mirror of
https://github.com/dtzp555-max/ocp.git
synced 2026-07-21 21:15:09 +00:00
1b324968f431569bcf35504fcf9b11b1e6222f38
33
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 
|
1b324968f4 |
feat(tui): real SSE streaming via claude's MessageDisplay hook (OCP_TUI_STREAM, default off) (#159)
* feat(tui): real SSE streaming via claude's MessageDisplay hook (OCP_TUI_STREAM, default off) Backlog #2. TUI-mode `stream:true` turns can now emit real SSE `delta.content` chunks as `claude` generates them, instead of buffering the turn and replaying it with streamStringAsSSE. Opt-in: with OCP_TUI_STREAM unset/0 the spawn argv, the SSE bytes and the cache behaviour are byte-for-byte unchanged (asserted by test). This PR does NOT mirror any cli.js function, so no `cli.js:NNNN` citation applies, and per CLAUDE.md's hard requirement #1 that is stated explicitly here rather than left implicit: - We consume claude's OWN `MessageDisplay` hook surface AS EMITTED — forwarding, not inventing. No new endpoint, no fabricated protocol, no new field. - The TUI spawn is OCP-owned surface: ADR 0007 owns it, not cli.js. - The SSE wire shapes are the OpenAI chat/completions streaming spec, adopted by ADR 0006. Every frame emitted here (role chunk, content-delta chunk, stop chunk, `[DONE]`, and the post-header {error:{message,type}} frame) is COPIED from callClaudeStreaming, the -p path. /health gains additive fields only (streamEnabled + 4 counters) — same grandfathered B.2 rationale as the existing tui block (ADR 0006). Existing keys are untouched. `claude` fires MessageDisplay per rendered block, handing the hook the RAW MARKDOWN SOURCE of an incremental delta on stdin. The hook is registered with `--settings` on the ordinary interactive spawn (no -p, no --bare) — verified to leave the billing pool alone. Sink: a static sh hook script appends each payload to `<streamDir>/<session_id>.jsonl`; OCP polls that file and forwards deltas as SSE. The per-session-id keying is MANDATORY, not an optimization — OCP_TUI_MAX_CONCURRENT defaults to 2, so two claude panes already run at once and a shared sink would splice one client's deltas into another's stream. Warm-pool compatible (a separate in-flight PR depends on this): the hook script AND the settings file are static — nothing request-specific is baked in at spawn time. The sink path reaches the pane through its own env (OCP_TUI_STREAM_FILE) and derives from the session-id, which a pre-booted pane fixes at boot. The hook is SYNCHRONOUS (forceSyncExecution: claude blocks on it), so the script writes and exits: one `cat` append, nothing else. Measured p50 7.2ms / p90 14.7ms per fire, ~50ms across a whole turn — noise against a 6-10s turn. It remains the terminal-turn signal, the source of the returned/cached text T, and the input to the honesty gates. The delta stream is a low-latency MIRROR, never a replacement: - the truncation gate (C-2) and auth-banner gate (C-1, issue #133) run BEFORE anything is committed or flushed, unchanged; - at end of turn the streamed bytes are asserted against T. Equal -> serve. A strict PREFIX of T -> top up from the transcript so the client still receives exactly T (counted). NOT a prefix -> REFUSE the turn: SSE error frame, no cache, no success, streamDivergences++. Serving text the transcript disagrees with is the failure class ALIGNMENT.md exists to prevent, so this fails loud rather than degrading quietly; - only T is ever cached — never the concatenated deltas. The auth banner needs prevention, not just detection (SSE deltas cannot be un-sent), so the first OCP_TUI_STREAM_HOLDBACK (100) chars are withheld: the default banner detector cannot match a message longer than 100 chars, so releasing past that provably cannot leak a banner. A custom CLAUDE_TUI_ERROR_PATTERNS has no such bound — OCP warns at boot. - BANNER, before/after the spawn change: `Sonnet 4.6 with low effort · Claude Max` both, including on the pane the server itself spawns. Never `API Usage Billing`. Transcript entrypoint stays "cli". --settings is not a --bare-class flag. - --settings MERGES with <HOME>/.claude/settings.json rather than clobbering it (the user-level settings' `env` block still reached the hook), so the isolated-HOME settings story (permissions / additionalDirectories) survives. - EXACTNESS: 8/8 varied prompts (short, long, markdown, code fence, multilingual, JSON, table, unicode) byte-exact vs transcript T, streamed AND buffered. 0 top-ups, 0 divergences over 15 streamed turns. - TTFT: buffered delivers NOTHING until the turn ends (TTFB == total, 7.5-15.8s). Streamed sends headers at ~25ms (heartbeat covers the pre-first-delta silence) and first content mid-generation, e.g. markdown 7.9s first chunk / 12.8s total; long 9.7s / 17.4s. - CONCURRENCY DEMUX: two concurrent streamed turns (ALPHA/BRAVO), tui.inflight peaked at 2, each read its own session-keyed transcript, ZERO cross-contamination. - AUTH-BANNER GATE under streaming, both layers: a short banner-like turn reached the client as 0 content chunks + an SSE error frame (never emitted); a long one was streamed but still ended on an error frame, not finish_reason:"stop", and was not cached. - DISCONNECT mid-turn: pane torn down and semaphore slot released within 1s (info-logged, not booked as a model error). - THINKING: not leaked. Opus 4.8 + xhigh turns carry a thinking block with a signature but `thinking:""` (the reasoning text is not persisted in interactive mode), both MessageDisplay text-extraction sites in the 2.1.207 bundle filter type==="text", and no reasoning prose appeared in any delta; concat===T held exactly on the single-message turn. - npm test: 282 passed, 0 failed (was 267 on main; +15). The transcript keeps only the model's LAST assistant message. A turn where the model narrates before calling a tool therefore has two messages, and T is only the second. If the narration exceeds the holdback it has already been streamed and cannot be retracted -> the turn is REFUSED. Reproduced live: Opus narrated 475 chars before a Bash call. The assembler discards a prior message's text when nothing has been emitted yet (so short narration is handled correctly and stays exact), and raising OCP_TUI_STREAM_HOLDBACK above the narration length rescues the turn — verified on that exact transcript: holdback>=500 -> served, exact=true. Documented in README and ADR 0007; this is why streaming is opt-in and off by default. ADR 0007 line 59 ("no real token streaming — deliberate") is amended, not silently contradicted. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(tui): re-integrate streaming onto the warm-pane pool (#158) — install the hook at BOOT Rebasing backlog #2 (streaming) onto #158 (warm pane pool) is not a textual merge: #158 split the monolithic runTuiTurn into bootTuiPane + runTuiTurn, and streaming had patched the monolith. Re-integrating it in the OLD shape would have compiled, passed every existing test, and been WRONG. The bug that shape would have shipped: the sink was derived at TURN time from a streamDir argument. But a POOLED pane is pre-booted long before any request exists — so on a pool HIT runTuiTurn never cold-boots, no hook was ever registered on that pane, and the turn would silently serve BUFFERED. Every miss streams, every hit does not; no error, no failing test. The operator sees "streaming does nothing in production" and has nothing to grep for. Fix — install the hook where the pane is born: - bootTuiPane({ streamDir }) registers the MessageDisplay hook at spawn and returns the pane's own sink (pane.streamFile), keyed by the pane's own --session-id. The hook script and settings file are STATIC (one pair per streamDir); the only per-turn thing is the sink path, and it is fixed at boot. So nothing request-specific is baked into a spawn. - runTuiTurn reads pane.streamFile — never recomputes it — so a warm pane and a cold pane stream through byte-for-byte the same path. - server.mjs threads the same streamDir into the pool's bootPane closure, so pre-booted panes carry the hook too. TUI_STREAM/TUI_STREAM_DIR now declare before the pool needs them. Three regression guards added (test-features.mjs), and the third was MUTATION-TESTED: with the fix reverted to the turn-time shape it fails ("the pooled pane's deltas must reach the client"), with the fix in place it passes. A guard nobody has watched fail is not a guard. /health: pool + stream* fields are now a union — the shape assertion asserts CONTAINMENT of the seven grandfathered keys plus an exact added-set, so a future field that silently REPLACED an original key cannot pass. Class B (ADR 0007, OCP-owned TUI spawn) — cli.js does NOT perform this operation. npm test: 313 passed, 0 failed. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VqgWJcjxrjjL9L9SkpZyXR * fix(tui): close the streaming auth-banner leak + 6 further review findings (PR #159) Independent review (Iron Rule 10) found a HIGH bug by EXECUTING the code, not reading it. All seven findings fixed. F1 and F3 were merge-blocking. F1 (HIGH) — the auth-banner holdback was bypassed after the first release. TuiDeltaAssembler.released was set once and never reset at a message_id boundary, so the holdback + detectError predicate guarded only the FIRST message of a turn. In production's own configuration (OCP_TUI_FULL_TOOLS=1, where multi-message tool-using turns are the norm): the model narrates past the holdback before a tool call -> released; credentials expire mid-turn -> claude renders the 401 as ordinary assistant TEXT as a NEW message -> push() took the `if (this.released)` branch and handed the banner verbatim to the client. That is precisely the silent-error case the C-1 gate exists to prevent. Detection survived (the turn was still refused at finalize) but PREVENTION did not. Fix: once a message boundary follows an emit, the turn is already unrecoverable — finalize() will refuse it — so push() now emits NOTHING further for the rest of the turn. Second hole in the same predicate: detectTuiUpstreamError() trims before applying its <=100-char rule, so 101 whitespace chars trimmed to "" -> detector had nothing to classify -> returned null -> release fired having screened nothing. Release now gates on the TRIMMED length, so both sides of the check talk about the same string. F2 — the "provably safe" claim in stream.mjs, ADR 0007 and README was unsound as written. Restated with both required halves: (i) nothing is emitted until the trimmed accumulation exceeds the detector's max banner length, AND (ii) no emission at all once a message boundary follows an emit. Half (i) alone only ever covered a turn's first message. F3 (blocker) — prepareStreamHook was write-if-missing, so md-hook.sh could never be updated OR repaired: a host that booted once under an older version was stuck on that HOOK_SCRIPT forever, and a non-atomic write interrupted mid-flight left a TRUNCATED script that existsSync() called fine — on a hook claude BLOCKS on synchronously. Now written unconditionally via tmp+renameSync (the pattern already used by ensureTuiCwdTrusted). F4 — the two spawn paths differed for non-streaming requests: the pool installed the hook whenever OCP_TUI_STREAM was on (correct — a pre-booted pane cannot know what request it will serve), but the cold path gated it on this turn's onDelta. So one stream:false request got --settings on a pool HIT and not on a MISS: two spawn argvs for the identical request, on this project's billing-classification surface. Both paths now gate on TUI_STREAM alone; whether the sink is POLLED remains correctly gated on onDelta. F5 — pool._drop() killed the pane but orphaned its sink file; the reap tick drains the whole pool, so sinks accumulated with no GC path. Now removed best-effort on every drop path. F6 — /health counters did not measure what they documented: streamTurns was incremented only AFTER the honesty gates, hiding exactly the turns an operator most wants to see (and making streamDivergences/streamTurns a meaningless ratio); streamDeltas counted every fire while claiming to count forwarded ones. Counters and docs now agree. F7 — total hook failure was silent: zero fires per turn still yields ok:true/exact:false and a normal, fully-buffered answer. Only streamTopUps moved, which the code itself calls benign. Added streamZeroDeltaTurns (+ a tui_stream_zero_deltas warning) to separate "the hook is dead" from "one fire was dropped". Tests: 316 passed, 0 failed (was 313). Every new guard MUTATION-TESTED — with each fix reverted the guard named for it fails, and passes with the fix restored: - drop the restartedAfterEmit guard -> 2 failed (incl. the strengthened old test) - revert trim() in the release gate -> 1 failed - revert F3 to write-if-missing -> 1 failed The pre-existing test "new message_id AFTER an emit" asserted finalize().ok === false but never checked what push() RETURNED — so it passed while F1 was live, documenting the leak instead of catching it. Strengthened to assert the emission, not just the verdict. Class B (ADR 0007, OCP-owned TUI spawn) — cli.js does NOT perform this operation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VqgWJcjxrjjL9L9SkpZyXR --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
|
9f5bc3264a |
feat(tui): warm pane pool — single-use pre-booted panes, opt-in via OCP_TUI_POOL_SIZE (−41%) (#158)
* feat(tui): warm pane pool — single-use pre-booted panes, opt-in via OCP_TUI_POOL_SIZE Backlog item #3 of docs/plans/2026-07-13-tui-latency/README.md. Every TUI request currently cold-boots a tmux+claude pane. This adds an OPT-IN pool of pre-booted panes. Recorded as ADR 0008 (docs/adr/0008-tui-warm-pane-pool.md), which extends ADR 0007. MEASURED (this host, Sonnet 4.6, --effort low, through a real OCP instance; a sample counts only if HTTP 200 AND the body carries the demanded marker): pool off (main code) n= 6 p50 10.17s [9164 9499 9760 10572 10774 11281] pool on, warm hits n=12 p50 6.00s [5286 5289 5520 5584 5621 5969 6040 6098 6280 7846 8036 11053] pool on, warm hits (post- n= 6 p50 5.62s [4729 4753 5236 6004 7548 9548] review-fix re-run) -> -4.17s / -41%. 12 hits / 1 miss / 0 bootFailures over 13 requests (and 6/1/0 on the post-fix re-run). Robust to counting the miss: n=13 p50 -> -40.6%. The plan doc predicted only -1.0s (the boot). It is ~4.2s because the cold path also pays ~2.9s INSIDE the first turn beyond claude's own reported turn_duration — post- input-bar init that an idle pane has already finished. Phase decomposition of the cold path (n=6 medians): prep 2ms | tmux spawn 27ms | boot->input-ready 1232ms | paste 8ms | paste-verify 426ms | submit->terminal 8458ms | teardown 8ms = 10162ms total, vs native turn_duration 5539ms => 4490ms of OCP-side overhead, of which the pool recovers ~1.26s of boot and ~2.9s of in-claude cold start. (The 426ms paste-verify is one 400ms poll tick; a real paste lands in ~80ms. Not addressed here — separate item.) DESIGN - SINGLE-USE panes. A pooled pane serves exactly ONE turn, then is killed and replaced in the background. Each carries its OWN fresh --session-id fixed at boot, so one session still holds one exchange. This is what keeps transcript.mjs's extractLatestAssistantText correct; its warning about a future warm pool reusing a session is answered in-place (comment updated) and left standing for anyone who later wants a second turn on a pane — that would be a cross-request TEXT LEAK and needs user-line scoping in the transcript reader first. - Pool keyed by model; --model is fixed at spawn. A miss falls back to the cold path with zero behaviour change. The pool warms the most recently requested model, so the first request after start (and after a model switch) is always a cold miss. - REAPER COEXISTENCE (the crux). An idle warm pane IS ours, and the periodic sweep runs precisely when we are idle. reapStaleTuiSessions() takes a `spare` set of EXACT live session names, and server.mjs DRAINS the pool immediately before the sweep: 1. a live pooled pane is never reaped — INCLUDING one still BOOTING (see below); 2. an orphaned pooled pane IS still reaped — membership is by exact name from a live in-memory registry, never by name shape, so a pane from a dead process generation has nothing claiming it. Omitting `spare` reaps MORE, never less (fail-safe); 3. kill-server is suppressed while any pane is spared — hence the drain, so the sweep still flushes <defunct> claude zombies (the only mechanism that can). - THE POOL TRACKS ITS IN-FLIGHT BOOT BY NAME, NOT AS A COUNT. bootTuiPane creates the tmux session SYNCHRONOUSLY and only then waits up to POOL_BOOT_MS (20s) for the input bar, so a pooled session can be LIVE for ~20s before its boot resolves. Tracking boots as a count meant the pool could not name that session, which caused two real bugs (found in review, reproduced, fixed, and now regression-tested): * the reap sweep KILLED the booting pane (it could not be spared), left the pool empty with nothing scheduled, and logged the exact tui_pool_boot_failed WARN operators are told to alert on — for a completely healthy drain; * graceful shutdown ORPHANED a live authenticated idle `claude`: gracefulShutdown calls process.exit(0) in the SAME TICK as the drain (TUI panes are tmux children, so activeProcesses is empty and the wait-for-children path exits immediately), so cleanup deferred to a .then() never ran. Fix: the pool mints each pane's identity up front ({sessionId, name}) and holds it in _bootingPane. liveNames() includes it; drain() kills it SYNCHRONOUSLY. A generation counter distinguishes "cancelled by us" from "genuinely failed", so a drain never inflates bootFailures and resume() reliably starts a fresh boot. Deriving the name from the session-id also makes `tmux ls` correlate to the transcript file. - SLOT ACCOUNTING. Refill boots take NO TuiSemaphore slot (those bound real turns and would be starved); they cannot leak one either, since they never hold one. Refills are SERIALIZED, one boot at a time — live at size=2, two cold boots racing an in-flight turn overran the readiness cap and a refill was discarded. A genuinely failed boot does not re-kick the chain (backoff; a broken claude must not respawn forever). Background boots get a more generous readiness cap (POOL_BOOT_MS = 5x BOOT_MS): BOOT_MS is tight because a client is blocked on it, which is not true of a pre-boot. - BOUNDED COST. A warm pane is a LIVE idle claude process held whether or not a request arrives. Peak processes = pool size + OCP_TUI_MAX_CONCURRENT + 1 booting replacement. Size clamped to POOL_MAX_SIZE=4; garbage values disable rather than guess. Panes have a 10-min TTL and a health check at hand-out (dead/degraded pane => miss, never a hang). Missing collaborators throw at CONSTRUCTION, not on a live request (refill() is called synchronously from the request path). DEFAULT OFF (OCP_TUI_POOL_SIZE=0). This is a stable production path and the pool holds standing processes, so the operator opts in. With the pool off, runTuiTurn takes the IDENTICAL code path as before (the `pool ? pool.acquire() : null` branch yields null, and tuiPool is null so no observer is attached and no new log line is emitted) — that is what establishes the default path is unchanged. A pool-off control run (n=6, p50 9.40s) is consistent with the 10.17s baseline but had 2/6 samples >12s, so it is corroboration, NOT proof: n=6 cannot establish "unregressed" on its own. The code-path equivalence can. BANNER: NO SPAWN ARGUMENT CHANGED. buildTuiCmd is byte-identical to main (verified by extracting the function body from both revisions and comparing). Live banner captured from two real POOLED panes anyway: "Sonnet 4.6 with low effort · Claude Max" — the subscription pool, never "API Usage Billing". /health: `tui.pool` added (null when off), incl. `cancelled` (boots WE killed — not a fault; do not alert on it). The tui block is ADR-0007-owned and post-dates ADR 0006's v3.16.4 grandfather snapshot; the addition is purely additive — every pre-existing key keeps a byte-identical value. Authorization recorded in ADR 0008. ALIGNMENT: Class B / ADR 0007 + ADR 0008 (OCP-owned TUI spawn machinery). cli.js does NOT perform this operation — there is no cli.js citation and none is required: this is not an Anthropic API surface, it is OCP's own process management around the claude CLI, exactly as the existing tmux session lifecycle and reaper already are (ALIGNMENT.md Rule 2). TESTS: 294 passed / 0 failed (was 267). +27 covering acquire/hit/miss, single-use (a pane is never handed out twice), bounded + serialized refill, TTL + health-check drops, model retarget, drain/resume, boot-failure backoff, identity linkage, all three reaper invariants incl. post-drain kill-server restoration, and — the coverage gap that let both bugs ship — FIVE mid-boot tests: the booting pane is nameable/spareable, the sweep's drain kills it and resume starts a fresh boot with no bogus WARN, shutdown kills it synchronously (asserted WITHOUT awaiting, since process.exit runs in the same tick), a stale settle cannot clear a newer boot's slot, and a model switch cancels an in-flight boot for the old model. Live verification (temporary 20s reap interval, reverted): sweep drained both panes -> reaped -> refilled with NEW panes; a foreign tmux session survived untouched; with no foreign session kill-server fired and the pool still recovered and served the next request. Both review bugs reproduced against a PRIVATE tmux server (-L pr3repro, so the reaper's internal kill-server could not touch the host) before and after the fix. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(tui): kill a cancelled boot's pane when it settles + make async tests actually count Folds in the independent review's remaining nit — and, in proving the nit's fix, uncovers two defects in the test suite itself. ## The nit (latent M1b, second costume) `_cancelBooting` kills BY NAME, but the tmux session only EXISTS once `bootPane` has run — and `bootPane` is queued on a microtask. So a caller doing `refill()` then `drain()` in the SAME synchronous block leaves `_cancelBooting` with nothing to kill (a no-op); it bumps the generation, and the boot microtask then CREATES the session, succeeds, and — under the old bare `return` on a stale generation — walked away from a LIVE authenticated `claude` that nothing owns. Reproduced: reverted: drain() kills nothing (no session yet) -> boot creates it -> ORPHAN: ['p1'] fixed : drain() kills nothing (no session yet) -> boot creates it -> boot kills it -> [] Not reachable from any current call site, so this is defense-in-depth — but ADR 0008 and the reap-tick comment in server.mjs BOTH explicitly contemplate a boot-time pre-warm, which is exactly the shape that reaches it. Killing an already-dead session is a harmless no-op, so the fix is idempotent whichever way the race lands. ## Defect 1 in the suite: async tests were never awaited (44 of them) Writing the regression guard exposed this. `test()` called `fn()`, got a promise back, and IMMEDIATELY printed ✓ and incremented `passed` — without awaiting it. For all 44 tests written as `test("...", async () => {...})`: - ✓ meant "did not throw SYNCHRONOUSLY", not "passed"; - a failed assertion escaped as an unhandled rejection, crashing the process (CI stays red on the non-zero exit) but never being COUNTED — so the summary could print "0 failed" and be wrong. The suite's headline number was therefore not evidence for ANY async test, including this PR's own M1a/M1b guards. `test()` now settles an async body before counting it, and the summary awaits them. ## Defect 2, exposed the instant defect 1 was fixed: a false guard `"a boot that resolves AFTER a drain kills its own pane ... no orphan process left behind"` asserted `killed.length === 1` — i.e. that kill was CALLED once. But `_cancelBooting`'s kill-by-name on a not-yet-existent session is a NO-OP that still increments that counter. So "kill was called once" and "a live session is orphaned" were both true at the same time: a test named for the absence of an orphan was passing while the orphan was present. Now asserts LIVENESS (`live.size === 0`) — the only honest question. ## Evidence fix present : 295 passed, 0 failed, exit 0 fix reverted: 293 passed, 2 failed <- BOTH liveness guards fire (the old kill-count guard did not) Also: `dropped`'s doc comment now lists `cancelled` (a cancelled in-flight boot lands there via _drop). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VqgWJcjxrjjL9L9SkpZyXR --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
|
5258d5d395 |
feat(tui): pin spawn effort via OCP_TUI_EFFORT (default low) (#156)
* feat(tui): pin spawn effort via OCP_TUI_EFFORT (default low) buildTuiCmd never passed --effort, so the pane's claude inherited a HOME-dependent effortLevel: real-home mode inherits the operator's ~/.claude/settings.json (high/xhigh on typical operator hosts), env-token scratch mode inherits claude's built-in default — proxied-turn latency silently depended on which HOME mode resolveTuiHome() picked and on an unrelated operator setting. Pass --effort explicitly, from new env var OCP_TUI_EFFORT (default "low"; allowlist low|medium|high|xhigh|max per `claude --help` 2.1.207; "inherit" restores the pre-flag argv byte-for-byte; an invalid value warns and falls back to "low" so a typo can never reach the pane argv). Not endpoint-touching: no server.mjs change, no wire-level change — the flag rides the existing interactive spawn (ADR 0007). Billing-pool safety verified per the docs/plans/2026-07-13-tui-latency banner protocol: startup banner stays "Claude Max" with "low effort". Measured through a test OCP instance (:3979, TUI mode, real-home, claude-sonnet-4-6, n=5+5, same ~1850-token prompt as floor.sh): before: median 11.30s range 9.05-12.38s (spread 3.32s) banner: high effort - Claude Max after: median 9.55s range 9.27-9.77s (spread 0.50s) banner: low effort - Claude Max OCP_TUI_EFFORT=inherit: banner back to "high effort" (pre-flag behavior restored) README: new row in the Environment Variables table (release_kit new_feature_doc_expectations: new env var -> README table). Tests: 4 new buildTuiCmd cases (default, explicit level, inherit, invalid fallback); suite 267 passed / 0 failed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VqgWJcjxrjjL9L9SkpZyXR * docs(readme): review nit — 'pre-v3.22' → 'pre-flag' (next version not fixed yet) Reviewer nit from the Iron Rule 10 independent review of PR #156. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VqgWJcjxrjjL9L9SkpZyXR --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
|
d96da46fa0 |
fix: honor runtime-lowered concurrency limit + cancel queued waiters on disconnect (#149)
* fix: honor runtime-lowered concurrency limit + cancel queued waiters on disconnect Fixes three findings from an independent concurrency audit of the -p/stream-json wait-queue (lib/tui/semaphore.mjs, reused by server.mjs as `claudeSemaphore`) and its acquireClaudeSlot()/callClaudeTui() callers in server.mjs: F1 (MEDIUM) — release() handed a freed slot straight to the next queued waiter without re-checking `this.limit`, so a PATCH /settings maxConcurrent decrease was silently ignored until every already-inflight task happened to finish on its own. release() now only re-grants when post-decrement inflight is still under the current limit, and a new setLimit() wakes queued waiters immediately when the limit is raised instead of only on the next incidental release(). F2 (MEDIUM) — a request queued behind the concurrency limit had no link to its HTTP connection, so a client that disconnected while still queued would still get a claude process spawned for it once a slot freed — burning subscription quota for a dead socket. acquire() now accepts an optional AbortSignal; server.mjs derives one from the client's res "close" event (closeSignalFor) and passes it into claudeSemaphore.acquire() / tuiSemaphore.acquire() while queued. On abort the waiter is spliced out of the queue (not just flagged), so `queued` accounting stays exact; the same "close" signal is wired into acquireClaudeSlot() (-p path, non-streaming + streaming + singleflight-wrapped) and callClaudeTui() (TUI path). If the response is already destroyed by the time we try to queue, we reject immediately without ever entering the queue. F8 (cosmetic) — acquireClaudeSlot() set `stats.queued = claudeSemaphore.queued + 1` BEFORE calling acquire(), over-reporting /health's queued count by 1 whenever a slot was granted immediately (the common, non-queued case). acquire() already updates its internal queue synchronously before returning a Promise, so reading claudeSemaphore.queued right AFTER calling it (instead of guessing "+1" before) is exact. No /health field was added, removed, or renamed. ALIGNMENT.md: this PR touches request-handler code (callClaude, callClaudeStreaming, callClaudeTui, acquireClaudeSlot) but is local concurrency-control/queue-accounting infrastructure with no cli.js wire analogue — it does not add, rename, or change any endpoint, header, request field, or response field, and does not touch the /v1/messages forwarding path or the OAuth bearer machinery (the two Class A surfaces this repo governs). The /health response shape is unchanged (same field set, same nesting; only the *value* of the pre-existing `stats.queued` field is corrected). Per CLAUDE.md hard-requirement #1, a cli.js citation is therefore declared ABSENT: there is no corresponding cli.js operation to cite because this is not a cli.js-mirror (Class A) change and not a Class B endpoint-contract change either. Tests: added 6 unit tests to test-features.mjs against the shared TuiSemaphore (lowering the limit mid-load does not over-admit; raising the limit wakes queued waiters up to the new headroom, FIFO; a queued waiter cancelled via AbortSignal is spliced out and never later acquires; an already-aborted signal never touches the queue; cancelling one of several queued waiters preserves FIFO for the rest). 238 pre-existing tests remain green; suite is now 244/244. Verification: `node --check server.mjs && node --check lib/tui/semaphore.mjs && npm test` — 244 passed, 0 failed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix: singleflight follower retry on leader disconnect + quiet disconnect handling (review M1/L1/L2) Addresses the independent reviewer's APPROVE-WITH-CHANGES findings on PR #149: M1 (MEDIUM, F2 regression) — when a singleflight LEADER disconnected while queued, its RequestDisconnectedError rejected the SHARED promise, so live followers fell into respondUpstreamError's generic branch and got a spurious 500 on a healthy socket. Fix: keys.mjs singleflight() gains an optional follower-side `retryIf` predicate. When a follower joins an existing flight and the shared promise rejects with an error retryIf() accepts, the follower does NOT inherit the rejection — it re-enters singleflight with its OWN fn (the map entry is guaranteed already deleted: the delete-finally is attached upstream of the promise followers await), becoming the new leader or joining a retrying sibling's fresh flight. The leader's own rejection is never retried (it IS that client's disconnect). server.mjs passes retryIf = (err) => err instanceof RequestDisconnectedError && !res.destroyed, so a follower whose own client is also gone still propagates quietly. Callers without retryIf keep byte-for-byte pre-existing share-everything semantics (pinned by the existing failure-fan-out test). L1 (LOW) — a disconnect-while-queued on the non-streaming paths was recorded as a usage FAILURE row and logged as a [proxy] error: metric noise for a non-error. Both non-streaming catch blocks now early-return on RequestDisconnectedError without recordUsage(success:false) and without console.error — mirroring the streaming path, which returns silently. The disconnect remains observable at info level (concurrency_wait_cancelled, now also emitted with path:"tui" from callClaudeTui for parity with acquireClaudeSlot's -p log). L2 (LOW, test gap) — added a unit test for the abort-after-grant race: a waiter granted its slot whose signal aborts afterward must see no rejection, no queue corruption, and its slot released exactly once via the normal path (the semaphore detaches the abort listener at grant; the onAbort idx===-1 guard is the in-dispatch backstop). Tests: +3 (2× M1 in the singleflight section, 1× L2 in the F2 section) — suite is now 247/247 green. ALIGNMENT.md: unchanged declaration — still local concurrency/dedup infrastructure with no cli.js wire analogue; no endpoint, header, request field, or response field added or changed; /health shape untouched. cli.js citation declared ABSENT per CLAUDE.md hard-requirement #1 (not a Class A mirror change, not a Class B contract change). Verification: node --check server.mjs lib/tui/semaphore.mjs keys.mjs && npm test — 247 passed, 0 failed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
|
2538233059 |
fix(server): serialize -p real-HOME token fallback + TTL-cache keychain + de-stale isolation decision (F3/F5/F6) (#150)
* fix(server): serialize -p real-HOME token fallback + TTL-cache keychain + de-stale isolation decision (F3/F5/F6) Three audit findings in the -p spawn-token resolution + HOME-isolation layer. All are infra/ process changes to how OCP READS and GATES an OAuth token it already holds; none touch the OAuth wire machinery. F3 (MEDIUM) — expiry-window fallback herds concurrent -p spawns into real HOME. When the keychain token is within 5 min of expiry, resolveSpawnToken() returns null and every concurrent spawn simultaneously falls back to the real HOME; each spawned claude then races a refresh_token grant against the SAME single-use refresh token — rotating it out from under the others and the operator's real claude (the credential-fork hazard, #112/#146 class). Fix: a promise-chain mutex (createSerialMutex) serializes ONLY the real-HOME fallback — one such spawn at a time. When a serialized waiter is admitted (prior holder torn down → its claude has refreshed the keychain), it re-runs resolveSpawnToken(): a now-fresh token means it proceeds ISOLATED instead of real-HOME, so the queue drains to the fast path. Isolated spawns never touch the mutex. F5 (LOW-MED) — per-spawn double keychain exec on the hot path. getOAuthCredentials() sync-exec'd `security find-generic-password` up to twice (wrong label first), worst case 5s×2, blocking the event loop and stalling in-flight SSE streams. Fix: (a) memoize the last-good keychain label and try it first (orderLabelsLastGoodFirst); (b) a 30s TTL cache of the read (createTtlCache). This does NOT reintroduce the #146 forever-memoized regression: the TTL bounds only how often we re-READ the keychain; resolveSpawnToken() still applies the 5-min expiry gate (isTokenExpiring) to the CACHED creds on EVERY use, so a token expiring within the window is still rejected → real-HOME fallback. Call sites stay synchronous (no async conversion). F6 (LOW) — memoized isolation decision goes stale; /health could misreport. getSpawnHomeMode() memoized the isolated/real-home decision forever: credentials appearing after startup never enabled isolation; deleting ~/.ocp/spawn-home at runtime ENOENT'd every isolated spawn until restart; during an expiry stint /health reported isolated:true while spawns ran real- HOME. Fix: re-evaluate the decision per spawn (cheap now that F5 caches the keychain read); ensureSpawnHome() re-verifies + re-prepares the scratch dir per isolated spawn; and /health now reports the EFFECTIVE decision (token presence AND expiry gate). The /health field set is UNCHANGED — no field added/removed/renamed — only the values are made truthful. Alignment: - Class: Not a wire/endpoint change for the spawn-token layer + Class B (B.2) for /health. - cli.js citation: DECLARED ABSENT. cli.js does NOT perform OCP's spawn-token resolution, HOME isolation, keychain caching, or fallback serialization — these are proxy-internal process concerns with no cli.js analogue, so no Class A cli.js:NNNN citation exists or is required (ALIGNMENT.md Rule 2: this is proxy infra, not an invented forwarded endpoint/header/body). - The /health change is authorized by ADR 0006 (grandfathered B.2 as of v3.16.4) and is a behaviour-preserving contract change: same fields, truthful values. - OCP still NEVER performs a refresh_token grant itself — that property is preserved; the fix only serializes/gates reads of a token refreshed by the spawned or real claude (#112). - No new endpoint, header, or request/response field. alignment.yml blacklist unaffected. Tests: extracted the pure primitives to lib/spawn-auth.mjs and added 11 unit tests (mutex serialization order + idempotent release; TTL cache freshness + null-miss; expiry gate; label ordering; and the combined invariant that the TTL cache respects the expiry gate). node --check clean; 249 tests pass (238 + 11). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(server): drain F3 fallback queue immediately — invalidate F5 keychain cache before the serialized re-check Follow-up to the F3/F5/F6 fix (independent-review observation). F5's 30s keychain TTL cache could make F3's post-refresh re-check see the stale (expiring) cached creds for up to ~30s, so a waiter admitted right after the prior real-HOME holder's claude refreshed the keychain would needlessly fall back to real HOME again instead of proceeding ISOLATED. Serialization safety was never at risk (still one real-HOME spawn at a time, no double-refresh); only the drain-to-fast-path optimization lagged. Fix: invalidateKeychainReadCache() clears the F5 TTL cache; resolveSpawnDecision() calls it under the fallback mutex, immediately before the re-check, so the admitted waiter reads FRESH keychain state and drains to the isolated fast path at once. The extra keychain read happens only on the rare real-HOME fallback path and only under the mutex (serialized, one at a time). Alignment: unchanged from the parent commit — proxy-internal keychain/HOME-isolation process logic, no cli.js analogue (cli.js citation DECLARED ABSENT, ALIGNMENT.md Rule 2), no endpoint/header/body, /health shape unchanged, OCP still never performs a refresh_token grant itself (#112). 249 tests pass. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
|
31e5a44099 |
fix(tui): scope session prefix + reap/kill-server to this instance's port (F7) (#148)
Audit finding F7 (LOW): lib/tui/session.mjs hardcoded SESSION_PREFIX =
"ocp-tui-" as a bare, host-wide constant. The boot-reap and periodic
idle-reap in server.mjs used it to decide which tmux sessions to
kill-session and whether to kill-server (which flushes defunct <claude>
zombies but tears down the WHOLE tmux server, including any live pane).
The coexistence guard only ever spared foreign product prefixes
(olp-tui-*); a SECOND OCP instance on the same host — e.g. a temporary
verification instance stood up alongside production, a real pattern
used during PR #144/#146 verification — was indistinguishable from
"ours" and could have its LIVE sessions reaped/kill-server'd by the
other instance's boot or periodic sweep.
Fix: scope the session-name prefix to this instance's own listen port
(the natural stable per-instance discriminator on one host — two OCP
instances cannot share a port): `ocp-tui-<port>-`. A sibling instance's
`ocp-tui-<otherPort>-*` sessions now fail the own-prefix startsWith
check and fall into the same "othersRemain" bucket as olp-tui-*,
so they are never touched and never used to justify kill-server.
lib/tui/session.mjs:
- sessionPrefixForPort(port) replaces the bare SESSION_PREFIX export.
- reapStaleTuiSessions({ tmux, port, includeLegacy }) now requires
port and computes its own prefix from it.
- runTuiTurn({ ..., port }) builds the tmux session name from
sessionPrefixForPort(port) instead of the old bare constant.
- LEGACY_SESSION_PREFIX / LEGACY_SESSION_NAME_RE (exact
"ocp-tui-<8-hex>" shape, no port segment) describe the OLD
pre-fix session-name shape, retained only for the migration below.
Legacy migration rule (chosen + reasoning): a bare-prefix legacy
session cannot be created by any post-fix OCP process, so if one is
seen it is presumed to be an orphaned zombie from THIS instance's own
PRE-fix process generation (left behind across an in-place upgrade),
not a stranger's. reapStaleTuiSessions() therefore accepts an
includeLegacy flag: server.mjs's one-time BOOT reap passes
includeLegacy: true (claims exact-legacy-shape sessions as its own,
enabling cleanup right after an upgrade); the periodic 15-min idle
sweep does NOT set it, so a lingering legacy-shaped session during
steady-state is conservatively treated as foreign and cannot trigger
kill-server on a routine tick. Residual (documented, accepted): a
genuinely-still-running PRE-fix OCP instance coexisting on the host at
the exact moment a new instance boots could have its live legacy
session reaped — the same class of residual risk the audit finding
itself accepts ("no live instance of the new version creates them");
this PR does not regress that scenario, it only removes the far more
common same-version collision that is the actual F7 finding.
LEGACY_SESSION_NAME_RE (`^ocp-tui-[0-9a-f]{8}$`) can never match the
new shape: the new shape always inserts a literal "-" between the
port digits and the 8-hex suffix, which the anchored 8-hex-only legacy
regex cannot satisfy.
server.mjs changes are local TUI session-lifecycle infrastructure
(tmux session naming, boot/periodic reap, kill-server) with no cli.js
wire analogue — verified via `strings` against the compiled claude
CLI 2.1.198 binary (this machine ships cli.js as a Mach-O binary per
ALIGNMENT.md's "OAuth token-host verification" precedent): cli.js
contains only `env.TMUX` detection (whether IT is running inside a
tmux pane) and an unrelated `--remote-control-session-name-prefix`
flag for its own remote-control feature — no session-prefix/reap/
kill-server mechanism of any kind. Per ALIGNMENT.md Rule 2, this is
declared absent: no endpoint, header, request, or response shape
changed; only OCP's own local process-lifecycle bookkeeping. No PORT
literal was hardcoded (CI port-SPOT check) — PORT is threaded through
from the existing server.mjs SPOT (lib/constants.mjs DEFAULT_PORT via
CLAUDE_PROXY_PORT).
test-features.mjs: rewrote the reaper suite's fixture session names to
the new port-scoped shape, added tests for sessionPrefixForPort(),
LEGACY_SESSION_NAME_RE's non-collision with the new shape, a sibling
same-host OCP instance being treated as foreign (F7 regression test),
and the includeLegacy boot-migration behavior (claims legacy zombies,
still spares a sibling instance's port-scoped session).
Verified: node --check server.mjs && node --check lib/tui/session.mjs
&& npm test → 243 passed, 0 failed.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
||
|
|
38da104b97 |
chore(release): v3.21.0 — TUI cleanup + client-tools/ToS docs + promotion plan (#145)
* refactor(tui): dead-code / footgun cleanup (A1/A2/A3)
ALIGNMENT.md Rule 2: infra-only, no new cli.js wire behavior.
No protocol, no endpoint, no credential changes.
A1 (session.mjs): delete resolveTuiEntrypointEnv() and its call site
+ the redundant env-strip block around the spawnSync call. The function
mutated a {env} object passed to spawnSync (tmux itself), but tmux does
NOT forward that env to the pane; the pane's claude gets its env ONLY
from the `env` prefix string built inside buildTuiCmd (verified live
2026-06-01). The spawnSync {env} is intentionally minimal now; only
env.HOME is retained (tmux binary reads it). All claude-specific vars
go via the buildTuiCmd prefix string, unchanged. Tests for the now-
deleted function removed; test count drops by 7 (expected).
A2 (transcript.mjs): delete encodeCwd() and transcriptPath() exports.
Production resolves transcripts exclusively via findTranscriptPath()
(glob by session-id); these two helpers carried a fragile path-encoding
rule used only by their own tests. grep confirms zero non-test importers.
Added a TODO comment near findTranscriptPath() noting a CI fixture-
contract test would make claude-schema drift fail loudly. Tests removed;
count drops by 2.
A3 (session.mjs + README): remove the CLAUDE_SKIP_PERMISSIONS branch
that pushed --dangerously-skip-permissions when OCP_TUI_FULL_TOOLS=1.
OCP_TUI_FULL_TOOLS=1 now always takes the --allowedTools path. Rationale:
claude v2.1.x shows an interactive bypass-acceptance screen that a
headless tmux TUI cannot answer — bricks the turn (tui_paste_not_landed
/ wallclock cap), not recoverable without a human. The working path is
--allowedTools + scratch-home settings.json additionalDirectories.
README OCP_TUI_FULL_TOOLS row updated to document the removal and the
correct alternative; CLAUDE_SKIP_PERMISSIONS row for the -p path is
unchanged (still used in server.mjs). Test updated: skip-permissions
case replaced with an assertion that --dangerously-skip-permissions is
absent from the full-tools command.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* docs: client-tools boundary, ToS honesty, promotion plan (B1/B2/B3)
ALIGNMENT.md Rule 2: docs-only, no new cli.js wire behavior.
No protocol, endpoint, or credential changes.
B1 (README): add 'Client-tools boundary' subsection under 'How It
Works'. Documents that OCP is a text-prompt bridge only — it does not
pass OpenAI tools/functions or Anthropic tool_use blocks to the client.
Clients receive assistant TEXT only; client-local tool execution is
not supported by design (bypassing cli.js = out of scope per
ALIGNMENT.md).
B2 (README): two updates to 'Why OCP?' and the LAN-sharing section.
(a) New bullet: OCP drives the official claude CLI as-is — no OAuth
token extraction, no binary patching, no protocol invention — so
traffic looks like genuine Claude Code (cc_entrypoint=cli).
(b) LAN-sharing paragraph strengthened: pooling one Claude subscription
across multiple distinct people may violate Anthropic's Consumer ToS
and risk account suspension by the abuse classifier. The defensible
framing is 'one person, your own devices'; friends/team sharing is
not. Replaces the softer 'account terms are your call' language.
Feature and auth-mode docs are unchanged.
B3 (docs/PROMOTION.md): new promotion strategy doc. Covers: goal
(polish + low-key OSS visibility, NOT growth-hacking given the live
ToS/billing risk), pre-requisites (stability first), honest ToS
disclosure requirement, items explicitly skipped (multi-backend
routing, gateway model-discovery — delegated to OLP; raw API
passthrough — ALIGNMENT.md scope), TUI toggle as billing-split
insurance, and low-key visibility actions. Framed as a recommendation
for the maintainer to review, not a committed plan.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* chore(release): v3.21.0 — TUI cleanup + docs honesty + promotion plan
ALIGNMENT.md Rule 2: release prep; no new cli.js wire behavior.
Bump version 3.20.1 → 3.21.0. CHANGELOG entry covers:
- A1/A2/A3 TUI dead-code removals (inert entrypoint-env path,
test-only transcript helpers, headless-unusable skip-permissions)
- B1/B2/B3 docs (client-tools boundary, ToS honesty, Why-OCP posture,
promotion plan)
- Previously-shipped v3.20.x items documented for completeness:
spawn-home isolation, bounded concurrency queue + 429, ocp restart,
ocp-plugin OpenClaw compat.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
|
||
|
|
5aaab5ea28 |
fix: -p spawn-home isolation (③ latency 3x) + concurrency queue/429 (⑥) + ocp restart env + /ocp plugin compat (#144)
* fix(server): isolate default -p spawn in credential-free scratch HOME (latency ③) The default (-p/stream-json) spawn inherited the operator's real HOME (global ~/.claude plugins/skills/hooks) and ran with cwd=~/ocp (project CLAUDE.md/skills), loading heavy host context on EVERY request. Measured: pure API floor for haiku "hi" ≈ 1–2s; same CLI in the operator's real HOME/cwd ≈ 10–28s; a clean minimal HOME + CLAUDE_CODE_OAUTH_TOKEN ≈ 3–7s with auth intact. When an OAuth token is resolvable (and OCP_SPAWN_REAL_HOME!=1), spawnClaudeProcess now runs claude under a credential-free minimal scratch HOME (<HOME>/.ocp/spawn-home, no .credentials.json / settings.json / plugins) with cwd = that neutral dir and the resolved token in CLAUDE_CODE_OAUTH_TOKEN (env token is authoritative for -p). Mirrors the TUI path's resolveTuiHome() env-token mode. Falls back to real HOME + inherited cwd when no token resolves (zero regression); OCP_SPAWN_REAL_HOME=1 is a kill-switch. The token is resolved ONCE (memoized via getSpawnHomeMode, reusing getOAuthCredentials, the same resolver the /usage probe uses) and never logged. Adds a startup log line and an additive /health `spawn` block so the operator can confirm isolation is on. ALIGNMENT.md Rule 2 justification: this is an INFRA / PROCESS-MANAGEMENT change (HOME/cwd/env isolation of the spawned process). It does NOT mirror a cli.js wire operation, introduces no new endpoint/header, and adds no API token to the wire path — so there is no cli.js function to cite. CLAUDE_CODE_OAUTH_TOKEN / CLAUDE_CODE_DISABLE_* are existing claude-CLI env contracts already used by the TUI path and getOAuthCredentials. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(server): bounded wait-queue + HTTP 429 for -p concurrency overflow (⑥) spawnClaudeProcess used `if (activeRequests >= MAX_CONCURRENT) throw` → the client got an opaque 500 AND the rejection was uncounted (a 15-concurrent stress run returned 7×500 while /health stats.errors stayed 0). The TUI path already had a bounded-queue semaphore (TuiSemaphore); the -p path did not. Now the -p path reuses TuiSemaphore as `claudeSemaphore = new TuiSemaphore(MAX_CONCURRENT, { maxQueue: CLAUDE_MAX_QUEUE })`. Requests beyond MAX_CONCURRENT WAIT (up to CLAUDE_MAX_QUEUE, default 16) instead of being rejected; only when the queue is ALSO full does the request get HTTP 429 + Retry-After (rate_limit_error, NOT 500), a distinct `concurrency_queue_full` log, and a stats.queueRejections counter surfaced on /health. callClaude / callClaudeStreaming now acquire a slot (acquireClaudeSlot) before spawning; the release fn is wired into the existing idempotent cleanup() so the slot is freed on EVERY exit path (close/error/timeout/abort) — the #37/#40 slot-leak guard. MAX_CONCURRENT semantics (max concurrent claude procs) are unchanged; only overflow handling changed from throw-500 to queue-then-429. claudeSemaphore.limit is kept in sync with runtime /settings maxConcurrent changes. Live-verified (fake claude, MAX_CONCURRENT=1 MAX_QUEUE=1): 3 concurrent → exactly one 429 (Retry-After: 7, rate_limit_error body) + two 200s, on BOTH the non-streaming and streaming paths; /health stats.queueRejections=1 while stats.errors=0; after drain activeRequests=0 / inflight=0 / queued=0 and a follow-up request returns 200 (no slot leak). 2 new unit tests (247 passed, 0 failed). ALIGNMENT.md Rule 2 justification: this is an INFRA / PROCESS-MANAGEMENT change (a concurrency queue + backpressure status code in front of the existing spawn). It does NOT mirror a cli.js wire operation, adds no new endpoint or wire header, and introduces no API token — so there is no cli.js function to cite. (Retry-After is a standard HTTP response header on OCP's own 429, not a claude wire header.) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(ocp): macOS restart uses bootout+bootstrap so plist env is re-read The macOS restart path used `launchctl kickstart -k gui/$uid/dev.ocp.proxy`, which only re-execs the process and reuses launchd's CACHED environment — so a plist EnvironmentVariables edit (CLAUDE_BIND, CLAUDE_CODE_OAUTH_TOKEN, etc.) was silently ignored until a full unload/reload. This is the documented pit-index footgun. `ocp restart` (macOS) now does a full `launchctl bootout` + `bootstrap` of the agent via a new `_launchd_reload` helper, which re-reads the plist EnvironmentVariables so env changes take effect. Success is keyed on the bootstrap (the env-reloading load), not the bootout (which may legitimately fail if the agent is not currently loaded). A missing plist returns failure so the `elif` chain falls through to the legacy label and then to the Linux `systemctl --user restart` path unchanged (systemctl already re-reads its EnvironmentFile). Updates `ocp restart` help text and adds a README Troubleshooting subsection ("Env var change doesn't take effect after restart") with the manual bootout+bootstrap commands and a ps-based verification one-liner. Verified `_launchd_reload` with a stubbed launchctl: missing plist → rc=1 (falls through, no launchctl call); bootout-fail + bootstrap-ok → rc=0; bootstrap-fail → rc=1; call order is bootout then bootstrap. `bash -n ocp` clean; npm test 247/0 (server.mjs untouched). ALIGNMENT.md Rule 2 justification: this changes ONLY the `ocp` CLI wrapper's local service-restart mechanism (launchctl invocation). It does not touch server.mjs, the wire path, any endpoint/header, or any API token — so there is no cli.js function to cite. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(readme): document new env vars CLAUDE_MAX_QUEUE / CLAUDE_QUEUE_RETRY_AFTER / OCP_SPAWN_REAL_HOME Release_kit contract (CLAUDE.md Iron Rule 5.5: "new env var → README § Environment Variables table") requires the three env vars added by the perf/concurrency fixes to be in the README table. Adds rows for CLAUDE_MAX_QUEUE (16) + CLAUDE_QUEUE_RETRY_AFTER (5) (FIX ⑥ -p wait-queue + 429) and OCP_SPAWN_REAL_HOME (FIX ③ spawn-home isolation kill-switch), each cross-referencing the additive /health.concurrency and /health.spawn fields. Addresses the independent reviewer's MEDIUM finding. Docs only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(ocp-plugin): add openclaw.extensions for OpenClaw 2026.5.27 compat + sync version (②/ocp) ocp-plugin/package.json's openclaw object lacked the 'extensions' field that OpenClaw 2026.5.27 requires to install/load a plugin (matches the sibling olp plugin). Without it the daemon refused to load the local path plugin, breaking /ocp Telegram commands. Version synced 3.12.0 -> 3.16.2 to match the manifest. Plugin-layer change; no server.mjs / ALIGNMENT cli.js surface touched. --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
|
60930f0ba4 |
fix(tui): credential-isolated env-token auth (ends recurring 401) + reap defunct sessions (#141)
* fix(tui): pass CLAUDE_CODE_OAUTH_TOKEN to spawned claude + reap defunct sessions
Root cause (PI231 incident): tmux does not forward the parent's env to the
pane, so the TUI claude never saw CLAUDE_CODE_OAUTH_TOKEN and fell back to
~/.claude/.credentials.json, whose single-use refresh token got corrupted to an
empty string by the per-request spawn + kill-session teardown racing claude's
token rotation -> permanent "Please run /login" 401 (re-login re-corrupted on
the next spawn). Connected leak: the pane's claude is a child of the tmux server
(not node), so kill-session left <defunct> zombies the server never reaped (25
over 30 days; tmux kill-server dropped it 25->3).
Fix 1: buildTuiCmd now adds CLAUDE_CODE_OAUTH_TOKEN=<shq-escaped> to the pane
env prefix when the env is set, so claude authenticates via the long-lived token
and never touches the credentials.json refresh path (matching stable hosts).
Unset -> no token added (credentials.json-only hosts unaffected).
Fix 2: reapStaleTuiSessions kill-servers after clearing our own sessions ONLY
when no foreign tmux session remains (never disrupts a co-hosted olp-tui-*).
kill-server is the only node-reachable action that ACTUALLY reaps -- server exit
reparents survivors to init, which waitpids them; a per-session kill cannot,
since node is not the zombies' parent. Added a 15-min periodic reap (server.mjs)
gated on TUI_MODE and on the TUI path being idle. Residual: a request whose pane
is created in the idle-check/kill-server window fails cleanly via the existing
honesty gates (documented).
ALIGNMENT: Class B (OCP-owned TUI spawn). cli.js does NOT perform either
operation -- there is no cli.js analogue for "how the TUI pane authenticates" or
"reaping tmux-server-owned zombies"; authorized by ADR 0007 (PR-C amendment) per
ALIGNMENT.md's Class B citation requirement. No Class A wire surface, no endpoint
shape, no alignment.yml token, and no models.json entry touched.
Tests: +6 in test-features.mjs (buildTuiCmd token set/unset/shq-injection;
reaper kill-server ours-only / foreign-present / no-server). 241 passed, 0 failed.
Co-Authored-By: Claude <claude-opus> <noreply@anthropic.com>
* fix(tui): isolate TUI auth to env-token-only home (no credentials.json shadowing)
Passing CLAUDE_CODE_OAUTH_TOKEN to the spawned interactive `claude` (commit
|
||
|
|
3322d7bdae |
feat(tui): per-path concurrency limit + /health observability (PR-B) (#139)
Two P1 audit fixes for the TUI subscription-pool bridge (ADR 0007). The
default path (CLAUDE_TUI_MODE unset) is unchanged except the additive
/health `tui` block (enabled:false when off).
C-4 — TUI path had NO concurrency bound. The global MAX_CONCURRENT gate
lives in spawnClaudeProcess (the -p/stream-json path); callClaudeTui never
calls it — it calls runTuiTurn, which cold-boots a full interactive claude
in tmux. So N concurrent TUI requests spawned N simultaneous cold boots (a
family burst of 5 on a Pi 4 = OOM risk + subscription rate-limit pressure).
Adds an independent in-process limiter (lib/tui/semaphore.mjs, TuiSemaphore)
gating callClaudeTui: OCP_TUI_MAX_CONCURRENT (default 2 — a TUI turn is heavy:
per-request cold-boot + up to 120s wallclock). Queues rather than rejects
(mirrors MAX_CONCURRENT intent), with a bounded wait queue (default 32x the
limit) → tui_queue_full (503) on overflow rather than unbounded growth. The
slot releases in a finally, so PR-A's honesty-gate throws / timeouts / paste
failures never leak a slot.
C-5 — no operator-visible TUI drift surface. The tui_entrypoint_mismatch
warning only reached journald; after the 6/15 flip a silent sdk-cli drift
(the documented top risk) would drain metered credits invisibly. Adds an
additive `tui` block to /health: { enabled, entrypointMode, lastEntrypoint,
entrypointMismatches, inflight, queued, maxConcurrent }. lastEntrypoint /
entrypointMismatches are recorded in callClaudeTui (same mismatch branch the
journald warning covers); inflight/queued come from the C-4 semaphore.
ALIGNMENT (Class B): cli.js does NOT perform this operation — both the TUI
path and /health are OCP-owned, so no cli.js citation applies. /health is a
grandfathered B.2 endpoint (ADR 0006, frozen at v3.16.4). The response-shape
change is authorized by the ADR 0007 PR-B amendment added in this commit and
is behaviour-preserving: the `tui` block is NEW fields only — no existing
/health field is changed, renamed, removed, or re-typed, and no existing
semantics change, so existing consumers (dashboard, ocp-connect, monitoring)
are unaffected. Per ALIGNMENT.md's grandfather provision, an additive
behaviour-preserving change to a grandfathered B.2 endpoint is authorized by
an ADR; ADR 0007 is the authority for the TUI observability surface. No
Class A forwarding path, no alignment.yml, no models.json touched — alignment
blacklist is unaffected (zero new network tokens).
Tests: 11 new (lib/tui/semaphore.mjs is importable, so the semaphore + the
two pure /health helpers are tested directly): limit=1 serializes two
overlapping calls; limit=2 runs two + queues the third; slot released on
throw; bounded queue → tui_queue_full; mismatch counter increments on
cli→other drift; auto mode never counts a mismatch; /health tui block shape
+ live inflight/queued. npm test: 235 passed, 0 failed (was 224).
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude <claude-opus> <noreply@anthropic.com>
|
||
|
|
a37ff713d9 |
fix(tui): honest error/truncation handling + version-robust entrypoint + short-prompt paste (PR-A) (#137)
TUI-mode (CLAUDE_TUI_MODE=true) is an OCP-owned subscription-pool bridge for the
2026-06-15 Anthropic billing split. This PR fixes four honesty/robustness defects
confirmed by a prior audit and live-reproduced on PI231 (2026-06-10).
C-1 (P1) — upstream AUTH-FAILURE banner returned/cached as a real answer.
The interactive claude CLI renders in-session errors as ordinary assistant text.
The R-1 case C-1 exists to catch is expired/invalid credentials, where EVERY turn
comes back as the same one-line auth-failure banner, e.g. the two live PI231
banners (2026-06-10):
"Please run /login · API Error: 401 Invalid authentication credentials" (69 chars)
"Failed to authenticate. API Error: 401 Invalid authentication credentials" (73 chars)
callClaudeTui returned that banner verbatim → OCP cached it, shared it via
singleflight, and recorded a model SUCCESS. New detectTuiUpstreamError()
(lib/tui/transcript.mjs) flags such a turn; callClaudeTui throws tui_upstream_error
+ logEvent("error", …) BEFORE recordModelSuccess/cache write-back, so the error
never enters the cache; the client gets a 5xx.
C-1 NARROWING (false-positive probe, 2026-06-10). An earlier generalised default
rule — ^<short auth-failure prefix>?API Error: <3-digit> <detail>$ — was TOO BROAD:
the unbounded ".*" detail tail let any short prefix + "API Error: NNN" + an
arbitrarily long sentence match, so it KILLED legitimate long answers that merely
DISCUSS an API error (e.g. "API Error: 500 happened because the server was
overloaded. To fix this, retry with exponential backoff …"). A false-positive costs
the user a missing answer AND a double-burn retry — strictly worse than the rare
false-negative (caching one transient error for the 5-min TTL). C-1 is therefore
reframed from "detect any API error" to "detect a claude-CLI AUTHENTICATION-FAILURE
banner", and is CONSERVATIVE: when unsure it PASSES. The narrowed default detector
flags a turn only when ALL of the following hold over the WHOLE trimmed text
(a conjunction; any one failing => PASS):
1. SHORT whole-message — length ≤ 100 (live banners are 69/73; cap gives headroom
while rejecting multi-sentence prose; a 226-char auth answer is dropped on
length alone).
2. Contains "API Error: 4\d{2}" — auth failures are 4xx (401/403); transient 5xx
and bare "HTTP 401 means unauthorized." (no API-Error core) are excluded.
3. Contains an auth keyword — authenticat | /login | credential (case-insensitive);
rejects "To debug a 401 … API Error: 401 Unauthorized …" (authoriz-, not
authenticat-).
4. Contains NO backtick/quote char (` ' ") — a real banner is plain text; quoted
text signals an answer QUOTING the error, e.g. "You'll see `API Error: 401` …
run /login to fix it." (short + 4xx + /login, excluded only by this signal).
Full required matrix (2 KILL + 7 PASS) encoded as tests; npm test green.
CLAUDE_TUI_ERROR_PATTERNS still overrides (a non-empty value REPLACES the default
with operator regexes; empty/whitespace DISABLES detection). New fixture
lib/tui/fixtures/error-401-failauth.jsonl + positive/negative tests retained.
C-2 (P1) — wallclock-truncated partial text returned as silent success.
readTuiTranscript returned {text, entrypoint} identically on the terminal-marker
path and the cap-with-partial path, so a cut-off turn was cached + returned as
finish_reason:stop. It now returns truncated:false on terminal-marker and
truncated:true on the cap-with-partial path (additive field; no-text cap path
still throws). callClaudeTui throws tui_wallclock_truncated on truncated.
C-3 (P1) — verifyEntrypoint only read the turn_duration line.
Some claude builds don't emit turn_duration (Mac mini: absent; PI231/2.1.104:
present), while the entrypoint field is on ordinary lines on BOTH. Reading only
turn_duration made the server.mjs tui_entrypoint_mismatch assertion get got:null
every turn on non-emitting builds. verifyEntrypoint now PREFERS the turn_duration
entrypoint and FALLS BACK to the entrypoint field on any line.
C-4 (P2) — short prompts 100% failed paste-landing.
tuiPromptLanded required needle.length >= 3, so a 1–2 char first line ("hi","ok")
never matched and 5s-failed with tui_paste_not_landed every time (live-repro:
"hi"). Threshold lowered 3 → 2; the input box starts empty (placeholder excluded
by the affirmative-signal design) so a 2-char needle present in the pane is the
prompt. Kept >=2 (not >=1) to avoid collisions with claude's chrome glyphs.
Tests: extended test-features.mjs with unit coverage for each fix + three fixtures
(lib/tui/fixtures/error-401.jsonl, error-401-failauth.jsonl, no-turn-duration.jsonl).
The C-1 block now encodes the full narrowed auth-banner matrix (2 must-kill + 7
must-pass + supporting regression guards incl. a length-cap-load-bearing test).
npm test: 224 passed, 0 failed. Default path (CLAUDE_TUI_MODE unset →
upstreamCall=callClaude) is byte-identical: the only server.mjs change is one import
+ the callClaudeTui body, and callClaudeTui is unreachable when TUI_MODE is off.
ALIGNMENT: Class B (OCP-owned compatibility surface). cli.js does not perform this
operation (TUI is an OCP-owned subscription-pool bridge); scope authorized by ADR
0007 (Class B). No Class A path touched (callClaude / handleUsage / OAuth unchanged);
no change to .github/workflows/alignment.yml or any blacklisted token.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude <claude-opus> <noreply@anthropic.com>
|
||
|
|
6d4751f983 |
feat(tui): OCP_TUI_FULL_TOOLS gate — -p-equivalent tools for single-user TUI (#135)
* feat(tui): OCP_TUI_FULL_TOOLS gate — -p-equivalent tool surface for single-user TUI Lets a SINGLE-USER / trusted TUI deployment run a tool-using / MCP agent (e.g. an OpenClaw assistant) on the subscription pool. When OCP_TUI_FULL_TOOLS=1, buildTuiCmd grants the interactive session the SAME tool surface as the -p A-path — --allowedTools (+ optional --mcp-config / --dangerously-skip-permissions, read from the same CLAUDE_ALLOWED_TOOLS / CLAUDE_MCP_CONFIG / CLAUDE_SKIP_PERMISSIONS env as buildCliArgs) — instead of the default MCP-walled, built-in-tools-only set. Motivation: the default TUI tool wall (--strict-mcp-config --disallowedTools mcp__*) exists for multi-tenant safety, but it also blocks a trusted single-operator agent from its MCP tools — forcing tool-using agents onto the metered -p pool after 2026-06-15. This gate resolves that for the single-user case. Safe to gate ON only because TUI is hard-incompatible with AUTH_MODE=multi (server.mjs refuses to boot, see existing guard), so this can NEVER widen a guest's tool surface. Default (gate unset) is unchanged: MCP wall + built-in tools only. Verified live on PI231 through the real OCP TUI path (spike 2026-06-02): - built-in tools: claude created + read a file on the host (2 tool_use entries); - MCP: claude invoked a custom MCP server's tool (mcp__spike__echo_marker) and returned its output; - billing: entrypoint=cli (subscription pool) WITH full tools + MCP; - completion: end_turn, no permission stall, no hang. Tests: +1 (full-tools branch: --allowedTools present + MCP wall dropped; skip-permissions supersedes; mcp-config threaded). 195 pass. README: new OCP_TUI_FULL_TOOLS env var entry. ALIGNMENT.md: changes lib/tui/session.mjs, not server.mjs — server.mjs hard requirements do not trigger. Mirrors buildCliArgs() permissions logic; the flags are documented Claude Code CLI flags, not invented endpoints. cli.js citation N/A under Rule 2. ADR 0007 (A-path). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tui): shq operator tool tokens in full-tools shell string (reviewer Finding 2) Independent reviewer (APPROVE WITH CHANGES) caught that buildTuiCmd returns a SHELL STRING (run by tmux via sh -c), unlike buildCliArgs which returns an argv array to spawn(). Operator-supplied CLAUDE_ALLOWED_TOOLS can be a scoped specifier such as "Bash(npm run test:*)" or "Read(~/**)", whose ( ) * ~ would break / inject the shell command if pasted bare. Now shq() each allowed-tool token (operator-self-injection only — guests cannot reach TUI per the multi-mode boot guard, but it's a real correctness bug). Also: tightened the README OCP_TUI_FULL_TOOLS wording per reviewer Finding 1 — the precise safety property is "no guest key can reach the TUI path (multi-mode boot is a hard exit)", and noted the AUTH_MODE=shared + OCP_TUI_ALLOW_LAN trust model is unchanged. Test: +scoped-specifier assertions (token shq'd; never appears unquoted). 195 pass. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
d291331998 |
fix(tui): never inject the host's CLAUDE.md / auto-memory into proxied turns (#4) (#132)
OCP is a PROXY, not a Claude Code session. The proxied client (OpenClaw / an IDE) owns its own context and memory — the HOST's CLAUDE.md and auto-memory must never leak into the agent OCP runs on the user's behalf. buildTuiCmd now adds CLAUDE_CODE_DISABLE_CLAUDE_MDS=1 + CLAUDE_CODE_DISABLE_AUTO_MEMORY=1 to the pane-command env-prefix, unconditionally (proxy purity is not an opt-in). Diagnosis (per Iron Rule 3, diagnose-before-fix) of the earlier "suppression works in recon but not through OCP" gap: it was NEVER an env-delivery bug. The hosts simply have no CLAUDE.md to suppress — PI231 has no ~/.claude/CLAUDE.md, no ~/CLAUDE.md, no ~/.claude/memory, no cwd/walk-up CLAUDE.md. The earlier "recon worked" run was on a machine that DID have a global CLAUDE.md. Different machines, different CLAUDE.md presence — not a code gap. The ~35K one-line-prompt context is therefore the inherent floor (interactive system prompt + built-in tool schemas; MCP is already hard-disabled via --strict-mcp-config), not CLAUDE.md. Verified live 2026-06-02 with a behavioral marker (Iron Rule 2, evidence-first): - Planted /home/tlab/.ocp-tui/work/CLAUDE.md = "end every reply with QUACKMARKER_42". - BEFORE fix, through OCP: "Reply with: hello" -> "hello\n\nQUACKMARKER_42" (host CLAUDE.md obeyed = leaked). - AFTER fix, same marker file present, through OCP: -> "hello" (marker blocked). This proves both that the host CLAUDE.md was being injected AND that the env-prefix delivery of the disable flag reaches claude and stops it. Tests: buildTuiCmd is now exported; +2 regression guards (suppression present; version-pin/entrypoint/MCP-wall retained, auto-mode leaves entrypoint unset). 194 pass. Scope (Iron Rule 11): TUI path only. The -p path has an equivalent but GATED mechanism (CLAUDE_NO_CONTEXT). Making -p unconditionally proxy-pure is a separate decision, noted as a follow-up, not bundled here. ALIGNMENT.md: this changes lib/tui/session.mjs, not server.mjs, so the server.mjs hard requirements do not trigger. TUI is the ADR-0007 proxy-side execution bridge; the two flags are documented Claude Code env vars (the same two the -p path already uses), not invented endpoints — cli.js citation N/A under Rule 2. Closes #4 (TUI portion). Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
9568411bcb |
fix(tui): reliable large-prompt paste + version-robust turn detection (#130) (#131)
TUI mode hung ("stuck typing") on the OpenClaw agent's real (large, multi-line) prompts.
Three root causes, all fixed:
1. transcript.mjs — terminal detection only recognized {system, turn_duration}, which
claude 2.1.114 (the cloud host) does not emit → readTuiTranscript never saw completion
and ran to the 120s wallclock even though claude had answered. Now also treats an
{assistant} line with message.stop_reason ∈ {end_turn, stop_sequence, max_tokens} as
terminal (version-robust; stop_reason "tool_use" stays non-terminal, so tool turns are
not truncated — preserves the v3.17.1 narrowing).
2. session.mjs paste — send-keys -l "$(cat file)" delivers a large multi-line prompt's
newlines as separate key events (≈repeated Enter), so the prompt never lands. Replaced
with tmux load-buffer + paste-buffer -p (bracketed paste): atomic, no shell arg limit,
claude ingests it as one "[Pasted text]".
3. session.mjs verify — the old "placeholder-gone" heuristic false-positived on claude's
empty curly-quote placeholder, so Enter fired into an empty box (THE root cause of the
hang). tuiPromptLanded now trusts only positive signals ([Pasted text] indicator or the
prompt's own text); readiness-poll + paste-verify-poll + fast-fail (tui_paste_not_landed)
turns a 120s wallclock hang into a deterministic ~5s error.
Also: env delivered to the pane via an `env`-prefix on the pane command (tmux does NOT
forward spawnSync's env, and new-session -e needs tmux ≥3.2 while the cloud host runs 2.7),
carrying DISABLE_AUTOUPDATER (version pin) + CLAUDE_CODE_ENTRYPOINT labeling.
Validated live on both hosts: a 300-line / ~30KB prompt that previously hung now returns
in ~5-8s; small prompts ~4-5s. 192 tests pass (new: terminal-detection across schemas,
positive-signal verify incl. the curly-quote false-positive guard, paste predicates).
Scope discipline: an attempt to also strip claude's CLAUDE.md/auto-memory injection (context
reduction) was REVERTED — it didn't measurably reduce context through OCP, caused an
off-response, and is a separate concern. It will be its own measured branch.
ALIGNMENT.md: TUI is the ADR-0007 proxy-side execution bridge, not a forwarded Anthropic
operation, so a cli.js citation is N/A under Rule 2. No blacklisted tokens / port literals.
Independent fresh-context reviewer (opus): APPROVE WITH CHANGES (Iron Rule 10) — validated
the three fixes, caught that a per-session tmux socket would break the startup stale-session
reaper, and flagged the context-suppression scope-creep; both were reverted per the review.
Closes #130.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
|
1b02f181fa |
refactor: extract isLoopbackBind to lib/net.mjs (#125) (#127)
Follow-up cleanup from #115's review. isLoopbackBind was defined in server.mjs and copy-pasted into test-features.mjs (with a "keep in sync" comment, because importing server.mjs would run server.listen()). Extracted to a new importable lib/net.mjs; server.mjs and the test now import the one definition, removing the drift surface. Pure refactor — the function body is byte-identical to the prior server.mjs version, the TUI LAN-gate call site is unchanged, and the 8 isLoopbackBind truth-table tests now exercise the real shared function. 181 tests pass. ALIGNMENT.md: touches server.mjs but adds/removes no operation — it relocates an existing (#115) helper into a module. cli.js citation N/A under Rule 2. Independent fresh-context reviewer (opus): APPROVE (Iron Rule 10) — byte-identical body, exactly one definition, wiring + scope correct, no test dropped, scope clean. Closes #125. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
aa1c65beb1 |
fix: TUI hardening — non-loopback LAN gate + assert cc_entrypoint (#115) (#122)
Two TUI-mode findings from the 2026-05-31 audit (ADR 0007):
1. The fail-loud LAN gate refused TUI boot only when CLAUDE_BIND === "0.0.0.0",
but binding to a concrete LAN IP, a Tailscale 100.x address, or IPv6 :: is
equally network-exposed and slipped through — letting any reachable peer drive
the operator's full-filesystem claude session. New isLoopbackBind() treats only
127.0.0.0/8 / ::1 / localhost / ::ffff:127.0.0.1 as safe; the gate now trips on
any non-loopback bind (OCP_TUI_ALLOW_LAN=1 escape hatch retained).
2. verifyEntrypoint() was implemented and unit-tested but never wired: readTuiTranscript
discarded the events and returned only text, so a silent degrade to the metered
sdk-cli (Agent SDK) billing pool — which still returns text but costs money — went
undetected. readTuiTranscript now returns { text, entrypoint }; runTuiTurn passes it
through; callClaudeTui logs a "tui_entrypoint_mismatch" warning when configured cli
mode yields a non-cli entrypoint (including null/unverified). callClaudeTui still
returns Promise<string>, so singleflight/cache/completion downstream is unchanged.
ALIGNMENT.md: TUI is a proxy-side execution-mode bridge (ADR 0007); this adds a local
bind classifier (startup gate) + a non-fatal log assertion of an existing classification
— no Anthropic operation forwarded, so a cli.js citation is N/A under Rule 2. No
blacklisted tokens or port literals introduced; alignment.yml passes.
Independent fresh-context reviewer (opus): APPROVE WITH MINOR (Iron Rule 10) — verified
isLoopbackBind via truth table (no exposed address misclassified as loopback — the
dangerous direction), the full string→{text,entrypoint} contract ripple across all
callers (Promise<string> preserved), the mismatch logic (warns on sdk-cli AND null in
cli mode, silent for auto/off, non-fatal), and ALIGNMENT N/A via a fetch/header grep.
Minors are process-only (this commit body; a future lib/ extraction of the test-mirrored
helper). npm test → 181 passed, 0 failed.
Closes #115.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
|
879b40fe93 |
fix: escape dashboard DB-sourced values + validate key names (#114) (#121)
The dashboard built table rows via template-literal innerHTML, interpolating
DB-sourced strings (key names, usage rows) with no HTML escaping, and an
onclick="revokeKeyUI('${k.name}')" sink a single quote could break out of. Key
names were unvalidated at creation. Admin-gated (self-XSS today), but a real
unescaped-sink gap that becomes cross-user if key creation is ever delegated.
- dashboard.html: added escapeHtml() and applied it to every DB/string-sourced
interpolation in refreshUsage and refreshKeys (key_name, name, keyPreview,
created_at, last_request, model). Replaced the inline-onclick revoke button with
a data-revoke attribute + addEventListener, so a name can never break out into an
event-handler string. (model is attacker-pickable via the request body, so its
escaping is the load-bearing one.)
- server.mjs: POST /api/keys now rejects names not matching /^[A-Za-z0-9 ._-]{1,64}$/
before createKey() — defense-in-depth so a <script>/quote name can never reach the
DB. Creation-only; existing keys unaffected; the default key-${Date.now()} passes.
Noted (out of scope, optional follow-up): the status/plan summary cards render
trusted server/Anthropic-upstream values unescaped — non-user-controlled, so not part
of this stored-XSS fix.
ALIGNMENT.md: the server.mjs change is proxy-policy input validation with no Anthropic
operation forwarded, so a cli.js citation is N/A under Rule 2. No blacklisted tokens or
port literals introduced; alignment.yml passes.
Independent fresh-context reviewer (opus): APPROVE WITH MINOR (Iron Rule 10) — verified
escapeHtml correctness, full sink coverage (incl. the attacker-pickable model field and
the data-revoke attribute), revoke round-trip via getAttribute, the anchored/bounded
key-name regex running before createKey, and that createKey has no other unvalidated
caller. Both minors non-blocking (trusted status-card escaping; this commit-body note).
Closes #114.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
|
68d58e7df4 |
fix: CLI/installer hardening — restart labels, key permissions, unit-secret escaping (#113) (#120)
Three CLI/installer findings from the 2026-05-31 audit (no server.mjs): 1. ocp-plugin `cmdRestart` hardcoded uid 501 + the legacy `ai.openclaw.proxy` label, and fell through to a dangerous `pkill -f 'node.*server.mjs' && cd ~/.openclaw/projects/*/; node server.mjs &` that could kill an unrelated node process and relaunch an env-less ghost proxy from a glob-ambiguous dir. Now uses process.getuid() + the live labels (dev.ocp.proxy on macOS, ocp-proxy on Linux systemd; the OpenClaw gateway label is unchanged) and drops the pkill fallback entirely (returns a manual `ocp restart` message on failure). 2. ocp-connect wrote the quota key unquoted into rc files and a world-readable environment.d/ocp.conf. Now single-quotes the value and chmod 600s the rc files and ocp.conf (matching the existing auth-profiles.json 0o600 convention). 3. setup.mjs interpolated the injected service-unit secrets (CLAUDE_BIN, OCP_ADMIN_KEY, PROXY_ANONYMOUS_KEY) raw into plist <string> and systemd Environment= lines. Added xmlEscape() for all plist <string> values and assertSafeInjectValue() which rejects control characters (\x00-\x1f — newline, CR, tab) before any unit is written, blocking a newline-injected rogue Environment= directive. Spaces are intentionally allowed (CLAUDE_BIN paths may contain them). XML-escaping on write also resolves plist-merge.mjs's [^<]* regex concern transitively (no raw < reaches it) — comment added, logic unchanged. ALIGNMENT.md: CLI/installer scripts only, no Anthropic operation forwarded → cli.js citation N/A. No blacklisted tokens or port literals introduced; alignment.yml passes. Independent fresh-context reviewer (opus): APPROVE (Iron Rule 10) — verified the control-char regex byte-exact via od (no space-rejection regression), the pkill fallback fully removed, restart labels match setup.mjs ground truth, OCP keys (base64url) cannot break the single-quoting, and the validator runs before any write. Closes #113. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
4a7d79c330 |
fix: record OAuth-host verification + models.json SPOT for usage-probe/default (#112) (#119)
Three alignment/SPOT findings from the 2026-05-31 audit: 1. The OAuth token-refresh host (platform.claude.com/v1/oauth/token, a Class A surface) was introduced in the 2026-04-11 drift commit and had no verification record. Verified against the compiled cli.js (claude.exe v2.1.154) via `strings`: OAUTH_TOKEN_URL and OAUTH_CLIENT_ID both appear in the binary byte-for-byte (in the same `prod` config object), and the legacy host console.anthropic.com/v1/oauth is absent (0 hits). Recorded this as an inline ALIGNMENT citation comment above the constants. No live OAuth probe was run — a refresh-token grant would rotate the operator's real credentials; the strings-on-binary evidence is decisive. (cli.js: verified against compiled claude.exe v2.1.154, 2026-05-31.) 2. fetchUsageFromApi() hardcoded the haiku model ID; now derives from modelsConfig.aliases.haiku (ADR 0003 SPOT). Prevents a silent /usage break on a future haiku ID bump. 3. [P3] The default request model hardcoded the sonnet ID; now derives from modelsConfig.aliases.sonnet (ADR 0003 SPOT). Both SPOT values are byte-identical to the literals they replace today, so zero behavior change — only drift-resistance. The alignment.yml blacklist pin for the wrong-host variant was deliberately NOT included here: extending the blacklist is a governance-layer change (alignment.yml inline policy) and belongs in its own PR. ALIGNMENT.md: finding 1 IS the verification (cli.js citation recorded inline); findings 2-3 are SPOT hygiene that forward no new operation. No blacklisted tokens or port literals introduced; alignment.yml passes. Independent fresh-context reviewer (opus) INDEPENDENTLY re-ran `strings` on the binary and confirmed the host/client_id present and the legacy host absent — APPROVE (Iron Rule 10; alignment hard-requirement #3 satisfied). Closes #112. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
c3b1f32c86 |
fix: error-output sanitization + process-lifecycle hardening (#111) (#118)
Four findings from the 2026-05-31 audit (3×P2 + 1×P3), all in the error/process
lifecycle layer:
1. sanitizeError() helper — streaming error paths sent raw claude error_message /
stderr to the client, leaking home-dir / credential-file paths that the
non-streaming path already redacted. Factored the path-strip regex into one
helper and applied it at all 9 client-facing jsonResponse/sendSSE error emits;
de-duped the 3 pre-existing inline .replace() sites. Operator-log calls
(logEvent/trackError) and admin-gated endpoints left raw by design.
2. res.on("close") SIGKILL escalation — a client disconnect sent only SIGTERM; a
SIGTERM-resistant child held its concurrency slot until the request timeout
(narrow #37 on the hottest exit path). Now escalates to SIGKILL 5s after SIGTERM,
cleared on proc exit. Per review: gated on the child still being alive
(exitCode===null && signalCode===null) so the normal-success close no longer
fires a spurious SIGTERM or leaks the 5s timer; killTimer.unref() so a genuine
disconnect timer never delays graceful shutdown.
3. Per-key quota TOCTOU — documented as best-effort/eventually-consistent (inline
comment + README note): concurrent requests at the boundary can overshoot by up
to MAX_CONCURRENT and cache hits are uncounted. Chose documentation over an
in-flight counter to avoid a decrement-on-all-paths liability (the #37 class) on
a low-blast-radius internal family rate-limiter — not a payment boundary.
4. [P3] overallTimer cleared on semantic completion — the request timer was cleared
only on proc exit, so a streamed response that res.end()'d before the child
exited could record a spurious post-success timeout. New clearOverallTimer()
(clears the timer ONLY, never touches the `cleaned` slot-accounting flag — no
slot leak) is called in the streaming stop-success path; cleanup() on exit still
clears it idempotently and decrements the slot.
ALIGNMENT.md: error-shaping / process-lifecycle / rate-limit documentation forward
no Anthropic operation, so a cli.js citation is N/A under Rule 2. No blacklisted
tokens or port literals introduced; alignment.yml passes.
Independent fresh-context reviewer (opus): APPROVE WITH MINOR (Iron Rule 10) — the
two critical concerns (clearOverallTimer slot-leak class, SIGKILL double-kill) were
verified clean; MINOR #1 (kill-timer leak on success path) folded in; MINOR #2
(pre-existing regex over-redaction of ratios/URLs) left as out-of-scope.
Closes #111.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
|
4458490caa |
fix: request validation + OpenAI-compat correctness (#110) (#117)
Three correctness/compat defects on the request path:
1. Non-array `messages` (e.g. {"messages":"x"}) passed the `!messages?.length`
guard but threw in `messages.reduce(...)`; since the handler runs without
await, the rejection silently hung the client until socket timeout. Now
guarded with `Array.isArray(messages) && length>0` → 400 invalid_request_error,
placed before the first use of `messages`.
2. OpenAI array `content` ([{type:"text",text},{type:"image_url",...}]) was
JSON.stringify'd into the prompt as literal noise. New `contentToText()` helper
flattens text parts and replaces non-text parts with a placeholder; used in
messagesToPrompt, extractSystemPrompt, and all promptChars char-count sites
(zero `JSON.stringify(m.content)` remain).
3. A streamed upstream error arriving after eager SSE headers emitted a bare
finish_reason:"stop" + [DONE] — byte-identical to a successful empty completion.
Both streaming error paths (the parsed.error result branch AND the non-zero-exit
close-handler branch — the latter folded in per reviewer) now emit an SSE
`data:{"error":{...}}` frame so clients can distinguish failure from empty.
Error-text sanitization across all emit sites is intentionally deferred to #110's
sibling issue #111 (security layer).
ALIGNMENT.md: these are OpenAI-compat shim + input-validation behaviors; OCP does
not forward, add, or alter any Anthropic API operation here (cli.js does not speak
the OpenAI wire format), so a cli.js citation is N/A under Rule 2. No blacklisted
tokens or port literals introduced; alignment.yml passes.
Independent fresh-context reviewer (opus): APPROVE WITH MINOR (Iron Rule 10) —
the close-handler sibling fold-in addresses MINOR #1; this commit body addresses
MINOR #2.
Closes #110.
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
|
||
|
|
36be723198 |
fix(security): gate /health anonymousKey behind opt-in PROXY_ADVERTISE_ANON_KEY (#109) (#116)
/health is an unauthenticated, LAN-reachable endpoint that returned the live anonymous bearer key (anonymousKey: PROXY_ANONYMOUS_KEY). Any device that could reach the port harvested a working, quota-spending credential (P0). The anonymousKey field is now included only when the caller is localhost (already fully trusted by the auth path, via the unspoofable req.socket.remoteAddress) OR the admin explicitly opts in with the new PROXY_ADVERTISE_ANON_KEY=1 env var (default off). ocp-connect's absent-field fallback (interactive --key / anonymous access) is unchanged — comment-only update there. ALIGNMENT.md: this gates an OCP proxy-policy field in /health; it does NOT forward, add, or alter any Anthropic API operation, so a cli.js citation is N/A under Rule 2. No blacklisted tokens introduced; alignment.yml passes. Independent fresh-context reviewer (opus): APPROVE WITH MINOR (Iron Rule 10). Closes #109. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> |
||
|
|
05a984df89 |
fix: OCP code-audit P1+P2 hardening (crash bugs + multi-tenant gates) (#106)
* fix(daemon): P1-1 guard proc.stdin against EPIPE crash
In spawnClaudeProcess, attach an error listener on proc.stdin BEFORE
the write/end calls so an EPIPE (child closed stdin mid-write) is
swallowed and logged rather than thrown as an unhandled exception.
The existing proc.on("error") listener is on the ChildProcess object,
NOT on the stdin Writable — it does not catch stdin write errors.
Hardening per OCP code audit; entry-surface contract unchanged for
single-user default path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(tui): P1-3 remove tool_use from isTerminalLine; only turn_duration is terminal
In interactive TUI mode, stop_reason=tool_use does NOT mean the turn is
complete. Claude handles the tool call internally and continues generating —
the transcript advances to another assistant entry. Treating tool_use as
terminal truncated tool-using turns mid-flight.
Only {type:"system", subtype:"turn_duration"} is the authoritative
completion marker (claude CLI v2.1.157+ interactive session transcript).
Updated two unit tests that previously asserted tool_use → true; they now
assert false (the correct behaviour). The real-fixture terminal detection
test is unaffected because the fixture uses turn_duration.
Hardening per OCP code audit; TUI path behaviour fix.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(tui): P2-5 add -l (literal) flag to send-keys prompt paste
A prompt that equals a tmux key token (e.g. "C-c", "Escape") would be
interpreted as that key binding rather than typed as literal text.
The -l flag forces literal character-by-character input. The separate
Enter key event afterward deliberately omits -l so tmux sends a real
carriage-return keypress to submit the prompt line.
Authority: tmux send-keys(1) § -l flag. Hardening per OCP code audit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(audit): String-coerce parsed.error + stale tool_use comment (review fast-follow)
Folds the 2 minor findings from the independent review of the audit fixes:
- String(parsed.error) before .slice/message in callClaude + callClaudeStreaming
(defensive: claude could emit a non-string result/error_message).
- correct the readTuiTranscript comment that still listed tool_use as terminal.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: dtzp555 <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
||
|
|
74260d7f6f |
feat(tui): opt-in CLAUDE_TUI_MODE — interactive subscription-pool bridge (PR-1..PR-4 squashed) (#104)
Reader + driver + home modes + server wiring + entrypoint hardening. 5 independent reviews folded; e2e green; default stream-json path byte-identical when flag off. See PR description + ADR 0007 for the full layer/authority/security detail. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
|
885f62addf |
feat: Phase 6c — claude stream-json port + --system-prompt + opus 4.8 (#103)
* wip(rescue): OCP Phase 6c stream-json port + opus 4.8 (uncommitted /tmp work) Rescue commit — preserves the subagent-produced Phase 6c port (claude -p → stream-json + --system-prompt + NDJSON parser) and opus 4.8 model addition done 2026-05-30 in a /tmp clone (volatile). NOT reviewed-for-merge; this is preservation only on a WIP branch so a /tmp reboot does not lose the work. Strategy context: OCP is being made the first-mover for TUI-mode (users are on OCP). Phase 6c here is the DEFAULT-path foundation (stream-json = sdk-cli = the "Max-user stable mode" fallback). TUI-mode (cc_entrypoint=cli subscription bridge) lands on top as an opt-in. Re-review before merging to main. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(phase6c): fold review findings — scrub hostname, README prose, parser tests, version citation P3-1 (binding): scrub private hostname "PI231" from server.mjs comment; replace with role-based term "the test server". Repo is public — no machine-specific hostnames may appear. P3-2: update README.md "How It Works" diagram and prose from stale "claude -p" to "claude --output-format stream-json", matching the Phase 6c spawn change already in server.mjs and CHANGELOG. P2-2: update all five authority citations in server.mjs from bare "claude CLI v2.1.104" to "claude CLI § <flag> (ported from OLP, verified v2.1.104; behavior stable through v2.1.158)" — honest about OLP provenance and version range, without inventing new facts. P2-1: add "Stream-JSON parsers" suite to test-features.mjs (17 new tests, 84 → 101 total). Copies parseStreamJsonLines and parseStreamJsonEvent verbatim from server.mjs with a MIRRORS header comment; logEvent stubbed to avoid live-server side-effects. Covers: content_block_delta deltas, assistant-aggregate fallback, no-double-count guard, aggregate-only short responses, partial-line buffering across two chunks, is_error result variants, malformed/non-JSON lines, system/user/unknown event types. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
|
7a69d72886 |
feat(snapshot): gcSnapshots + ocp update --rollback --gc + auto-GC; v3.16.0 (#94)
Adds snapshot garbage collection with retention policy: keep last 5, keep snapshots within 30 days, always keep the most recent. Configurable via keepCount / keepDays opts. Wire-up: - ocp update --rollback --gc (manual trigger; --dry-run supported) - runFullUpgrade auto-GC after successful upgrade (best-effort, swallows errors) 4 unit tests: keepCount enforcement, keepDays override, never-delete- most-recent safety, dry-run mode. Bumps to v3.16.0 (bundles PR #93 --check oauth + this GC feature). No cli.js citation needed: this is OCP-internal snapshot lifecycle with no corresponding cli.js operation. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
|
a8601a6d30 |
feat(doctor): --check oauth fast path (#93)
* feat(doctor): --check oauth fast path Implements the --check oauth fast path documented in cmd_doctor_help but previously unimplemented. Skips version detection, from-version check, git operations, and models endpoint — runs only the curl against /health + auth.ok extraction. Use cases: - After `claude auth login`, fast verify OCP can spawn cli.js - After a known service blip, quick health gate before larger ops - AI agent's setup-repair loop: ./ocp doctor --check oauth in a retry-after-fix step 3 unit tests cover: PASS path, OAuth FAIL → fix_oauth, service down → fix_service. No cli.js citation needed: this is OCP-internal doctor logic with no corresponding cli.js operation. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore(doctor): nit fixes for --check oauth (N3 body=null test + N4 skipped sentinel comment) --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
|
cd6ec2a212 |
fix(doctor): dynamic latest_version from origin/main; release v3.15.1 (#92)
v3.15.0 doctor used a hard-coded `latest = "v3.14.0"` fallback, causing any v3.15.0+ install to report kind=upgrade against a stale value. `ocp update` would then attempt `git checkout v3.14.0` — a downgrade. Doctor now fetches `git -C ~/ocp show origin/main:package.json` to determine the actual latest. On failure (offline, fresh clone, no remote), falls back to currentVersion so kind=noop instead of recommending a downgrade. Regression test added: doctor with unreachable ocpDir falls back to currentVersion as latest (not the old hardcoded v3.14.0). Caught during v3.15.0 post-deploy verification on home-mac: ./ocp doctor reported `kind=upgrade` immediately after v3.15.0 install, which would have been a critical user-facing bug. No cli.js citation needed: this is OCP-internal doctor logic with no corresponding cli.js operation. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> |
||
|
|
ab03c13332 |
feat(upgrade): ocp doctor + cross-version ocp update + rollback + AI prompts (#91)
* feat(doctor): add ocp doctor with --json + next_action contract Implements scripts/doctor.mjs with semver-aware path selection (noop/update/upgrade/fresh_install/fix_oauth/fix_service) and the JSON contract documented in the design spec. Service health + OAuth checks integrated; mockable via opts.mockHealth for unit tests. 8 unit tests cover the kind dispatch tree and the next_action shape for each kind. No cli.js citation needed: this is OCP-internal tooling with no corresponding cli.js operation. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(ocp): wire cmd_doctor into bash CLI; dispatch to scripts/doctor.mjs Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(doctor): handle unparseable version + empty health body Three issues raised by code-quality reviewer on |
||
|
|
55c576bbb1 |
fix(setup): preserve user-customised plist/systemd env vars on re-setup (#90)
* fix(setup): merge plist/systemd env vars instead of overwriting
setup.mjs previously wrote the launchd plist and the Linux systemd unit
with writeFileSync(path, NEW_TEMPLATE_STRING), which silently dropped any
user-customised env vars (CLAUDE_HEARTBEAT_INTERVAL, CLAUDE_CACHE_TTL,
etc.) on every re-setup or upgrade. This change introduces
scripts/lib/plist-merge.mjs which preserves keys present only in the
existing file and lets the template's known keys win. Linux variant uses
the same logic against Environment=KEY=VALUE lines.
Tests added in test-features.mjs cover preserve / override / first-install
for both formats.
No cli.js citation needed: this is OCP-internal installer behaviour with
no corresponding cli.js operation.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(setup): plist-merge code-quality follow-up
Three issues raised by the code-quality reviewer on
|
||
|
|
5ff30ac9b6 |
feat(cache): singleflight stampede protection on non-streaming path (#66)
cli.js does not perform proxy-layer stampede protection. The singleflight layer is a value-add proxy operation that exists only inside OCP, between concurrent client requests and the single upstream cli.js spawn. It does not introduce, alter, or remove any endpoint, header, request field, or response field that cli.js emits or expects — no client-observable wire shape change. Justification under ALIGNMENT.md Rule 2: the singleflight Map deduplicates concurrent identical non-streaming cache-miss requests so only one cli.js spawn runs per unique hash window. All followers receive the same resolved (or rejected) content. This is a proxy-internal concurrency optimization, not a wire-level protocol change. Spec: docs/superpowers/specs/2026-05-07-cache-upgrade-design.md (D4) Changes: - keys.mjs: add singleflight(hash, fn) + getInflightStats() exports; in-memory Map cleared via Promise.finally() on each settlement - server.mjs: import singleflight + getInflightStats; wrap non-streaming cache-enabled path through singleflight with inner recheck; add TODO comment in callClaudeStreaming (streaming-path dedup is explicitly out of scope for v3.13.0, see spec D4 streaming caveat); extend /cache/stats to return inflight + requesters fields (additive, no removed fields) - test-features.mjs: 7 new PR-B singleflight tests covering basic dedup, failure fan-out, map cleanup (success + failure), different-hash independence, getInflightStats shape, and sequential-call non-sharing; all 31 tests pass (24 existing + 7 new) Streaming-path singleflight is explicitly out of scope; TODO left in callClaudeStreaming for a future follow-up ticket. Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
|
16eeb66557 |
feat(cache): per-key isolation, cache_control bypass, chunked stream replay (#65)
* docs(governance): ADR 0005 (no multi-provider) + cache upgrade spec Governance prelude for the cache upgrade work: - docs/adr/0005-no-multi-provider.md — locks in the decision that OCP stays single-provider (Anthropic via cli.js spawn). Cache improvements are explicitly in scope (decision §3); multi-provider refactor is explicitly out of scope, with three documented trigger conditions for revisiting. - docs/adr/README.md — index updated to reference 0005. - docs/superpowers/specs/2026-05-07-cache-upgrade-design.md — design for the cache upgrade work split into PR-A (per-key isolation, cache_control bypass, chunked stream replay) and PR-B (singleflight stampede protection). Each design decision has a written rationale. server.mjs is not modified; this commit is doc-only and therefore exempt from the cli.js citation requirement (ALIGNMENT.md Rule 5 applies only to commits that touch server.mjs). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * feat(cache): per-key isolation, cache_control bypass, chunked stream replay cli.js does not perform response caching at the proxy layer. OCP's response cache is a value-add operation internal to OCP, between the wire and the cli.js spawn. It does not introduce, rename, or alter any endpoint, header, request field, or response field that cli.js emits or expects — this change qualifies under ALIGNMENT.md Rule 2's value-add carve-out for non-wire- affecting proxy operations. no client-observable wire shape change. Spec: docs/superpowers/specs/2026-05-07-cache-upgrade-design.md D1 — Per-key cache isolation (cacheHash v2 format) keys.mjs: prepend `v2|k:<keyId or "anon">|` before existing hash fields. Backward-compatible: absent/null/empty keyId folds to "anon". v1-format rows in response_cache are abandoned naturally; TTL cleanup at server.mjs:185 reaps them within one window. No migration needed. server.mjs: pass keyId: req._authKeyId at the single cacheHash call site (line ~1221). D2 — cache_control bypass keys.mjs: export hasCacheControl(messages) — walks messages and nested content arrays for presence of cache_control field. server.mjs: if hasCacheControl(messages) is true, set req._cacheHash = null and log cache_skipped{reason: cache_control_present}; existing `if (CACHE_TTL > 0 && req._cacheHash)` guards on write-back handle the skip. D3 — Chunked stream replay (80 codepoints/chunk, no artificial delay) server.mjs: replace single-chunk cached.response emission with an Array.from(cached.response) loop in steps of CACHE_REPLAY_CHUNK_SIZE=80. Array.from ensures multibyte UTF-8 codepoints (e.g. CJK) are never split. Tests: 12 new cases in test-features.mjs (36 total, 0 failed). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> |
||
|
|
97ca91341c |
feat(server): per-key quota + response cache (v3.8.0) (#18)
Add two new features for LAN sharing governance:
Quota (budget control):
- Per-key daily/weekly/monthly request limits (NULL = unlimited)
- Idempotent schema migration for quota columns
- Single-query check (SUM/CASE) for all 3 periods — no N+1
- PATCH /api/keys/:id/quota (partial update, input validation)
- GET /api/keys/:id/quota (current limits + usage)
- 429 with structured error when exceeded
- Only applies to identified per-key users, not admin/anonymous
Response cache:
- SHA-256 hash of model + messages + temperature/max_tokens/top_p
- Opt-in via CLAUDE_CACHE_TTL env var (0 = disabled, default)
- Cache hit serves both streaming (simulated SSE) and non-streaming
- Streaming responses accumulated and cached on success
- Skips multi-turn sessions (conversationId present)
- GET /cache/stats, DELETE /cache admin endpoints
- Runtime-tunable cacheTTL via PATCH /settings
- 10-minute periodic cleanup of expired entries
Bug fix discovered during testing:
- SQLite datetime('now') stores 'YYYY-MM-DD HH:MM:SS' but JS
.toISOString() produces 'YYYY-MM-DDTHH:MM:SS.sssZ'. String
comparison breaks for same-day ranges. Added sqliteDatetime()
helper for correct format matching.
Code review fixes:
- DELETE /api/keys/:id no longer shadows /quota sub-routes
- updateKeyQuota uses partial UPDATE (only SET provided fields)
- cacheHash includes temperature/max_tokens/top_p in hash
- Replaced raw getDb() in server.mjs with findKey() encapsulation
- Unified UTC midnight calculation across checkQuota/getKeyQuota
Includes 24-test integration suite (test-features.mjs).
Co-authored-by: Tao Deng <dtzp555@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|