bug-reproduction-brief — GitHub Copilot skill for bug reproduction
Turns a vague or intermittent bug report into a minimal reproduction with evidence. Use it when the failure depends on the environment or is hard to repeat.
npx skills add github/awesome-copilot --skill bug-reproduction-briefWhat bug-reproduction-brief does
The skill records the observed error, environment, and the difference between expected and actual results. It then trims unrelated data and steps until the smallest failing scenario remains. For intermittent failures, it asks for observed frequency and duration instead of calling the result deterministic.
The brief includes the target and commit, environment, reproduction steps, minimal fixture, commands, and outputs. The workflow stops once the reproduction is verified; implementation changes belong to a separate diagnosis and repair process. Keep secrets and personal data out of logs, and use read-only or reversible checks first. Install it when you need evidence before a fix; it is not the right workflow when the task is to edit the implementation immediately.
What it helps with
- reduce an intermittent failure to a minimal reproduction
- separate observed behavior from a suspected cause
- record the environment and commit linked to a bug
- check repeatability with commands and captured output
- gather evidence before changing implementation code
Example prompts for the agent
- Use the Bug Reproduction Brief skill on the failing checkout test. Find the smallest safe reproduction and do not fix it yet.
- Compare the expected and actual test results, then record the exact commands and output.
- Create a Bug Reproduction Brief for this report and label any unverified details.
How to install bug-reproduction-brief
1Skills CLI (Vercel)
npx skills add github/awesome-copilot --skill bug-reproduction-brief2GitHub CLI
gh skill install github/awesome-copilot bug-reproduction-brief3Manually: put the files in place
.github/skills/bug-reproduction-brief/SKILL.md1Skills CLI (Vercel)
npx skills add github/awesome-copilot --skill bug-reproduction-brief -a claude-code2Manually: put the files in place
~/.claude/skills/bug-reproduction-brief/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/bug-reproduction-brief/
Manually: put the files in place
.agents/skills/bug-reproduction-brief/SKILL.mdFor all projects use ~/.agents/skills/
Manually: put the files in place
.cursor/skills/bug-reproduction-brief/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
Platform CLI
gemini skills install https://github.com/github/awesome-copilot --path skills/bug-reproduction-briefManually: put the files in place
.devin/skills/bug-reproduction-brief/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
From the same set554
The whole set →Similar
Questions about bug-reproduction-brief
How do I install bug-reproduction-brief in GitHub Copilot?
Run npx skills add github/awesome-copilot --skill bug-reproduction-brief or gh skill install github/awesome-copilot bug-reproduction-brief.
What does bug-reproduction-brief do?
It helps build a minimal, evidence-backed reproduction of a bug before diagnosis or repair begins.
Should I change implementation code while building the brief?
No. The workflow stops after verifying the reproduction because code changes can affect the evidence and mix reproduction with repair.
Source text
SKILL.md · MIT
--- name: bug-reproduction-brief description: 'Turn a vague, intermittent, or environment-specific bug report into a minimal evidence-backed reproduction before proposing a fix.' --- # Bug Reproduction Brief Use this skill when a bug report is incomplete, intermittent, environment-specific, or mixed with an assumed cause. The goal is to prove the smallest observable failure before diagnosis or repair begins. ## 1. Record the observed failure Capture the exact error, incorrect output, timestamp, affected route or command, and the smallest known input. Preserve relevant logs without secrets or personal data. Label second-hand descriptions as unverified. ## 2. Identify the environment Record only facts you can inspect: - repository and commit; - runtime and package-manager versions; - operating system or container; - dependency lockfile; - relevant feature flags; - whether the target is local, test, staging, or production. Never guess credentials or production configuration. ## 3. Separate expected from actual behaviour Write two explicit observable statements: ```text Expected: [observable result] Actual: [observable result, including status or error] ``` Do not put the suspected cause in either statement. ## 4. Reduce the reproduction Start from the reported path, then remove unrelated data, services, and steps one at a time. Keep the smallest fixture that still fails. If the failure stops, restore the last removed condition and record it. Prefer an isolated test, minimal script, or smallest safe request over reproducing against production. ## 5. Prove repeatability Run the minimal reproduction at least twice where safe. Record commands and outputs. If the failure is intermittent, report the observed frequency and duration instead of calling it deterministic. ## 6. Stop before repair A verified reproduction is the deliverable. Do not edit implementation code while building the brief because that can destroy the evidence or mix diagnosis with remediation. ## Output ```markdown # Bug Reproduction Brief - Target and commit: - Environment: - Expected: - Actual: - Minimal steps: - Minimal fixture: - Reproduced: yes / no / intermittent - Evidence: - Unknowns: - Safe next hypothesis to test: ``` ## Safety boundaries - Do not change production data merely to reproduce a bug. - Do not publish secrets, customer records, or private source. - Do not claim a root cause from correlation alone. - Use read-only or reversible discovery first. - Stop after a verified reproduction; diagnosis and repair are separate workflows. ## Example prompt ```text Use the Bug Reproduction Brief skill on the failing checkout test. Do not fix it yet. Reduce it to the smallest safe failing fixture and report the exact command evidence, expected result, actual result, and remaining unknowns. ``` ## Source and licence Adapted from the MIT-licensed workflow at https://github.com/skyestrela/ai-agent-skill-preview.
© github · MIT · github/awesome-copilot