analyse-issue

📁 tenfyzhong/skills-hub 📅 3 days ago
4
总安装量
2
周安装量
#51323
全站排名
安装命令
npx skills add https://github.com/tenfyzhong/skills-hub --skill analyse-issue

Agent 安装分布

cursor 2
claude-code 2

Skill 文档

Analyse Issue

Overview

Fetch the issue details, verify repo alignment, and inspect the local codebase to confirm or refute the reported problem. Provide an evidence-backed analysis with clear next steps.

Workflow

1) Parse the issue input

  • If input is a GitHub issue URL, extract owner/repo and issue number.
  • If input is just a number, assume the current repo unless the user specifies another repo.
  • If the provider is not GitHub or is unclear, ask for clarification before proceeding.

2) Verify local repo matches the issue

  • Ensure the current directory is inside a git repo (git rev-parse --show-toplevel).
  • Identify the target remote (prefer origin); normalize SSH/HTTPS to owner/repo.
  • If the issue URL repo does not match the local owner/repo, stop and ask the user to switch directories or confirm the target repo.
  • If no remote or multiple candidates exist, ask the user which repo to use.

3) Fetch issue content

Prefer gh when available:

  • gh issue view <num> --json title,body,labels,comments,author,createdAt,updatedAt
  • If URL provided, run against that repo: gh issue view <num> -R owner/repo --json ...

Fallbacks:

  • gh api repos/{owner}/{repo}/issues/{num} and .../comments if you need more fields.
  • If gh is unavailable but GITHUB_TOKEN exists, use curl with the GitHub API.
  • If neither works, ask the user to paste the issue content.

Capture at least: title, body, labels, environment details, repro steps, expected/actual behavior, and key comment insights.

4) Analyze the codebase

  • Translate the issue into concrete signals (keywords, error messages, stack traces, config names).
  • Use rg to locate relevant code, tests, and configs.
  • Trace the execution path: entry points -> core logic -> dependencies.
  • Identify likely failure points: missing checks, edge cases, incorrect assumptions, data shape mismatches, concurrency/timing issues.
  • If appropriate and safe, run focused tests; otherwise propose targeted tests to validate the hypothesis.

5) Confirm or qualify the issue

  • Provide evidence with file references and reasoning.
  • If you can only reason without running tests, state assumptions and confidence.
  • If evidence is insufficient, list the exact missing info needed.

6) Respond with a structured analysis

Include:

  • Issue summary (expected vs actual)
  • Repo match verification
  • Evidence (files/functions)
  • Root-cause hypothesis or confirmed cause
  • Suggested fix approach
  • Questions or missing data
  • Next steps (tests, logs, repro)

Output conventions

  • Respond in the user’s language when clear; default to Chinese for Chinese prompts.
  • Keep analysis concise but include concrete file pointers and evidence.
  • Do not claim confirmation without code-based evidence or reproduction.