implement-issue

End-to-end workflow for implementing an OpenDataSuite GitHub issue — read the issue and roadmap, plan against acceptance criteria, implement with tests, verify, and open a PR. Use when asked to "work

SKILL.md

GitHub

Implement an ODS issue

1. Understand

  1. Read the issue (GitHub MCP issue_read or gh issue view N) including comments.
  2. Find it in docs/ROADMAP.md / .github/milestones.json: note milestone and whether it is critical-path.
  3. Check dependencies: issues listed earlier in the same milestone table, or in an earlier milestone, that this one builds on. If a hard prerequisite is not done, say so and propose doing the prerequisite (or a minimal stub of its contract) first.
  4. If it is listed as a duplicate in ROADMAP §5, work on the canonical issue instead.

2. Plan

  • Turn every acceptance-criteria bullet into a checklist item with the test that proves it.
  • Identify crates touched. If a new crate or contract is needed → new-crate skill.
  • If the issue says "ADR" or introduces a new persisted format / contract / dependency class → write the ADR first with the adr skill (can be the same PR).
  • For non-trivial issues, delegate to the issue-planner subagent and review its plan.

3. Implement

  • Follow AGENTS.md rules (provider neutrality, conservative defaults, explainability, determinism). Contracts go in ods-sdk; vendor code in providers/.
  • Add a fake implementation + conformance test for any new contract.
  • Human output + --json for any new CLI command; snapshot-test both.
  • Keep scope to the acceptance criteria; file follow-ups as new issues (ask first).

4. Verify

  • Run the verify skill. All checks must pass.
  • Ask the architecture-guard subagent to review the diff; fix any violations.

5. Deliver

  • Conventional Commit messages; reference the issue (Part of #N / Closes #N).
  • Push to the assigned branch. Open a PR only if the user asked; the body should map each acceptance criterion to evidence (test name, command output, file).