Agent Skills › openinary/openinary › issue-fixer

issue-fixer

GitHub

用于在开源仓库中协作实现 GitHub Issue。流程包括同步代码、分析问题与方案、获取人工审批后创建分支,并进行增量开发与原子提交,最终生成可验证的 PR。

.github/skills/issue-fixer/SKILL.md openinary/openinary

Trigger Scenarios

需要按 Issue 逐步实现功能 需要人工审批技术方案后编码 需要在开源项目中协作开发

Install

npx skills add openinary/openinary --skill issue-fixer -g -y
More Options

Non-standard path

npx skills add https://github.com/openinary/openinary/tree/main/.github/skills/issue-fixer -g -y

Use without installing

npx skills use openinary/openinary@issue-fixer

指定 Agent (Claude Code)

npx skills add openinary/openinary --skill issue-fixer -a claude-code -g -y

安装 repo 全部 skill

npx skills add openinary/openinary --all -g -y

预览 repo 内 skill

npx skills add openinary/openinary --list

SKILL.md

Frontmatter
{
    "name": "issue-fixer",
    "description": "Implement GitHub issues collaboratively in open-source repositories with explicit approval gates, incremental delivery, atomic commits, and PR-ready validation. Use when working issue-by-issue with a human operator who decides scope and product direction.",
    "argument-hint": "Issue URL or description, target branch policy, and any non-negotiable constraints",
    "user-invocable": true,
    "disable-model-invocation": false
}

Open Source Issue Implementation

Outcome

Produce a production-ready issue implementation with:

  • Shared understanding of scope and constraints
  • Operator-approved technical approach before coding
  • Incremental, reviewable changes
  • Clean commit history
  • Verified build/lint/type/tests status
  • PR-ready summary linked to the issue

When to Use

Use this skill when:

  • Working on a single GitHub issue in an open-source repository
  • A human operator must validate decisions before implementation advances
  • Scope control and traceable collaboration matter more than speed

Do not use this skill when:

  • Doing broad discovery without a concrete issue
  • Running large autonomous refactors without operator checkpoints

Collaboration Contract

At all times:

  • Propose -> operator decides -> implement
  • Do not make product decisions alone
  • Do not expand scope without explicit approval
  • Prefer clarity over speed
  • Ask concise clarification questions when ambiguous

Required Inputs

Collect or confirm:

  • Issue link or full issue text
  • Target base branch and contribution conventions
  • Definition of done (tests, lint/type, docs, screenshots, changelog)
  • Constraints (backward compatibility, performance, API stability)

Procedure

1. Repository Sync

Run and report results:

  1. git checkout main
  2. git pull origin main
  3. git fetch --all
  4. git status

Validation checks:

  • Working tree is clean (or operator explicitly approves proceeding with local changes)
  • main is current
  • No unresolved conflicts

If validation fails:

  • Stop and ask the operator how to proceed before editing code

2. Issue Analysis (No Coding)

Deliver a concise analysis containing:

  • 2-3 sentence summary
  • Goal
  • In-scope items
  • Out-of-scope items
  • Constraints
  • Edge cases
  • Open questions

Decision gate:

  • Wait for operator validation before moving forward

3. Solution Proposal (No Coding)

Propose implementation details:

  • Technical approach
  • Impacted files/modules
  • Data flow (if relevant)
  • Risks and trade-offs
  • Test strategy

Decision gate:

  • Wait for explicit operator approval before coding

4. Branch Creation

Create a branch only after proposal approval:

  1. Choose one prefix: feat/, fix/, refactor/, or chore/
  2. Create branch: git checkout -b <prefix><short-description>

5. Implementation Loop (Incremental)

For each iteration:

  1. Explain the next small step
  2. Implement that step
  3. Report what changed and why
  4. Ask for operator validation

Suggested iteration order:

  • Setup/refactor prerequisites
  • Core functionality
  • Edge cases and error handling
  • Cleanup and documentation

Rules:

  • Keep diffs focused and reviewable
  • Avoid unrelated file churn
  • Halt when requirements become ambiguous

6. Atomic Commits

Commit each logical unit separately using conventional commits, for example:

  • feat: add overlay support
  • fix: handle empty queue payload

Commit quality checks:

  • One logical change per commit
  • Message reflects actual change
  • No debug artifacts

7. Validation Before PR

Run or confirm:

  • Build succeeds
  • Lint/type checks pass
  • Tests pass (or clearly state missing tests and rationale)
  • No debug code, temporary logging, or dead code remains

If checks fail:

  • Fix failures or escalate with a concise blocker report

8. Sync Before Push

Before opening PR:

  1. git fetch origin
  2. git rebase origin/main

If conflicts appear:

  • Resolve cleanly
  • Re-run validation checks
  • Summarize conflict decisions for operator visibility

9. Pull Request Preparation

Create a structured PR description containing:

  • Problem summary
  • Solution summary
  • Key implementation details
  • Validation evidence (build/lint/type/tests)
  • Visuals/screenshots when relevant
  • Linked issue

10. Post-Merge Cleanup

After merge confirmation:

  1. git branch -d <branch>
  2. git push origin --delete <branch>

Completion Criteria

This skill is complete when:

  • Issue scope is implemented as approved
  • Operator accepted each decision gate
  • Validation checks are green (or exceptions documented and accepted)
  • PR is ready or merged with cleanup completed

Message Format (Recommended)

For each operator-facing update, use:

  1. Intent
  2. Action taken
  3. Result
  4. Decision needed

Keep updates concise and structured.

Version History

  • fdcff9e Current 2026-08-20 09:59

Same Skill Collection

.agents/skills/loops-api/SKILL.md
.agents/skills/loops-cli/SKILL.md
.agents/skills/loops-email-sending-best-practices/SKILL.md
.agents/skills/loops-lmx/SKILL.md

Metadata

Files
0
Version
1f05325
Hash
146de45a
Indexed
2026-08-20 09:59

trang chủ - Wiki
Copyright © 2011-2026 iteam. Current version is 2.155.2. UTC+08:00, 2026-10-25 05:10
浙ICP备14020137号-1