mirror of
https://github.com/EveryInc/compound-engineering-plugin.git
synced 2026-10-07 23:49:12 +02:00
Trevin Chow
58164ee99e
`auto_babysit: false` was named as the standing opt-out in the babysit handoff but never read there. The only config read lived in the Step 4 reference, scoped to the teaching-gate keys, so the Step 5 gate depended on remembering an earlier step. Compaction destroys exactly that, which is why the reporter saw Codex babysit PRs they had opted out of. The resolution now happens at the gate, and an opted-out handoff is stated as a successful terminal — without that, the gate's own "stop and report it blocked" language turned a deliberate choice into a failure. A per-run `babysit:` token now decides the run whatever the config says, which also covers `continuous`/`checkpoint` under a standing opt-out. `lfg`'s in-pipeline babysit is bounded and stays out of scope; that is now stated at the step and in the configuration reference rather than reading as an oversight. The contract test asserted the literal string `auto_babysit: false`, which prose that never read anything satisfied. It now pins the read mechanism and the successful-terminal framing. Fixes #1601
Skill Documentation
End-user-facing documentation for compound-engineering plugin skills. Each page covers the skill's high-level purpose, novel mechanics, use cases, and chain position relative to other skills.
For runtime behavior and contributor reference, the SKILL.md in each skill's source folder under skills/ is authoritative.
Checkout-local defaults shared across skills are documented in Compound Engineering configuration.
Artifact paths shown throughout these pages (docs/plans/, docs/solutions/, docs/ideation/, and the rest) are the defaults. A project can relocate every CE artifact folder under one repo-relative root with docs_root; when it is set, read the shown paths as <your-docs_root>/plans/, <your-docs_root>/solutions/, and so on. See Artifact root.
The compound-engineering core loop
[/ce-ideate] (optional) "What's worth exploring?"
│
▼
┌─→ /ce-brainstorm "What does this need to be?"
│ │
│ ▼
│ /ce-plan "What's needed to accomplish this?"
│ │
│ ▼
│ /ce-work "Build it."
│ │
│ ▼
└── /ce-compound "Capture what we learned."
/ce-compound is the closer that makes the loop compound: it writes learnings into docs/solutions/, which the next iteration's /ce-brainstorm and /ce-plan read as grounding. That return arrow is the whole point. /ce-ideate is an optional prelude for when you don't yet know what to work on. Everything else in this catalog is either an anchor around the loop or an on-demand tool used when a specific need arises, not a step you walk through every time.
The Core Loop
The steps of every engineering iteration. /ce-ideate runs only when you need to find a direction first; the other four run in order per piece of work.
| Skill | Description |
|---|---|
/ce-ideate |
Optional first step: discover grounded directions worth exploring (six frames, tagged basis, adversarial cut) |
/ce-brainstorm |
Define what something should become: one question at a time, ask only decisions, named gap lenses, requirements-only unified plan |
/ce-plan |
Bound execution with guardrails (U-IDs, test scenarios, automatic confidence check). WHAT decisions, not HOW code |
/ce-work |
Execute an implementation-ready plan: figure out the HOW with code in front of you, then ship through quality gates |
/ce-compound |
Close the loop by writing what you learned into docs/solutions/ so the next iteration can read it |
Around the Loop
Skills that anchor, feed, or maintain the loop without being steps inside it.
| Skill | Description |
|---|---|
/ce-strategy |
Create or maintain STRATEGY.md, the upstream anchor ce-ideate, ce-brainstorm, and ce-plan read as grounding |
/ce-product-pulse |
Outer observation loop: a time-windowed report on usage, performance, errors, and follow-ups, saved to docs/pulse-reports/ |
/ce-sweep |
Recurring feedback sweep: ingest Slack/GitHub items (email experimental), acknowledge at source, and keep an /lfg-ready rolling plan |
/ce-compound-refresh |
Maintain docs/solutions/ over time (Keep / Update / Consolidate / Replace / Delete), Interactive or Autofix |
On-Demand
Invoked when a specific need arises, not part of any chain.
| Skill | Description |
|---|---|
/ce-pov |
A project-grounded verdict: adopt/hold/reject, a document take, or a position on supplied approaches. Optional named/oracle panel. |
/ce-explain |
A durable teaching document for a concept, a diff, an idea, or a window of recent work. Optional opt-in check-in. |
/ce-prototype |
Build a throwaway prototype so someone can experience how the product should work, feel, or read, then write those decisions into an existing plan or continue into brainstorm or plan |
/ce-debug |
Find the root cause of broken behavior: causal chain, predictions, then an optional fix and PR handoff |
/ce-code-review |
Structured review of a diff or PR: skill-local personas, confidence-gated findings, and the rules you write in CODING_STANDARDS.md |
/ce-doc-review |
Structured review of a requirements or plan document: findings, not a holistic verdict |
/ce-simplify-code |
Refine recently changed code for reuse, quality, and efficiency, with behavior preserved |
/ce-optimize |
Metric-driven optimization loops with parallel experiments and a durable experiment log |
/ce-retune |
Retune a skill corpus for a new model: baseline, noise floor, then measured cut passes |
Research & Context
| Skill | Description |
|---|---|
/ce-riffrec-feedback-analysis |
Turn a Riffrec recording into structured feedback: a quick bug in chat, or an extensive analysis that hands off to ce-brainstorm |
Git Workflow
| Skill | Description |
|---|---|
/ce-commit |
Local git commit(s) only: convention-aware, named-file staging, file-level splits (up to three). No push. |
/ce-commit-push-pr |
Working changes to an open PR. Three modes: full ship, rewrite an existing description, or description-only from a URL. |
/ce-babysit-pr |
Watch an open PR over time: incoming review via /ce-resolve-pr-feedback, CI via /ce-debug. Does not merge under target or stack-ready; stack-land can merge a confirmed managed stack. |
/ce-worktree |
Isolate work in a git worktree: detect existing isolation, prefer the host's native tool, else plain git |
Autonomous Pipeline
| Skill | Description |
|---|---|
/lfg |
Hands-off pipeline through an open PR (plan, implement, review, ship, bounded CI watch). Pushes without prompting when a remote exists; local commits only otherwise. Does not merge. |
Frontend Design
| Skill | Description |
|---|---|
/ce-polish |
Conversational UX polish on a feature that already works: start the dev server, open a browser, iterate. Manual invoke only. |
Collaboration
| Skill | Description |
|---|---|
/ce-proof |
Publish, view, comment on, or pull markdown via Proof. One-way publish; not a review skill. |
Workflow Utilities
| Skill | Description |
|---|---|
/ce-promote |
Draft announcement copy for a shipped feature (X, changelog, LinkedIn, email, blog, demo). Drafts only; never posts. |
/ce-resolve-pr-feedback |
One pass to evaluate, fix, and reply to PR review comments, including nitpicks. Babysit is the watch that calls this. |
/ce-dogfood |
Hands-off browser QA of the branch: map flows, fix small breakages, write a report. Manual invoke only. |
/ce-test-browser |
End-to-end browser tests of the current diff using a host-native browser with agent-browser fallback. Does not check out a PR or branch. |
/ce-test-xcode |
Build and test an iOS app on the simulator (screenshots, logs, human verification). Not XCUITest. |
/ce-setup |
Diagnose optional tool capabilities and create or repair repo config.yaml |
/ce-handoff |
Write a session handoff, or find and orient from a selected source. Does not auto-continue. |
See also
For the top-level install and usage guide, see README.md. Each skill's authoritative runtime spec is in skills/<skill>/SKILL.md.