feat(extensions): bench-reflective-dreaming + claude-code-bridge by LightDriverCS · Pull Request #1 · BenchAGI/openclaw

LightDriverCS · 2026-04-18T22:22:41Z

Summary

Adds two BenchAGI-owned extensions to our fork that together unify OpenClaw's native memory-core.dreaming with a structured reflective + fleet-canon layer, synced to the Firebase agent wiki via our own cloud-mirror daemon.

`extensions/bench-reflective-dreaming/` (new)

Complements — does not replace — memory-core's existing sleep phases. Where memory-core is mechanical (score-based MEMORY.md promotion) and poetic (dreaming-narrative writes dream-diary prose), this adds the ops-log dimension: per-agent structured reflection + a coordinator-written fleet canon + a nightly consolidation pass.

02:30-02:46  per-agent structured reflection (5 cron jobs)
03:00        memory-core's existing sweep (light → REM → deep + narrative)
03:15        fleet canon (coordinator synthesizes across agents)
03:40        memory consolidation (prune duplicates, stage risky patches)

Shape (lightweight — prompts + defaults + scripts, not a full TS plugin yet):

assets/prompts/{dreaming,fleet-canon,memory-consolidation}-protocol.md — version-controlled prompt templates.
defaults/cron-schedule.json — agent list, minute offsets, coordinator agent.
scripts/install.mjs — idempotent reconciler. Adds/updates/prunes 7 managed cron jobs to match defaults; migrates un-tagged orphan jobs (from manual bootstrap) by name.
scripts/uninstall.mjs — removes managed cron jobs; preserves vault protocols.
openclaw.plugin.json — manifest with config schema. Runtime is `"scripts"` today; upgrade path to a full TS plugin later.

`extensions/claude-code-bridge/` (was untracked)

Already present in working trees but never committed. Now version-controlled:

cloud-mirror.mjs — the daemon that mirrors ~/.openclaw/wiki/main/* → https://benchagi.com/api/v1/wiki/ingest (X-API-Key auth). This is how vault writes reach the Firebase Agent Wiki.
ai.openclaw.wiki-mirror.plist.template — launchd template.
serve.mjs, statusline.mjs, mirror.mjs — supporting files already in use.

Verification

Ran scripts/install.sh tonight against Cory's local gateway:

3 protocol files installed into ~/.openclaw/wiki/main/concepts/ (content-hashed — re-runs report `0 written, 3 unchanged`)
7 cron jobs reconciled with `[managed-by=bench-reflective-dreaming]` tag
Idempotency verified: re-running reports `0 added, 0 updated, 0 removed`
First dream ran on `main` at ~21:37 UTC: produced a real 5.4 KB reflection (access-gate patterns, crew-as-relay-for-cofounders, dream logs as cross-session continuity mechanism), mirrored to Firebase in ~30 s, visible in the BenchAGI agent-wiki admin

How it fits with upstream

No upstream code modified. This PR only adds files under extensions/.
Runs alongside memory-core.dreaming. Both dreaming systems coexist: our structured reflection at 02:30–02:46, upstream memory-core at 03:00 for promotion + narrative prose, our canon at 03:15 synthesizing across both.
Safe to merge upstream updates. Fork-sync task will continue to work; no conflicts expected.

Test plan

bash extensions/bench-reflective-dreaming/scripts/install.sh on a fresh machine → expect 3 protocols written + 7 crons added.
Run a second time immediately → expect `0 added, 0 updated, 0 removed`.
Edit defaults/cron-schedule.json (change a minute offset) → re-run install → expect 1 reconciled.
bash extensions/bench-reflective-dreaming/scripts/uninstall.sh → expect 7 cron jobs removed, protocols preserved.
Tonight 02:30–03:40 MDT: observe 5 dream files + 1 canon entry + 1 consolidation audit appear under ~/.openclaw/wiki/main/.

Follow-ups

Upgrade to a full TS plugin (index.ts using definePluginEntry) once the API surface stabilizes — would enable runtime config reconciliation without re-running install.sh.
Consider upstreaming a "reflection" phase alongside memory-core.dreaming's light/REM/deep once pattern is proven.

🤖 Generated with Claude Code

Adds two BenchAGI-owned extensions to the fork that together unify OpenClaw's native dreaming with a structured reflective + fleet-canon layer, synced to the Firebase agent wiki via our cloud-mirror daemon. extensions/bench-reflective-dreaming/ Lightweight extension (prompts + defaults + install scripts, no TS runtime yet) that runs alongside memory-core.dreaming. Where memory-core is mechanical (score-based MEMORY.md promotion) and poetic (dreaming- narrative diary prose), this adds the ops-log dimension: 02:30-02:46 per-agent structured reflection (5 cron jobs) 03:00 memory-core's existing sweep (light/REM/deep + narrative) 03:15 fleet canon (coordinator synthesizes across agents) 03:40 memory consolidation (prune duplicates, stage risky) - assets/prompts/{dreaming,fleet-canon,memory-consolidation}-protocol.md version-controlled prompts; install.mjs copies into vault concepts/ (content-hashed, so re-runs are cheap) - defaults/cron-schedule.json: agent list + minute offsets + timezone - scripts/install.mjs: idempotent reconciler. Adds/updates/prunes the 7 managed cron jobs to match defaults; migrates untagged orphan jobs into managed ones by matching on name - scripts/uninstall.mjs: removes managed jobs; preserves vault protocols - openclaw.plugin.json: manifest with config schema; runtime=scripts today, upgrade path to a full TS plugin later extensions/claude-code-bridge/ Was already present in the working tree but untracked. Now version- controlled. cloud-mirror.mjs is the daemon that pushes ~/.openclaw/wiki/main deltas to benchagi.com/api/v1/wiki/ingest. Paired with the launchd plist template. Also tracked: serve.mjs (MCP bridge) and statusline.mjs. Verified tonight: first dream on `main` produced a real 5.4 KB reflection that mirrored to Firebase in ~30s. install.mjs is idempotent — re-running reports 0 changes once reconciled. Coexists with memory-core.dreaming cleanly; no upstream code modified. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Commits the bench-reflective-dreaming bundler stub (untracked since 2026-04-20). Should have shipped with #1 / #4 but was missed. Codex Anvil (gpt-5.5 xhigh) verdict: REPAIR. Initial bare `export const PLUGIN_RUNTIME` satisfied tsdown but failed the OpenClaw plugin loader contract (`src/plugins/loader.ts` rejects modules without `register`/`activate`). Repair: replaced with no-op `definePluginEntry`. Verified pnpm build emits dist/extensions/bench-reflective-dreaming/index.js + loader returns status: "loaded". 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Two small Codex W3 findings: #1 — Null-body guard. A JSON body of `null` (or a primitive/array) would skip the LlmTurnWireRequest cast and throw on field access, producing a 500 from the dispatch error handler instead of a clean 400. Added top-level object guard at the start of validateLlmTurnRequest that returns invalid_field('<root>', 'request body must be a JSON object'). Renamed local var `body` → `wire` for the rest of the function so TypeScript sees the narrowed type. #2 — Fractional max_tokens. Original `Number.isFinite(x) && x > 0` let `max_tokens: 0.5` pass and floor to a 0-token budget. Now require `Number.isInteger(x) && x > 0`. Reason text updated to 'must be a positive integer'. Tests: 14 → 16 (added fractional + non-object body cases). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

P1 #1 — thinking.budget_tokens < max_tokens constraint. Anthropic rejects requests where thinking.budget_tokens >= max_tokens; my budget map (high=16384, xhigh=32768) exceeded typical max_tokens=8192 which would 400 immediately. Fix: cap budget to (max_tokens - 1024) reserving room for actual output. If the cap drops below Anthropic's minimum thinking budget (1024), disable thinking entirely rather than send an invalid request. P1 #2 — ANTHROPIC_AUTH_TOKEN env var. Anthropic SDK reads ANTHROPIC_AUTH_TOKEN for OAuth tokens and ANTHROPIC_API_KEY for API keys. My resolver only checked ANTHROPIC_API_KEY, so OAuth-only setups got 'no_anthropic_credential'. Fix: read both, with ANTHROPIC_AUTH_TOKEN taking precedence (matches SDK behavior). API_KEY-shaped OAuth tokens (sk-ant-oat*) still detected and routed via authToken. P2 (deferred) — `as never` casts. Codex flagged that imports of SDK param types would catch shape bugs (e.g. Tool.InputSchema requires `type: "object"`). Validation layer already enforces shape, but upgrading to proper SDK types is a follow-up — not landing this commit to keep the patch focused. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* feat(gateway): W3 scaffold for /v1/llm_turn cloud-brain endpoint Phase 1B W3 of BenchAGI ADR-0002. Route registration + body validation + auth check for the cloud-brain dispatch target. Anthropic call is TODO (returns 501 so relay claim path can land first). Also: minor lint-fix in extensions/claude-code-bridge/serve.mjs to unblock the repo-wide pre-commit hook (oxlint curly-rule on a pre- existing single-line return). Out of W3 scope but required for commit. Spec: ~/.openclaw/wiki/main/_boards/specs/phase-1a-design-gate-2026-05-05-v2.md §7 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(gateway-w3): address Codex GO-WITH-PATCHES feedback Two small Codex W3 findings: #1 — Null-body guard. A JSON body of `null` (or a primitive/array) would skip the LlmTurnWireRequest cast and throw on field access, producing a 500 from the dispatch error handler instead of a clean 400. Added top-level object guard at the start of validateLlmTurnRequest that returns invalid_field('<root>', 'request body must be a JSON object'). Renamed local var `body` → `wire` for the rest of the function so TypeScript sees the narrowed type. #2 — Fractional max_tokens. Original `Number.isFinite(x) && x > 0` let `max_tokens: 0.5` pass and floor to a 0-token budget. Now require `Number.isInteger(x) && x > 0`. Reason text updated to 'must be a positive integer'. Tests: 14 → 16 (added fractional + non-object body cases). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(gateway-w3): wire real Anthropic call (env-credential MVP) Replaces the 501 stub with an actual Anthropic SDK call. Resolves the credential from process.env.ANTHROPIC_API_KEY, distinguishing OAuth tokens (sk-ant-oat*) from API keys, and surfaces the resolved profile name back as 'used_auth_profile' so the cloud orchestrator (W2) can decide between OAuth-mode (no aiUsageRecords) and API/coin-mode (write aiUsageRecords) per spec §6. NEW IN llm-turn-http.ts: - callAnthropicForLlmTurn(request) — translates camelCase request to Anthropic SDK params. Handles cache_control on system prompt (ephemeral), thinking_level → thinking config with conservative budget_tokens (low=4096, medium=8192, high=16384, xhigh=32768). Calls messages.create, returns wire-format response (snake_case). - resolveAnthropicCredential() — env-based credential resolution. Reports 'env-api-key' or 'env-oauth-token' as the profile name. - LlmTurnWireResponse type — explicit snake_case shape per spec §7 line 741. ERROR PATHS: - 'no_anthropic_credential' (500) — env not set - 'anthropic_auth_failed' (401/403) — credential rejected - 'anthropic_call_failed' (502 default) — SDK error EXPLICIT NEXT-ITERATION work captured inline: - Real auth-profile-machinery integration (loads the agent's configured profile via OpenClaw's auth-profile store, handles OAuth refresh, etc.). The env approach gets Cory's local OpenClaw to a working smoke test path; production-grade auth-profile resolution is the follow-up. - Idempotency-key write-ahead store (lease-recovery semantics per spec §7 line 851). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(gateway-w3): address Codex Anthropic-call P1 feedback (2 fixes) P1 #1 — thinking.budget_tokens < max_tokens constraint. Anthropic rejects requests where thinking.budget_tokens >= max_tokens; my budget map (high=16384, xhigh=32768) exceeded typical max_tokens=8192 which would 400 immediately. Fix: cap budget to (max_tokens - 1024) reserving room for actual output. If the cap drops below Anthropic's minimum thinking budget (1024), disable thinking entirely rather than send an invalid request. P1 #2 — ANTHROPIC_AUTH_TOKEN env var. Anthropic SDK reads ANTHROPIC_AUTH_TOKEN for OAuth tokens and ANTHROPIC_API_KEY for API keys. My resolver only checked ANTHROPIC_API_KEY, so OAuth-only setups got 'no_anthropic_credential'. Fix: read both, with ANTHROPIC_AUTH_TOKEN taking precedence (matches SDK behavior). API_KEY-shaped OAuth tokens (sk-ant-oat*) still detected and routed via authToken. P2 (deferred) — `as never` casts. Codex flagged that imports of SDK param types would catch shape bugs (e.g. Tool.InputSchema requires `type: "object"`). Validation layer already enforces shape, but upgrading to proper SDK types is a follow-up — not landing this commit to keep the patch focused. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Cory Shelton <coryshelton@Corys-Mac-mini.local> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

LightDriverCS merged commit e598b4e into main Apr 18, 2026
2 checks passed

LightDriverCS mentioned this pull request Apr 18, 2026

fix(memory-core): safe statistical fallback when dreaming narrative can't run #2

Merged

LightDriverCS mentioned this pull request Apr 28, 2026

chore(dreaming): commit bench-reflective-dreaming bundler stub #15

Merged

1 task

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(extensions): bench-reflective-dreaming + claude-code-bridge#1

feat(extensions): bench-reflective-dreaming + claude-code-bridge#1
LightDriverCS merged 1 commit into
mainfrom
feat/bench-reflective-dreaming

LightDriverCS commented Apr 18, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

LightDriverCS commented Apr 18, 2026

Summary

extensions/bench-reflective-dreaming/ (new)

extensions/claude-code-bridge/ (was untracked)

Verification

How it fits with upstream

Test plan

Follow-ups

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

`extensions/bench-reflective-dreaming/` (new)

`extensions/claude-code-bridge/` (was untracked)