- Read README.md and follow CONTRIBUTING.md.
- Read documentation relevant to the component being changed.
- Verify commands against the current Makefile, scripts, package manifests, and CI workflows. Flag discrepancies rather than assuming older examples work.
- Keep changes focused on one concern and follow existing package patterns.
- Avoid unrelated refactoring, formatting, and dependency updates.
- Discuss new dependencies and significant API or behavioral changes with maintainers before introducing them.
- Preserve existing user changes and update documentation for public changes.
- Review compatibility and migration requirements for API, configuration, storage, and event-format changes.
- Consult tiger-style-rules.md and
.golangci.yml. - Handle errors explicitly and retain useful diagnostic context.
- Consider concurrency, bounded resource use, cancellation, and timeouts.
- For queues and deliveries, consider retries, duplicates, ordering guarantees, and recovery after interruption.
- Preserve authentication, authorization, tenant isolation, and outbound destination validation. Never commit secrets or log credentials or tokens.
- Consult the dashboard README and package scripts.
- Follow existing frontend conventions.
- Exercise affected browser flows against a running backend.
- Add meaningful regression coverage for bug fixes where practical.
- Run formatting, vetting, linting, and tests appropriate to the change.
make testruns Go tests with race detection, excluding E2E packages.- Select relevant E2E targets from the Makefile and check prerequisites first.
- See evals/README.md for historical coding-agent tasks, independent grading, calibration requirements, and privacy rules.
- Existing product tests support grading; running them alone does not measure an agent's ability to implement a task. Uncalibrated tasks are candidates.
- For agent comparisons, freeze the task, starting revision, and independent grading checks; isolate attempts and record model, duration, and tool failures.
- Report checks actually run, results, and pending validation. Never claim an unexecuted check passed or present historical test provenance as an eval run.
- Use established generators rather than manually editing generated output.
- Consult
mockgen,generate_docs, anddocs-generatedMakefile targets. - Review generated diffs for unintended changes.
- Explain the problem, resulting behavior, and validation; link relevant issues.
- Use meaningful commit messages and disclose compatibility and operational risks.
- Review outgoing files and descriptions for secrets and local-only information.
- Before release work, read RELEASE.md, the monthly release workflow, and current CI.
- Resolve conflicting release instructions with a maintainer before publishing.
- Distinguish prepared, tested, published, and deployed work.