Git Bisect Assistant Agent Skill

Git Bisect Assistant

Automates the git bisect process to efficiently identify the first bad commit responsible for a bug or test failure.

Quick Start

Basic usage pattern:

python scripts/git_bisect_runner.py \
  --good <known-good-commit> \
  --bad <known-bad-commit> \
  --test "<test-command>"

Example:

python scripts/git_bisect_runner.py \
  --good v1.0.0 \
  --bad HEAD \
  --test "pytest tests/test_feature.py::test_specific_case"

Workflow

• Gather Information
• Identify the known good revision (commit, tag, or branch)
• Identify the known bad revision (defaults to HEAD)
• Determine the test command that fails on bad commits and passes on good commits

• Run Bisect
• Execute the git_bisect_runner.py script with appropriate parameters
• The script will automatically test commits and narrow down the culprit

• Review Results
• Examine the identified bad commit
• Review the bisect log showing all tested commits
• Check confidence level and assumptions

Parameters

Required

• --good: Known good revision (commit hash, tag, or branch name)
• --test: Shell command to test each commit. Exit code 0 = good, non-zero = bad

Optional

• --bad: Known bad revision (default: HEAD)
• --repo: Repository path (default: current directory)
• --retries: Number of test runs per commit for flaky tests (default: 1)
• --timeout: Test execution timeout in seconds (default: no timeout)

Handling Flaky Tests

For non-deterministic tests, use --retries to run the test multiple times per commit:

python scripts/git_bisect_runner.py \
  --good abc123 \
  --bad HEAD \
  --test "npm test" \
  --retries 3

The script uses majority voting: if a test passes 2 out of 3 times, the commit is marked as good.

Test Command Guidelines

The test command should:

• Exit with code 0 for good commits (test passes)
• Exit with non-zero code for bad commits (test fails)
• Be deterministic or use --retries for flaky tests
• Complete within reasonable time or use --timeout

Examples:

# Python test
--test "pytest tests/test_auth.py -v"

# Shell script
--test "./scripts/verify_build.sh"

# Compilation check
--test "make && ./bin/app --version"

# Multiple commands
--test "npm install && npm test"

Output Report

The script generates a comprehensive report including:

• First Bad Commit: Hash and commit message of the culprit
• Confidence Level: Assessment based on test stability and retry logic
• Assumptions: Any assumptions made during bisect (retries, timeouts)
• Tested Commits: Complete list of all commits tested with results
• Bisect Log: Detailed log of the bisect process

Common Scenarios

Scenario 1: Test Started Failing

User: "The integration tests started failing sometime in the last 20 commits"

python scripts/git_bisect_runner.py \
  --good HEAD~20 \
  --bad HEAD \
  --test "pytest tests/integration/"

Scenario 2: Feature Broke After Release

User: "Feature X worked in v2.1.0 but is broken now"

python scripts/git_bisect_runner.py \
  --good v2.1.0 \
  --bad HEAD \
  --test "python -c 'import app; assert app.feature_x() == expected'"

Scenario 3: Flaky Test Investigation

User: "A test fails intermittently, need to find when it started"

python scripts/git_bisect_runner.py \
  --good main \
  --bad feature-branch \
  --test "pytest tests/test_flaky.py" \
  --retries 5 \
  --timeout 30

Tips

• Ensure clean state: Commit or stash changes before running bisect
• Fast tests: Use focused tests rather than full test suites for faster bisect
• Build requirements: Include build steps in test command if needed
• Dependencies: Ensure test command handles dependency installation if needed across commits
• Timeout wisely: Set timeout slightly longer than expected test duration
• Retry count: Use 3-5 retries for flaky tests to get reliable results

Troubleshooting

Bisect fails to start: Verify good and bad revisions exist and are valid git references

Test command fails unexpectedly: Test the command manually on a known good/bad commit first

Inconsistent results: Increase --retries or check for environmental factors affecting tests

Timeout too short: Increase --timeout or optimize test command