memory-continuity/memory/2026-03-07.md

# 2026-03-07

## Memory/Embeddings incident + external API inventory (OpenClaw)
- Issue: `memory_search` (semantic memory retrieval) failed with `429 insufficient_quota` due to embeddings provider quota exhaustion.
- Root cause: OpenClaw `agents.defaults.memorySearch.provider = openai` with `remote.apiKey` present (OpenAI embeddings key). Not related to Brave web search.

### External API / Provider inventory (from ~/.openclaw/openclaw.json)
- Chat/model routing:
  - Primary: `openai-codex/gpt-5.4`
  - Fallbacks: `openai-codex/gpt-5.2`, `github-copilot/claude-opus-4.6`
- Memory semantic search:
  - `memorySearch.enabled = true`
  - `memorySearch.provider = openai`
  - `memorySearch.remote.apiKey` present
  - `memorySearch.sources = [memory, sessions]`
- Web tools:
  - `tools.web.search.enabled = true` (Brave Search key present)
  - `tools.web.fetch.enabled = true`
- Auth profiles present: `anthropic:manual`, `github-copilot:github`, `openai:manual`, `openai-codex:default`.

## User request
- Tao requested: archive this inventory info and keep it updated whenever APIs/providers are added/removed in the future.

## Comms protocol update (task progress)
- Tao reported a recurring issue: after assigning a task, the agent may go silent (stuck or working) unless Tao explicitly asks.
- Preference selected: progress updates at **milestones** (Option B).
- Policy: acknowledge within ~30s with plan+ETA; then send updates at each milestone; if ETA slips, proactively report block/reason/next-step.

## Memory architecture preference (agents)
- Tao preference: each Agent should have **independent soul + independent memory**; do **not** share memory across agents.
- Disaster recovery preference: restore automation should be maximized; when restoring, default to **safe mode**: rename existing `~/.openclaw` before restoring (Option B).

## Clawkeeper project + gateway safety incident
- Created and published new open-source repo: https://github.com/dtzp555-max/clawkeeper
  - Purpose: OpenClaw Memory Ops Kit (doctor + embeddings probe + provider switch + backup plan + verify-backup + restore tooling).
  - Added `doctor` to perform a real OpenAI embeddings probe (minimal request) to detect 429/insufficient_quota early.
  - Added `backup-plan --json` and `verify-backup <tgz>` to ensure backups include required paths.
  - Added `restore-guide` and `restore-apply` with SAFE default: do not restart gateway unless explicitly requested.
- Incident: calling `openclaw gateway restart` during restore testing could leave LaunchAgent in a bad state (not loaded / restart fails), forcing `openclaw gateway install` to recover; can cause temporary messaging disconnects.
- New hard rule from Tao: do not harm the primary Mac gateway during development/testing; use a separate test machine.

## Test machine ("lobster" 232)
- Host: 172.16.2.232 (ubuntu-srv), user: administrator.
- SSH key: ~/.ssh/openclaw_backup_key
- Tao confirmed key was installed; verified login: `ssh -i ~/.ssh/openclaw_backup_key administrator@172.16.2.232 'echo OK'`.
- Policy: perform potentially disruptive OpenClaw install/restart tests on 232 instead of the primary Mac.

## Multi-agent dev workflow preference
- Tao clarified the intended architecture for complex development:
  - main = PM / architect / coordinator / reviewer / reporter.
  - execution agents do the hands-on implementation.
- Dedicated execution agent for Codex work:
  - Name: `codex_worker`
  - Workspace: `~/.openclaw/workspaces/codex_worker`
  - Agent dir: `~/.openclaw/agents/codex_worker/agent`
  - Model target: `openai-codex/gpt-5.4`
- Preference: `codex_worker` should avoid automatic fallback; if GPT-5.4 is unavailable, report to Tao for model decision.
- Preference: execution agents should get the strongest execution model; main should remain strong for PM/summary work, but not necessarily do the coding itself.
- New delegation rule clarified: unless work is truly tiny and can be finished in one short pass, main should avoid personally editing/coding and should delegate implementation to `codex_worker` first.
- New agent-splitting rule clarified: create multiple execution agents when work is parallelizable, responsibilities differ, contexts would otherwise get mixed, or separate validation/release tracks are needed; avoid splitting tiny or tightly coupled work.
- Long-lived execution agents initialized locally: `docs_worker`, `qa_worker`, `ops_worker` (with dedicated agent dirs + workspaces + basic role files). `codex_worker` remains the default general execution worker.
- This architecture is the current default, but Tao may revise it as needs change.

## OCM / workflow / agent architecture updates (late evening)
- OCM work completed today:
  - Rewrote README opening around first-run value, support positioning, and product framing.
  - Refreshed README hero screenshots using newer redacted Dashboard / Agents / CLI captures.
  - Adjusted README screenshot layout to use one large Dashboard hero image plus a secondary row (Agents / CLI / Actions) because mixed aspect ratios looked awkward in a single row.
  - Unified OCM built-in CLI completion text to English:
    - success: `✅ Done (exit 0)`
    - non-zero exit: `❌ Exit code N`
- OCM GitHub flow today confirmed again that the repo is effectively PR-only even when approvals are set to 0:
  - PR #5: README positioning + screenshot refresh
  - PR #6: version bump to `v0.9.1`
  - PR #7: English CLI completion messages
  - PR #8: README screenshot layout A
  - Release published: `v0.9.1`
- Tao explicitly said that for status updates, accuracy matters more than token saving; concise is fine, but not at the cost of ambiguity.
- Tao approved a multi-agent dev architecture as the default pattern:
  - main = PM / architect / reviewer / user-facing coordinator
  - execution agents = hands-on implementers
  - use `codex_worker` as the default coding execution agent
  - create more execution agents when work is parallelizable, responsibilities differ, contexts would mix, or separate validation/release tracks are needed
  - avoid splitting tiny or tightly coupled work
- Tao approved a stronger delegation rule:
  - unless work is truly tiny and can be completed in one short pass, main should avoid personally implementing/editing and should delegate execution to `codex_worker` first
- Long-lived execution agents initialized locally for repeat use:
  - `docs_worker`
  - `qa_worker`
  - `ops_worker`
  - `codex_worker` remains the default general execution worker
- Two workflow/architecture skills were created and published to GitHub:
  - `gh-pr-release-flow` → https://github.com/dtzp555-max/gh-pr-release-flow
  - `execution-agent-planner` → https://github.com/dtzp555-max/execution-agent-planner
- Tao's preferred lifecycle rule for execution agents was clarified:
  - keep long-lived role-based agents for reuse
  - delete or archive project-specific temporary agents after the project ends to avoid an “agent graveyard”
- Next improvement approved: add a standardized execution-agent dispatch / handoff protocol so workers receive clearer task packets and report blockers/results consistently.
- New hard reporting rule approved: when a worker reports acceptance, milestone, blocker, or completion, main must update Tao before continuing other downstream work. A worker completion that is not forwarded counts as a main-process failure, not “still in progress”.