planr-fix

GitHub

用于执行代码修复、测试修复及文档修正的技能。针对失败测试、Bug回归或审查发现,进行根因分析并实施修复,确保范围准确和验证证据充分。

.codex/skills/planr-fix/SKILL.md regenrek/codex-planr

Trigger Scenarios

直接修复Bug 处理回归问题 修复失败的测试 根据planr-review结果改进代码 完成未结束的.planr计划阶段

Install

npx skills add regenrek/codex-planr --skill planr-fix -g -y
More Options

Non-standard path

npx skills add https://github.com/regenrek/codex-planr/tree/main/.codex/skills/planr-fix -g -y

Use without installing

npx skills use regenrek/codex-planr@planr-fix

指定 Agent (Claude Code)

npx skills add regenrek/codex-planr --skill planr-fix -a claude-code -g -y

安装 repo 全部 skill

npx skills add regenrek/codex-planr --all -g -y

预览 repo 内 skill

npx skills add regenrek/codex-planr --list

SKILL.md

Frontmatter
{
    "name": "planr-fix",
    "description": "Implement remaining scoped work in this repository to verified completion. Use for direct bug fixes, regressions, failing tests, `planr-review` findings, or unfinished `.planr` plan phases when the next step is to change code, tests, or docs, keep live `.planr` status honest, and prove the result. Not for writing a new execution contract (`planr-plan`), giving a verdict-only status answer (`planr-status`), or running a findings-first audit (`planr-review`)."
}

Planr Fix

Use this skill when the next job is to finish real implementation work, not just describe it.

A planr-fix is invalid if it claims completion without owned scope, honest live status, root-cause reasoning for bug-driven work, and exact verification evidence.

CLI-First Rule

  • Read ../planr-shared.md first.
  • Use ./.planr/tooling/planr status show, ensure-scope, set-checklist, set-blocker, and set-verification for .planr state whenever they fit.
  • There is no planr.py fix command today. Use normal repo tools for implementation, tests, and scoped verification.

Required Inputs

  • the trigger: user request, failing test, bug doc, task note, or planr-review finding
  • the governing .planr/plans/*.plan.md when one exists
  • any explicitly referenced historical source doc still in scope
  • the exact implementation files and tests for the owned scope

Core Rules

  • Convert every in-scope request, finding, or failing behavior into a real fix or an explicit blocked or unverified item.
  • For bugs, regressions, contract failures, unexpected requests, restore or hydration issues, and hidden writes, debug root-cause first:
    • state the expected behavior and invariant
    • trace the call path
    • identify the canonical source of truth and the first unintended side effect or write
  • Fix the correct owner layer, not only the downstream symptom.
  • Do not make contracts, parsers, or validation more permissive unless that behavior is proven canonical.
  • Apply the hard-cut policy by default: remove fallback, compatibility, shim, dual-path behavior, and one-value shells unless transition support was explicitly requested.
  • Checked phase items, completed plan todos, and live-status entries are claims until the owned diff and verification evidence prove them.

Required Workflow

  1. Start from the concrete trigger, not from the diff.
  2. Define the owned implementation scope and exclude unrelated dirty paths.
  3. Use the CLI to ensure or inspect the live scope entry before major edits when the command surface fits.
  4. Translate each in-scope issue into explicit checklist items with owned scope and required proof.
  5. Implement the canonical fix in the correct layer, including hard-cut cleanup where required.
  6. Update or add the smallest proof that can fail for the right reason.
  7. After each substantial step, sync checklist, blocker, and verification state through the CLI when supported.
  8. Reconcile checked plan items, completed todos, and live-status claims against the owned diff and actual verification before marking anything complete.
  9. Return completed work separately from blocked or unverified work, quoting exact commands and real results.

Output Format

Use this structure:

  • Execution scope
  • Checklist status
  • Verification evidence
  • Remaining blockers or unverified items
  • Brief summary

Rules for the response:

  • Say which findings, direct requests, or failing behaviors were closed.
  • If the work started from a bug or regression, name the root cause and the layer fixed.
  • Quote the exact verification command for each completed claim.
  • If something could not be verified, say that explicitly rather than smoothing it over.
  • Keep the summary short and factual.

Additional Resource

  • Read ../planr-shared.md first for shared CLI coverage and shared .planr rules.
  • For the full execution checklist, verification ladder, and response template, see reference.md

Version History

  • 15aeaf4 Current 2026-07-25 10:11

Same Skill Collection

.codex/skills/planr-plan/SKILL.md
.codex/skills/planr-review/SKILL.md
.codex/skills/planr-status/SKILL.md
.codex/skills/planr-summary/SKILL.md

Metadata

Files
0
Version
15aeaf4
Hash
33d8a7bd
Indexed
2026-07-25 10:11

trang chủ - Wiki
Copyright © 2011-2026 iteam. Current version is 2.155.2. UTC+08:00, 2026-09-01 03:11
浙ICP备14020137号-1 $bản đồ khách truy cập$