mirror of
https://github.com/dtzp555-max/olp.git
synced 2026-07-21 21:15:10 +00:00
40f9453d88b82e86cd388c2f0dddac7e1e0e57b0
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 
|
40f9453d88 |
fix(ir): accept OpenAI role=developer at entry surface, normalize to system (#65)
* fix(ir): accept OpenAI role=developer at entry surface, normalize to system Hermes Agent (v0.13+) and other modern openai-completions clients (Cline, Continue.dev) default to role=developer for the high-priority instruction slot when the model id matches OpenAI o1/o3+ reasoning family. OLP IR's validator rejected this with 400 IR validation failed: role must be one of system|user|assistant|tool, got "developer". Reproduced 2026-05-27 on PI230 Hermes v0.14.0 trying olp-codex/gpt-5.5 through PI231 OLP. olp-claude (Anthropic models) was unaffected because Hermes sends system role for Claude models (no developer-role concept on Anthropic side). Fix: extend openai-to-ir.mjs normalizeRole to map developer to system at the entry boundary. The IR canonical-four-roles invariant is preserved; every provider plugin's role-handling stays unchanged. Same pattern as the existing function-to-tool normalization that already lives in normalizeRole (function role was OpenAI-deprecated). Normalize-at- entry centralizes role-spec-evolution handling in one file rather than bloating the IR schema and forcing every provider plugin to handle each new role. Why not add developer to VALID_ROLES: it would require branches in three provider plugins (anthropic, codex, mistral) all mapping to system-style annotation anyway, plus wider IR surface for any future OpenAI role addition. Tests: two new pin tests in Suite IR translation: - developer role to system translation - mixed-role array including developer validates cleanly 768 to 770 tests, 0 fail. Verified Hermes via olp-codex no longer hits the IR rejection error after this fix (smoke run from PI230). ADR 0003 Amendment 3 documents the rationale + future-role policy (normalize-at-entry unless a role genuinely conveys provider-distinguishable semantics). Authority: OpenAI Responses API spec developer-role + Hermes Agent v0.14.0 reproduction. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * test+docs: PR #65 fold-in reviewer N2 + N3 N2 — Negative control test added: unknown role (e.g. "admin") still raises BadRequestError. Pins that the developer-to-system normalization is the only entry-surface escape hatch; any future widening of normalizeRole that returns role as-is for unknown inputs will be caught by this test. N3 — ADR 0003 Amendment 3 gains a "Cache-key impact" bullet documenting that a developer-form and system-form request with otherwise-identical content now share the same IR and thus the same cache key (per ADR 0005). This is by design and matches OpenAI's own backward-compat semantics; the note exists so a future debug session investigating "why does my developer request hit a cache entry from an old system request" finds the answer. Test count: 770 to 771, all pass. Skipped reviewer N1 (URL precision) — Amendment already cites multiple authorities including the Hermes/Cline tracking convention and live reproduction transcript; single-URL precision is over-spec. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: dtzp555 <dtzp555@gmail.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> |
||
|
taodengandClaude Opus 4.7
|
f784fdb947 |
fix+docs: D33 — round-5 cleanup batch (F1/F3/F5/F8/F9/F10/F11/F12)
cold-audit catch from 2026-05-24 (round 5)
Round-5 cold-audit cleanup batch. 8 items + 1 release-discipline
reconciliation. Largest batch by line count (582+/39-) but every item
is small-and-focused. 3 P2 items (F1/F3/F5 of which F3 + F1 are real
correctness/observability fixes; F5 backfills /health to spec).
Changes (10 files, +583/-39):
**P2 fixes**
1. **F1 — ALIGNMENT.md mistral authority pin self-contradicted plugin**
(ALIGNMENT.md): row cited `vibe --prompt --output json` but mistral.mjs
uses `--output streaming` (the plugin header at lines 360-369 even
justifies WHY: `--output json` emits single blob, breaks NDJSON
line-buffered parser). Constitution self-contradicting itself —
missed across 4 prior rounds. Pin updated to `--output streaming`
with DOCS-1 reference.
2. **F3 — Deterministic function_call synth ID**
(lib/ir/openai-to-ir.mjs): deprecated `function_call` translation
produced `id: \`fc-${Date.now()}\`` → ID flows into normalized
tool_calls → cache key SHA-256. Two identical requests separated
by ≥1ms → different cache keys → cache always misses for
`function_call` request shape. Violates ADR 0005 invariant
"same inputs → same key, no random, no timestamp."
Fixed: id is now `fc-<16-hex>` from SHA-256 of `${name}\0${arguments}`.
NUL separator prevents the (name='ab',args='c') vs (name='a',args='bc')
collision. 2^64 collision resistance is more than sufficient for
tool_call ID disambiguation (per-request semantic key, not crypto
primitive).
**P3 fixes**
3. **F5 — /health invokes per-plugin healthCheck()** (server.mjs +
docs/openai-spec-pin.md): ADR 0002 says "healthCheck — startup AND
/health endpoint use this." Pre-D33 /health returned only
{enabled, available} counts. Now async, iterates loadedProviders,
awaits each plugin's healthCheck() in try/catch. Returns
`providers: {enabled, available, status: {<name>: {ok, latencyMs?, error?}}}`.
4. **F8 — X-OLP-Cache reports fallback-hop cache hits** (server.mjs):
pre-D33 cacheStatus computed from `preCheckHit && fallbackHops === 0`
— only counted primary-hop cache hits. When fallback fires + the
fallback hop's getOrCompute returns from cache, header reported
`miss` despite no spawn happening.
Fixed: peek BEFORE getOrCompute inside executeHopFn, set
`lastHopWasCached` closure variable on every hop (last-write-wins
= serving hop's state). cacheStatus combines
`lastHopWasCached || (preCheckHit && fallbackHops === 0)`.
F8 chose option (b) peek-then-getOrCompute over option (a)
getOrCompute API change because option (a) would break ~15 test
callsites for marginal benefit. Accepted race window same as
existing preCheckHit pattern.
5. **F9 — validateProvider hints error message updated** (lib/providers/
base.mjs): pre-D33 message listed cacheable as missing and
maxSpawnTimeMs as required. Now: `'hints must be an object with
{ requiresTTY, concurrentSpawnSafe, maxConcurrent } + optional
{ maxSpawnTimeMs, cacheable }'`.
6. **F10 — Dead cache-write branch removed** (server.mjs): the
`if (hasStopChunk)` check in the streaming stop-less exhaustion
branch was unreachable (the stop-chunk completion path returns
earlier inside the for-await loop). Removed the dead code + added
a comment documenting the invariant.
**Governance/policy**
7. **F11 — Phase rolling mode policy formalized** (CLAUDE.md +
CHANGELOG.md): 22+ D-day commits accumulated under "Unreleased"
without per-D version bumps — Iron Rule 5 (release-kit bump-before-
push) appeared to be silently violated. Reality: per-D bumps would
produce 30+ noise tags during Phase 1. F11 formalizes the policy:
intra-Phase D-day commits accumulate under Unreleased; bump+tag
fires explicitly at Phase close (maintainer-triggered, not
automated). CLAUDE.md release_kit overlay gains `phase_rolling_mode`
block documenting the exception with self-pointer ("if Rule 5
appears silently violated, check this section first"). CHANGELOG
"Unreleased" gets a notice at top.
**No version bump, no git tag in D33** — policy formalization only.
8. **F12 — /v1/models created is stable per-model timestamp**
(models-registry.json + lib/providers/index.mjs + server.mjs +
docs/openai-spec-pin.md): pre-D33 used Math.floor(Date.now()/1000)
per request — violates OpenAI spec which treats `created` as
per-model attribute. Clients caching models by created would see
spurious updates on every poll.
Fixed: models-registry.json gains `bootstrapCreated: 1778630400`
top-level constant + per-model `created` fields where known
(anthropic claude-{opus,sonnet,haiku} with estimated release dates;
devstral models from "25-12" suffix; codex models pinned to
bootstrap pending verified release dates). handleModels uses
`getModelCreated(modelId)` helper from lib/providers/index.mjs.
Aliases share canonical's timestamp.
**Tests** (test-features.mjs): 401 → 414 (+13):
- F3 ×3 (same input → same id → same cache key; different name → different)
- F5 ×4 (empty/single/multi/throwing-plugin /health shapes)
- F8 ×1 (2-hop primary-fail + secondary-cache-hit → X-OLP-Cache: hit)
- F12 ×5 (stability/fallback/alias-equals-canonical)
Pre-commit fold-in (per evidence-first checkpoint #4):
- **D33 reviewer flagged F3 empty-args asymmetry** (Concern #1): hash
input used `?? ''` (empty stays) but emitted IR field used
`|| '{}'` (empty becomes '{}'). Consequence: `arguments: ''` and
`arguments: '{}'` emit identical IR but compute different ids →
different cache keys for semantically-identical requests. The exact
cache-stability bug F3 was supposed to fix.
Folded in: canonicalize empty-args to '{}' BEFORE hashing. Hash
input now matches IR emission exactly. Same line change resolves
the asymmetry.
Authority:
- ALIGNMENT.md self-amendment (F1 pin correction)
- ADR 0005 invariant "same inputs → same key, no random, no timestamp"
(F3 restoration)
- ADR 0002 § Provider contract "/health uses healthCheck" (F5)
- ADR 0004 § Observability headers (F8 X-OLP-Cache correctness)
- ADR 0005 § Cache write conditions item 1 (F10 truncation-not-cached
invariant explicit)
- Iron Rule 5 (F11 release-kit reconciliation)
- OpenAI /v1/models spec — `created` per-model stable (F12)
- CC 开发铁律 v1.6 § 10.x — Round-5 Cold Audit caught all 8
Reviewer (Iron Rule v1.6 § 10.x Mode A, fresh-context opus, independent
of drafter): APPROVE_WITH_MINOR. Verified:
- F1 plugin cross-reference (mistral.mjs:360-369) accurately documents
the rationale
- F3 collision resistance + NUL separator + restored cache invariant
- F5 all 4 cases (empty/single/multi/throwing) work
- F8 closure semantics across multi-hop chains (verified hop-fail +
fallback-hit case)
- F10 dead code removal preserves the stop-chunk completion path
- F11 phase_rolling_mode policy honest about what happened and what
the going-forward rule is
- F12 stability across consecutive /v1/models calls; alias-canonical
parity
- 414/414 tests pass
3 remaining non-blocking suggestions (F3-vs-modern-tool_calls path
canonicalization symmetry; F12 codex models explicit-vs-fallback
writeup mismatch; F8 servingHopWasCached naming) tracked as future
polish; not folded.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
||
|
taodengandClaude Opus 4.7 (noreply@anthropic.com)
|
e2e67de23a |
feat(phase-1): land IR + plugin loader + server skeleton (D3)
Phase 1 Day 1. First executable code lands. Zero providers wired yet
(per ALIGNMENT.md "v0.1 ships 0 Enabled Providers"); the server starts
clean and POST /v1/chat/completions returns 503 with no_enabled_provider.
Files added:
lib/ir/types.mjs - IR v1.0 schema + validators (ADR 0003)
lib/ir/openai-to-ir.mjs - OpenAI Chat Completions to IR
lib/ir/ir-to-openai.mjs - IR chunks to OpenAI SSE / non-stream
lib/providers/base.mjs - Provider contract + validateProvider + ProviderError
lib/providers/index.mjs - Static empty registry stub (ADR 0002)
server.mjs - HTTP listener with createOlpServer factory + main guard
test-features.mjs - 61 tests across 7 suites (IR / provider / HTTP)
Files modified:
package.json - main and scripts.start/test added back; targets now exist.
Authority citations:
IR fields and translation direction: ADR 0003 sections Decision and
Translation direction model.
Provider contract (9 fields): ADR 0002 section Provider contract v1.0
interface.
Entry surface routes (health, v1/models, v1/chat/completions): OLP v0.1
spec section 4.1 single-protocol entry; ALIGNMENT.md Authority 2.
Zero-Enabled-Providers behaviour: ALIGNMENT.md Provider Inventory.
Architectural decisions worth recording:
1. server.mjs uses a createOlpServer factory plus an import.meta.url
main guard. The factory returns an unbound http.Server; only the
main-script invocation calls .listen(). Tests import the real
server.mjs and exercise the real router. No parallel implementation
in the test file.
This pattern was a fold-in from the orchestration step. The initial
sonnet draft put a top-level server.listen call in server.mjs, which
forced test-features.mjs to reimplement the router inline (a false-
confidence trap because the real server logic would never be tested).
Refactored before reviewer dispatch.
2. lib/providers/index.mjs ships an empty STATIC_REGISTRY array, not a
placeholder with dummy entries. ALIGNMENT.md Provider Inventory says
v0.1 ships zero Enabled Providers; the registry honors that exactly.
Phase 1 Day 2 adds the first import (Anthropic) when its plugin lands.
3. BadRequestError lives in openai-to-ir.mjs and ProviderError in
base.mjs. Reviewer suggested relocating to a shared lib/errors.mjs
once the count exceeds two; deferred to Phase 1 Day 2 to ship with
the third typed error class.
4. contractVersion: '1.0' on each provider plugin: not enforced at D3
because no providers exist yet. Reviewer flagged for Phase 1 Day 2
tightening when the first provider lands.
Reviewer chain (Iron Rule 10):
Initial implementer: sonnet (general-purpose).
Refactor (createOlpServer + main guard) by the orchestrator after
catching the inline-router parallel-implementation issue.
Fresh-context reviewer: opus (ecc:code-reviewer). Verdict
APPROVE_WITH_MINOR.
Reviewer's two non-blocking findings folded in:
F1: removed unused createServer import from test-features.mjs line 12,
left over from the refactor.
F2: replaced finish_reason value 'error' with 'stop' in both the
streaming error chunk path (lib/ir/ir-to-openai.mjs line 72) and
the non-streaming error aggregation path (lib/ir/ir-to-openai.mjs
line 153). The 'error' value is not in OpenAI's documented
finish_reason enum (stop / length / tool_calls / content_filter /
function_call / null), so emitting it would violate ALIGNMENT.md
Rule 2 (b). Provider errors are now surfaced via a top-level
response.error object plus an inline content marker. The matching
test assertion at test-features.mjs line 325 was updated to verify
finish_reason stays within the OpenAI enum.
Note on the F2 fold-in:
Reviewer pointed only at the streaming path (line 72). After applying
that fix I ran grep across lib/ and test-features.mjs for the same
invention pattern and caught a second hit at line 153 (non-streaming
aggregation). This is the "fold-in must grep the full repo, not only
the file the reviewer named" discipline from
~/.cc-rules/memory/feedback/evidence_first_under_speed_pressure.md.
Both hits are fixed in this commit.
Verification:
node --check on all 7 new files plus modified package.json plus
server.mjs plus lib/ir/ir-to-openai.mjs - all clean.
npm test - 61/61 pass in 209ms, no flakes, no skipped.
OLP_PORT=14001 node server.mjs followed by curl /health returns
proper JSON; curl /v1/models returns 200 empty list; server shuts
down cleanly on signal.
grep "finish_reason.*error" returns zero hits across lib/ and tests.
Co-Authored-By: Claude Opus 4.7 (noreply@anthropic.com)
|