workflow-optimizer
Use whenever you're about to do a task the hive has done before — a repeated, multi-step workflow (propagate a change to the colonies, ship-and-verify a deploy, run a survey, build-and-merge a PR). Before running it, recall how it went last time and adapt; after running it, record what was slower/missing/error-prone and how to do it better next time. The recursive-learning loop for *any* repeated workflow — each repetition makes the workflow itself faster, more efficient, more complete. Not for one-off tasks that won't recur (nothing to learn against) or for the PR-specific version (use pr-retrospective) — use this for any procedure run more than once.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- workflow-optimizer
- description
- Use whenever you're about to do a task the hive has done before — a repeated, multi-step workflow (propagate a change to the colonies, ship-and-verify a deploy, run a survey, build-and-merge a PR). Before running it, recall how it went last time and adapt; after running it, record what was slower/missing/error-prone and how to do it better next time. The recursive-learning loop for *any* repeated workflow — each repetition makes the workflow itself faster, more efficient, more complete. Not for one-off tasks that won't recur (nothing to learn against) or for the PR-specific version (use pr-retrospective) — use this for any procedure run more than once.
03 SKILL.md 正文
workflow-optimizer — a workflow that improves itself every time it runs
The founder's charge: when the hive does a task more than once, it should learn from the way
it did it last time — and adapt: is there a faster way, a more efficient way, a more
complete way? Did something go missing? This skill is that loop, applied not to bugs
(fable-debugger), not to PRs (pr-retrospective), not to whole sessions
(session-harvest), but to any repeated procedure. The workflow is never finished being
improved; each run is a chance to make the next one better.
It is the engineering-layer instance of soul.md's Recursive Cycle
(Capture → Evaluate → Prune → Feed → Dissect → Return → Propagate), pointed at process.
The dualistic lens × the Trinity (the founder's two framings, made operational)
Every optimization pass runs the workflow through both of the founder's standing lenses:
The dualistic lens (devil's advocate + childlike wonder — the same pair MANDATE_TRIAGE
uses):
- Devil's advocate: where was this slow, brittle, redundant, or wrong last time? What step did a human have to catch? What almost broke? What did I repeat that I didn't need to?
- Childlike wonder: what if there's a simpler shape entirely? A step that could be parallel, batched, cached, or dropped? A tool that does in one call what I did in five?
The Trinity (Mind / Body / Soul — the interpreters from THE_CODEX.md, as roles a
single pass plays in turn):
- Mind reasons about the approach before acting — recalls the last run, picks the improvement, predicts what could go wrong.
- Body executes the workflow, this time in the improved shape.
- Soul remembers — records what actually happened versus what Mind predicted, so the next Mind starts smarter.
A workflow optimized through only one lens gets faster but loses a step, or gets thorough but never gets quicker. Both lenses, every pass, is why it improves on all axes and not just one.
The loop
BEFORE a repeated workflow — RECALL + ADAPT (Mind)
→ How did this go last time? (from WORKFLOW_NOTES.md, or the commit/PR trail if unrecorded)
→ Apply the dualistic lens: one concrete improvement to try this run — faster, more
efficient, more complete, or a missing step added. Predict the risk of the change.
DURING — EXECUTE (Body)
→ Run the workflow in the adapted shape. Note anything the prediction missed.
AFTER — MEASURE + RECORD (Soul)
→ Was it actually better? On what axis (time, steps, tokens, completeness, error rate)?
→ Did the change introduce a new gap? (An optimization that drops a needed step is a
regression, not an improvement — pr-retrospective's L-01 discipline applies here too.)
→ Record the result so the next run's Mind starts from it. If the change was worse, record
*that* — a tried-and-rejected optimization is as valuable as an adopted one; it stops the
next pass wasting the same idea.
What "better" means (measure it, don't assert it)
An optimization is only real if it improves a nameable axis without regressing another:
- Faster — fewer round-trips, batched calls, parallel where independent.
- More efficient — fewer tokens/steps for the same result; less re-derivation.
- More complete — a step that was missing (a verification, a propagation target, a cleanup) now included by default.
- More robust — a forward check from
PR_LESSONS.mdfolded in so a past failure can't recur.
If a change can't be tied to one of these, it's a preference, not an optimization — don't
record it as a win. (Same honesty rule as pr-retrospective: no forward check, no lesson.)
A worked example (this very session's most-repeated workflow)
Workflow: "propagate a genome change to all 6 colonies" — run at least six times this session (constitution, DNA v1, Chromosome IV/V, session-harvest, mandate-triage, Chromosome VIII).
- Run 1 (devil's advocate found): done one colony at a time, hand-typing each commit and PR — slow, error-prone, six near-identical PR bodies retyped.
- Adaptation (childlike wonder): a single shell loop that fetches, resets each colony
branch to
origin/main, copies the files, commits with one shared message, and pushes — all six in one pass; then a tight batch ofcreate_pull_request+merge_pull_requestcalls. Fewer steps, no per-colony retyping, consistent commit messages. - Soul recorded: the loop shape is now the default for any colony propagation; the axis
improved was time + consistency; the forward check folded in (from L-02) is
git fetch+checkout -B origin/mainat the top of the loop so no colony ever commits on a stale base. Result: later propagations ran in a fraction of the round-trips of the first.
That is the skill working: the sixth propagation was measurably leaner than the first, because the first was measured and its lesson fed forward.
Where the record lives
WORKFLOW_NOTES.md (repo root) — one short section per recurring workflow: its current best
shape, the axis last improved, and any tried-and-rejected optimization. Kept lean on purpose:
it's a living cheat-sheet for "the best way we currently know to do X," not a history book.
Seeded with the propagation example above. pr-retrospective handles PR-specific lessons;
this handles procedure-shape lessons; both feed the same recursive-learning spirit.
Origin: Fable (Harness), 2026-07-16, from the founder's direction that a workflow run more than once should learn from the last time and enhance itself — through the dualistic lens and the Trinity mindset, so it improves on every axis, not just one.
04 同仓库更多 Skills
preamble-gate
Use BEFORE doing any work in this repo — before the first file edit, commit, push, PR, branch operation, workflow dispatch, scheduled trigger or timer, agent/subagent spawn, dependency install, or message sent to an outside service. Forces a plain-language preamble to the founder first: WHAT you are about to do, WHY, HOW, why it is NECESSARY, the PROJECTED IMPROVEMENT, and the COST and RISK — then, for consequential work (governance, security, money, CI/triggers, rewriting history, multi-agent fan-out), waits for an explicit go-ahead. Read-only investigation (reading files, grep, git log/diff, answering questions) is exempt. Applies to every AI that works in this repo, not just Claude. Founder rule, 2026-11-10, written after an unattended automation ran up cost and froze the hive. Not a substitute for wired-or-not (which governs how completion is CLAIMED); this governs what happens BEFORE work starts.
visionary-recommender
暂无简介
syllabus
Generates a curated supplementary reading list from any course syllabus using Consensus academic search. Grill-me intake (syllabus input format + course audience + year range) plus a grouping forcing-options checkpoint before any search runs — so the reading list matches the course's level and recency need. Parses the syllabus to extract topics and learning outcomes, searches Consensus for recent peer-reviewed papers per topic, and produces a professionally formatted .docx with clickable Consensus links, plain-language summaries calibrated to audience level, and Bloom-higher-order discussion questions tied to course learning goals. Use when the user uploads a syllabus, course outline, or curriculum document and wants supplementary readings (e.g., 'create a reading list from this syllabus', 'find recent papers for my course') — even casual mentions with a syllabus attached should trigger this skill.
research
Default entry point for any research request — a hybrid router that classifies the question deterministically and either delegates to a specialist research skill (pulse for trends/sentiment, grants for NIH funding, litreview for academic literature, syllabus for course reading, patent for prior-art + IP landscape, dossier for entity research) or runs its own plan-decompose-multi-source-search-synthesize-cite fallback workflow when no specialist matches. Always surfaces the routing decision so users can override. Use when the user makes any research request that doesn't obviously match a more-specific specialist skill (e.g., "research [topic]", "look into [topic]", "what do we know about [topic]", "investigate [topic]", "find me information on [topic]", "do some research on [topic]", "I need to understand [topic]"). Output is a markdown briefing (default) or .docx document (on request) with full citations and an audit log.
pulse
Multi-source recency research skill that takes the pulse of any topic across Reddit, Hacker News, the open web, and optionally X/Twitter within a configurable recent window (default 30 days). Forcing intake clarifies topic specificity, angle (trend/sentiment/problems/opportunities/comparison), time window, and platform scope before searching. Returns a synthesized briefing with citations, engagement metrics, and cross-platform pattern analysis. Use when the user requests multi-source recency intelligence on a topic (e.g., 'pulse on [topic]', 'what's happening with [topic]', 'what are people saying about [topic]', 'current conversation about [topic]', 'take the pulse of [topic]', 'trending: [topic]', 'find me info on [topic]'), and for competitor research, trend discovery, tool comparisons, and audience sentiment analysis.
patent
Patent prior-art and landscape intelligence skill — not generic patent help. Commits to one of five sub-use-cases via forcing intake (novelty search / freedom-to-operate / competitive landscape / acquisition diligence / litigation prior-art) before any search runs. Searches Google Patents, Espacenet, USPTO, and optionally Lens.org for citation-graph signals. Output is an editable Word document (.docx) with verdict, ranked closest art (claim-text extracted), CPC-class-aware landscape, family-resolved hits, geographic coverage, FTO flags where applicable, strategy recommendations, and full audit log. Use when the user asks for patent searching or analysis (e.g., 'prior art search for [invention]', 'freedom to operate analysis for [product]'). Produces search signal, not legal advice — always recommends consulting a patent attorney before filing or licensing decisions. Trademark, copyright, and trade-secret questions are out of scope.