## What Problem This Solves
Skill Workshop experience review ran as an isolated agent with a re-rendered 60k-character transcript, its own bootstrap, and a trimmed tool surface. Every review was a cold request: no prompt-cache reuse, a large re-serialized trajectory, and a reasoning/tool profile that differed from the foreground turn. Autonomous updates could target any workspace skill, size limits allowed 40 KB skills to grow unchecked (one deployed workspace hit a 21 KB `SKILL.md`), the collection reviewer had to re-list every skill in its plan and read them under a fixed 24k-character budget (a 16-skill workspace failed every pass with "Read every current skill before reconciling"), collection review only recorded successes so a failing pass retried on every Gateway restart, and `openclaw skills curator status` showed nothing about what the last reviews did.
## Why This Change Was Made
- **Warm fork.** Experience review now continues the finished turn from the same in-memory session prefix (same session identity, bootstrap, skills prompt, tool schemas, `promptCacheKey`), appends one short review message, and runs with `sessionPersistence: "detached"` so nothing it writes reaches the foreground transcript or session record. Only `skill_workshop` executes; every other tool keeps its schema for cache parity and fails with a next-step message. The gate wraps the core tool list before Code Mode / Tool Search compaction, so catalog-hidden tools stay gated while `tool_call`/`exec` remain callable. `src/agents/embedded-agent-runner/run/attempt.skills-policy.test.ts` asserts identical system-prompt and tool digests between a foreground turn and its review, and that `tool_call` reaches `skill_workshop` but not `read` during review. Code Mode swarm globals (`phase`, `log`, `agents.run`) are the `sessions_spawn` capability and gate on the same allowlist, so a review cannot emit foreground lifecycle events or launch collectors. The review shares the foreground session, so it no longer retires that session's MCP runtime on run end; it reuses the warm runtime like any turn. Accepted tradeoff: the review inherits foreground tool construction (it creates the session MCP runtime only when the session has none, and spawns per-attempt LSP servers only when configured), because splitting tool construction from the foreground turn would break the cache-parity invariant this PR exists for.
- **One apply owner, rechecked at the write boundary.** `src/skills/workshop/autonomous-apply.ts` decides apply-vs-pending for both the post-review loop and foreground repair: creates and updates to Workshop-authored skills apply; updates to user-authored skills stay pending with a recorded reason. `applySkillProposalTransition` rechecks path-based ownership under the commit lock for non-operator actors (agent, or no actor), so a claim released after the pre-check cannot let an agent write a user-authored skill; gateway and CLI operators still approve any pending proposal. The user-authored pending write runs under the same commit lock and rereads the row, so an operator apply that lands first is kept. The old `auto-apply.ts` path and name-based `listWorkshopAuthoredSkillNames` are deleted.
- **Sparse collection plan and model-relative reads.** The collection reviewer returns only `write` and `drop` entries; unlisted skills stay untouched, so a 16-skill workspace no longer has to restate 16 keeps. The skill read budget is derived from the model's context window (35%) instead of a fixed 24k characters, with the 10,000-character skill cap still enforced per skill. No windowed read parameters: a skill is read whole or not at all.
- **Weekly cadence.** Collection review runs once every 7 days instead of daily (`REVIEW_INTERVAL_MS`), matching how slowly a skill library actually changes; the attempt is recorded before the model starts, so a failed pass does not retry on every restart.
- **Lean cap.** `AUTONOMOUS_SKILL_MAX_CHARS = 10,000` in `src/skills/workshop/collection-contracts.ts` is enforced by the tool and by collection reconcile; an oversized skill may only shrink. Tool description, experience prompt, and collection prompt were rewritten around procedures-not-records, one trigger per branch, and "NOTHING_TO_LEARN is the correct answer for most turns".
- **Detached runs end quietly.** A detached run writes no transcript or session record and runs under the foreground session key, so `attempt-finalize` now skips agent_end side effects for it: plugins do not observe the review as a foreground turn, and a deep review cannot schedule a successor review of itself.
- **Recorded outcomes.** Experience review records `applied | proposed | nothing | failed` with usage; one attempt per turn, drop on error. `openclaw skills curator status` prints the last collection and experience review outcome.
## User Impact
- Default `auto` mode: experience review reuses the foreground prompt cache and skips message/other tool execution, so review cost is one bounded continuation instead of a cold re-render.
- Autonomous edits touch only Workshop-authored skills; handwritten skills surface as pending proposals for operator approval.
- Autonomous `SKILL.md` results stay ≤ 10,000 characters.
- Collection review runs weekly and succeeds on larger skill libraries. A workspace with no recorded attempt reviews at the first daily check after Gateway start; an upgraded install keeps its recorded last attempt and reviews once it is older than 7 days. The 10,000-character cap applies to the next autonomous write; existing oversized skills are only ever shrunk.
- `openclaw skills curator status` shows the last collection and experience review outcome, time, and tokens.
- Docs: https://docs.openclaw.ai/tools/self-learning, https://docs.openclaw.ai/tools/skill-workshop
Related: #123866, #128871 both patch skills above the reviewer read budget; this PR caps autonomous skill size and restricts oversized skills to shrink-only rewrites.
## Evidence
- Live gateway (Linux, Telegram, `xai` provider) at `133ffe3`: manual experience review on a real foreground turn → `nothing` (usage: 0 uncached input, 13,902 cached, 36 output — the forked review hit the foreground prompt cache); manual collection review over the workspace → `succeeded`; the three oversized skills were rewritten under the cap (21,002 → 9,449; 11,882 → 7,735; 10,300 → 7,413 bytes, originals in `skill-workshop/collection-backups/`), the rest untouched, and the pre-PR "Read every current skill before reconciling" failure is gone. The next foreground turn's automatic experience review recorded `nothing` and `openclaw skills curator status` shows both outcomes. The apply recheck, MCP-runtime change (`26e821f`), and detached agent_end skip (`7e4a506`) landed after that run and are covered by the regression tests below.
- `pnpm test src/skills/workshop src/agents/tools/skill-workshop-tool src/gateway/server-methods/skills` plus `src/agents/embedded-agent-runner/run/attempt.skills-policy.test.ts`, `src/agents/embedded-agent-runner/run/attempt.tool-search-catalog-abort.test.ts`, `src/cli/skills-cli.curator.test.ts`, `src/agents/harness/tool-surface-bridge.test.ts` — green. New regression tests (`service.test.ts` agent-vs-operator apply on a user-authored skill and operator apply kept over a stale pending snapshot; `experience-review.apply.test.ts` no `cleanupBundleMcpOnRunEnd`; `attempt-phase-lifecycle.test.ts` no agent_end for a detached run; `code-mode-swarm.test.ts` swarm globals refused under the review allowlist) fail on the pre-fix code.
- Dependent sweep: 46 test files importing the touched modules — green.
- `oxfmt`, `scripts/run-oxlint.mjs` on changed files, `git diff --check` — clean.
- `pnpm tsgo && pnpm check:test-types` on Blacksmith Testbox — clean.
- Local ClawSweeper review (`gpt-5.6-terra`, high): `133ffe3` raised two findings (ownership recheck removed from the apply boundary; review retiring the shared session's MCP runtime), fixed in `26e821f`; `26e821f` raised one (detached review re-entering agent-end scheduling), fixed in `7e4a506`; `7e4a506` raised three: pending write racing an operator apply and Code Mode swarm globals bypassing the execution gate, both fixed in `e4b4322`; "prove detached review does not start configured MCP or LSP runtimes" is skipped as the cache-parity tradeoff stated above (LSP tool schemas come from the LSP runtime; the MCP runtime is session-owned and already warm). Maintainer decision on `e4b4322`: accepted — a detached review reuses the session MCP runtime and re-spawns configured per-attempt LSP servers exactly like a foreground turn; it still cannot execute them.
- Unrelated CI on `26e821f`/`7e4a506`: `check-lint-core-2` (`max-lines` in `src/gateway/server-methods/models-list-result.ts`, unused param in `models.test.ts`, both from #129332) and `checks-node-compact-small-8` (`doctor-auth.profile-health.test.ts`, Claude CLI auth from #129052) fail identically on `main` run 32857954734; `checks-node-compact-small-31` pins plugin SDK export counts (`4340` vs `4342`) that #129052 moved — this PR touches no `src/plugin-sdk` file. On `e4b4322` the failing set is the same twelve `checks-node-compact` shards that fail on `main` run 32857954734 (`large-5/12/13/14/15/18/22`, `small-8/14/20/21/26`) plus `small-31`; `check-lint-core-2` passes here. None are fixed here; #129357 carries the `main` fix.
- LOC (raw numstat): production +712 / −664 (net +48; the four review-fix commits after the live run add +76 / −26); tests + docs +1269 / −1238 (net +31).
Skill Workshop could autonomously rewrite or delete handwritten skills and discarded its own drop rationale. Workshop mutations are now gated on ledger-derived ownership (applied create proposals; unknown provenance fails closed to user-owned read-only), ownership claims end at drop and return on restore with failure-safe ordering, legacy backups stay restorable, every reconcile outcome persists with drop reasons in an additive table, and a bounded skill_workshop history action exposes them. Adopt/disown opt-in: #125711.
Co-authored with @jalehman (original ownership design).
Fixes#125652
Replace divergent per-skill curation with one daily isolated collection review that can consolidate, rewrite, create, and drop writable skills atomically.
Autonomous auto mode remains automatic. Adds bounded review, transactional recovery, durable cadence, shared locking, and reuse-first /learn behavior.
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
* feat(skills): repair used skills in turn
* fix(skills): bind repair to used skill receipt
* fix(skills): preserve explicit repair review
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
Teach the semantic reviewer to improve skills the agent actually used, keep review input provider-bound, and preserve bounded deterministic retries.
Co-authored-by: Ayaan Zaidi <hi@obviy.us>
Self-learning reviewer edits existing skills via targeted patches: it quotes the exact live text (or appends a section) and the service composes the full body inside the receipt-pinned read that hash-binds the proposal — untouched content survives by construction, and patches auto-apply through the scanner-gated pipeline. Full-body rewrites and oversized-skill edits stay pending for the operator. The review prompt shifts to active capture within the existing evidence gates, and shallow same-sender turns accumulate per session (provider-identity scoped, zero-iteration and duplicate-run contracts honored, aborted provenance carried, bounded state) so quick corrections get reviewed with their own transcripts. Replaces closed#119856.
Deletes the deterministic regex capture path that templated raw chat text into skill proposals (junk like a proposal whose whole procedure was one slugified user message). All autonomous learning now flows through the isolated experience reviewer: it sees a bounded workspace skill list, prefers revising pending proposals or updating the governing skill over creating new ones, and treats durable user corrections as first-class evidence. Update proposals are reviewer-only (explicit opt-in) and never auto-apply, since the reviewer drafts them without the live skill body. Removes the producerless pending-suggestion session machinery. Regression test proves the junk path is gone; real-Telegram E2E verdict in the PR body.
Skill Workshop lifecycle actions now run without an additional Gateway approval by default, while explicit `approvalPolicy: "pending"` keeps the operator approval gate.
Prepared head SHA: 06e907797e
Co-authored-by: Shakker <165377636+shakkernerd@users.noreply.github.com>
Reviewed-by: @shakkernerd
Surfaces skills.workshop.autonomous.enabled in the Control UI Workshop tab:
a Self-learning header toggle plus an enable pitch card on the empty
proposal board. Reads the include-resolved config snapshot, patches the
canonical key over the existing runtimeConfig RFC 7396 merge-patch seam,
and surfaces patch failures in the workshop error banner. Copy states the
token cost and that drafts arrive as pending proposals (no unconditional
review guarantee, which would be false under approvalPolicy "auto").
Splits workshop empty states, header controls, and the self-learning
module out of the oversized page/view files. Locale bundles synced
(fallbacks=0 across 20 locales).
Closes#106054
Tracks per-skill usage from the skill.used diagnostic event (trusted-only
delivery, file-scoped identity), sweeps workshop-created skills daily from
gateway maintenance (active -> stale 30d -> archived 90d, pinned bypass,
restore-only unarchive, files never touched), filters archived skills from
snapshots fail-open, reports workspace-scoped overlap candidates, and adds
openclaw skills curator CLI, additive gateway methods, and a warn-only
doctor finding. Zero new config keys; SQLite/Kysely state only.
Reactive corrections ("that's not what I asked", "stop doing X") now
count as durable signals: expanded extraction patterns, vocabulary
routing to existing workspace skills, per-skill grouping. Both capture
modes share one invariant: a bounded signal-fingerprint ring on the
session entry prevents replaying applied/rejected corrections, pending
autocapture-owned proposals are revised instead of skipped, /learn-style
turns suppress duplicate agent-end capture, and extraction runs before
any skill discovery. Autonomy off keeps the suggest-tier offer; autonomy
on files/revises proposals directly.
Co-authored-by: Peter Steinberger <steipete@gmail.com>
Adds the middle tier between capture-off and autonomous capture: when
autonomy is disabled, detected durable-instruction signals record a
one-shot pendingSkillSuggestion on the session entry (signal-hash
fingerprint prevents transcript-history replay), and the next
non-heartbeat turn atomically consumes it and injects one bounded
user-role line offering to save the skill. The agent offers, the user
decides; skill_workshop approval flow unchanged; no new config.
* feat(skills): diagnose skill_workshop hidden by tool policy (#87570)
Workshop can be enabled and auto-capturing while tools.profile hides the
skill_workshop tool; every inspection surface looked healthy. plugins
inspect and openclaw doctor now name the excluding policy layer (global/
agent/provider profile, allowlist, denylist) and the exact alsoAllow
grant to add, via a shared resolveSkillWorkshopToolPolicyAvailability
helper that /learn's guard now reuses instead of composing policy
itself. Diagnosis only; no policy behavior change.
* ci: retrigger
Approval wait now fits inside the Codex dynamic-tool watchdog (70s +10s
gateway grace under the 90s kill), approval cards carry proposal id,
skill name, description, file count, and body size (spoof-safe
rendering), and timeouts return a structured pending-not-failed outcome
instead of a bare error. Expired requests cannot execute late; no
auto-apply; generic plugin approvals unchanged.
* feat(commands): add /learn to draft skills from recent work (#100408)
/learn rewrites the turn into a standards-guided Skill Workshop authoring
instruction: the agent gathers named sources (or distills the current
conversation) and files ONE pending skill proposal via skill_workshop.
Approval flow unchanged; sandboxed/tool-restricted agents get a clear
unavailable reply. Extracts the harness OpenClaw-tools predicate into
shared helpers and reserves the command name against plugin shadowing.
* docs: regenerate docs map for /learn section
Ayaan Zaidi