AGENT SKILL

fix

Write a current-feature.md spec for an ad-hoc bug fix or small unplanned change, then stop before implementation. Use for /fix, confirmed bugs, or small changes that do not belong in the build plan.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 510
Forks
132

01 前言 Front Matter

name
fix
description
Write a current-feature.md spec for an ad-hoc bug fix or small unplanned change, then stop before implementation. Use for /fix, confirmed bugs, or small changes that do not belong in the build plan.
disable-model-invocation
true

03 SKILL.md 正文

fix - document an ad-hoc fix, then build it like anything else

Context reuse: Reuse any required file already loaded in project instructions or the current session. Read it again only if absent, changed, or exact current bytes or line references are needed.

First action: Before project inspection, preflight, or any other tool call, publish running to blueprint/.state/run.json when blueprint/.state/ exists. From the project root, run the first helper that exists, node .agents/skills/blueprint-doctor/scripts/run-state.mjs or node .claude/skills/blueprint-doctor/scripts/run-state.mjs, with start --command <skill> --summary "<truthful summary>" --boundary <read-only|reviewed|local-only>. Use update at milestones, with --status blocked --resume <command> for a blocker, and finish --status ready|completed at the end; --help lists every option and blueprint/workflow.md has the full activity contract. A missing or failing helper is a warning, not a blocker.

Where this sits in the workflow:

/fix  ->  /implement  ->  /complete  ->  back to your features
(spec     (build it,      (log to blueprint/history/fixes/
 the fix)  reviewed)       + merge)

A fix is a bug or small change that isn't a planned build-plan feature. It runs through the same loop as a feature (build with review gates, iterate, then merge); it just starts here instead of /feature, and is logged separately.

Input

A description of the bug or change, for example /fix "password reset email never sends". If the user just reported the problem in chat, use that.

The input may also be a finding ID from blueprint/context/findings.md, alone or with a description, for example /fix F-03. Pull the problem statement from that ledger entry. Use this form only between work items, when current-feature.md is the reset stub: this skill overwrites that file, so while a spec is active, repair its findings through /implement instead. Never replace another worker's active spec. Parallel work uses one clone or Git worktree per work item, with a separate branch in each checkout.

Step 1 - write the fix spec

Pull context from blueprint/context/project-overview.md and blueprint/context/coding-standards.md, then write a short spec to blueprint/context/current-feature.md (this file holds whatever is being built now, feature or fix). Its first heading must be exactly # Fix: <title>, and it must retain the **Type:** Fix contract below. Keep it lighter than a feature spec:

  • Title - the bug or change in a few words.
  • Type: Fix (so /complete logs it to blueprint/history/fixes/, not blueprint/history/features/).
  • Status: not started - /implement updates this durable workflow state as work and verification progress.
  • Branch: the full fix branch from the configured prefix plus the fix title in lowercase kebab-case.
  • Fixes: <finding id> - only when the fix targets a ledger finding. The stamp makes the repair traceable: /implement marks that finding fixed when the repairing step lands, and /audit re-reviews it before it closes.
  • The problem - what's wrong or what needs to change, and where.
  • The fix - the root-cause repair, and anything it must not break. Do not add an abstraction, dependency, service, configuration surface, compatibility layer, or security mechanism unless the defect is in that layer or an established repository requirement needs it now. Adding a missing authorization, ownership, validation, escaping, or redaction check is a root-cause repair, not new machinery.
  • Build steps - usually one small step; split only if the diff would be too big to read. Each ends with an observable "done when".
  • Verify - how to confirm it's fixed (what to click or test).

Then stop. Tell the user to review the fix spec, then run /implement to build it.

Rules

  • A fix is not a build-plan item; don't add it to build-plan.md.
  • Removing a completed feature is a fix. Keep its original feature archive. The plans stay user-owned: to build it again later, the user unchecks its item; if it is abandoned, they edit the plans and rerun /overview.
  • Keep it small. If it's really a new feature, use /feature and the build plan instead.
  • Same conventions as everything else (blueprint/context/coding-standards.md).

Formatting

Format the output to match the project's conventions in blueprint/context/ai-interaction.md: concise, scannable markdown, with lists for enumerations and tables for matrices rather than dense paragraphs.

04 同仓库更多 Skills