frame
Surface the open questions before planning a multi-branch feature, and shape the feature plan mode drafts. Use before entering plan mode for work that obviously won't fit in one branch and one pull request -- never for a change that already does.
- License 许可
- BSD-3-Clause
- 所属仓库
- urban233/Canon
- 最近更新
- 2026-10-01
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- frame
- description
- Surface the open questions before planning a multi-branch feature, and shape the feature plan mode drafts. Use before entering plan mode for work that obviously won't fit in one branch and one pull request -- never for a change that already does.
03 SKILL.md 正文
Frame
This runs once, before any branch exists for the feature. Framing decides what's worth building and what shape it takes; plan mode decides how one piece gets done. The branch is the line between them.
Ask only what's still open
A competent engineer would want to know: who has this problem, what changes for them, what does success look like, what's explicitly out of scope, and the shape at feature altitude -- the data model, the API surface, not implementation detail. If the opening request already answers some of these, don't re-ask them. Cover whatever's still open in one combined question, not one per item.
What the draft should contain
Once the developer approves the feature plan drafted in /plan mode,
write it yourself to .canon/plans/features/<slug>.md (the slug comes
from the plan's own # title -- Codex has no ExitPlanMode tool to
name a file for you, so a PostToolUse hook, normalize_plan.py,
derives this file's header from what you write, the same way the plan
skill's hook does for a branch plan; see that skill for the underlying
mechanism). Only write it once the developer has actually approved it --
this platform has no tool call that confirms that for you, so treat the
write itself as the consequential step. Make sure the plan states:
-
# <Feature name>as its title -- this becomes the saved plan's filename. -
## Why,## Success, and## Non-goals-- the same disciplineplanasks for at branch altitude, one level up. -
## Shape-- the technical approach -- and## Decisions, only where something was genuinely contested (an alternative seriously considered, not just the obviously correct way of doing it). -
## Steps-- an ordered list, each entry one branch-sized piece: one branch, one pull request. A non-empty## Stepssection is what makesnormalize_plan.pytreat the file as a feature plan rather than an ordinary branch plan.Lead each step with its branch slug, then a colon:
## Steps - slugs: make duplicate slugs raise, with a regression test - permalinks-api: expose the permalink endpointThat slug is how
canon_positionanswers "which step am I on" -- it matches the slug against branches that exist and pull requests that merged. A branch may carry a prefix (feature/slugsmatches the stepslugs), but a step written without a slug can never be matched to anything, andcanon_positionwill report it asunmatchedrather than guess. Matching step prose against branch names is exactly the wrong-but-plausible inference Canon refuses to make.Say what a step waits for where it isn't the step above it. No annotation means it waits for the one above -- what an ordered list already implies -- so write
(after: ...)only where that's wrong:## Steps - eval-corpus: 30 graded requests with their expected outcomes - pymol-tools (after: none): expose the library to an agent - baseline (after: eval-corpus, pymol-tools): record the score(after: none)waits for nothing; a comma-separated list waits for all of them.canon_positionreports every step whose dependencies have merged as startable, so this is what lets two sessions pick up different steps without colliding.Claim independence only where it's true. A step wrongly marked independent gets started against a base without its prerequisite in it -- a wasted branch at best, a silent conflict at worst.
No gate, no new artifact
Nothing here is accepted or advanced through a status. The feature plan
is a document the developer reads, edits, and reorders like any other
file in the repo -- there's no separate design doc, no brief, no launch
plan. When work starts on step one, use the plan skill as usual; it
already asks for the parent feature plan when one exists.
04 同仓库更多 Skills
testing-craft
Provides Google-derived testing strategy, test-writing craft, and test-suite health reference, distilled from Software Engineering at Google (the Flamingo book), the Google Testing Blog, and Google's published testing-infrastructure research. Canon's `reviewer` subagent reads the writing-craft and strategy references as review criteria. Invoke directly to design a test strategy, audit an existing suite's health, or triage a flaky or brittle test.
ship
Decide whether a change is ready to hand to a human, and open the pull request when it is. Use once the work is done and verified, before opening a PR or telling the developer the change is ready.
review
Dispatch Canon's independent reviewer subagents for this branch and read their verdicts back from canon_review. Use once a change is built and verified, before shipping. This skill only dispatches and reads back -- the review itself runs inside the subagent, in a fresh context, and never in this session.
review-change
暂无简介
plan
Shape what plan mode drafts, so the plan Canon saves needs no follow-up. Use whenever drafting a plan for approval in plan mode, before writing it to .canon/plans/.
decide
Draft and commit a decision record for a design or architecture choice made in this session, once it has survived implementation. Use at ship time, right before opening the pull request, when the branch made a decision worth recording.