agentskills.codes
CO

coder

by eq-labSource

Implement a planned Pipeline GitHub Issue by number. Use when Codex is asked to execute an active plan end-to-end, update code/tests/docs, mark plan steps complete, run the relevant verification commands, and report results without editing issue labels, closing the Issue, moving the plan, or committ

Install

mkdir -p .claude/skills/coder && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/16537" && unzip -o skill.zip -d .claude/skills/coder && rm skill.zip

Installs to .claude/skills/coder

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Implement a planned Pipeline GitHub Issue by number. Use when Codex is asked to execute an active plan end-to-end, update code/tests/docs, mark plan steps complete, run the relevant verification commands, and report results without editing issue labels, closing the Issue, moving the plan, or committing.
304 chars✓ has a “when” triggerlonger than Claude Code's old 250-char listing cap (fine on current versions)

About this skill

Coder

Implement exactly one GitHub Issue. The manager owns lifecycle labels, commits, pushes, and plan archival.

Start and finish with:

MODEL: <model> | EFFORT: <effort>

Required Context

  1. Read AGENTS.md.

  2. Read the Issue and comments:

    gh issue view <number> -c
gh issue view <number> --json title,body,labels,assignees,url

  3. Read .codex/skills/issue/SKILL.md.

  4. Read the active execution plan: docs/exec-plans/active/issue-<number>-*.md.

If the caller explicitly says Flow: trivial-frontend, there may be no plan. In that case, work directly from the Issue body and comments.

Workflow

  1. Confirm the user supplied an Issue number.
  2. Read the Issue, comments, labels, and active plan.
  3. If no active plan exists and this is not Flow: trivial-frontend, stop and report that planning is missing.
  4. Implement the Issue end-to-end:
    • Follow plan steps in order.
    • Add or update tests where useful.
    • Update docs listed in Docs to Update.
    • If the Issue or plan references Figma, use it as an implementation reference.
  5. Mark completed plan steps as you go. Do not create a plan for trivial frontend mode.
  6. Run verification scaled to the changes:
    • Rust changes: cargo clippy --all -- -D warnings.
    • TypeScript or docs changes: npx tsx scripts/lint-docs.ts.
    • Frontend changes: run the relevant frontend lint/build scripts from package.json or package docs.
    • Run the repo's fast test workflow when available.
  7. Fix failures before reporting done. If blocked by environment or dependency failures, report exact commands and errors.

Rules

  • Do not edit Issue labels, assign, close, commit, push, or move the plan to completed.
  • Do not ask for approval during implementation; execute the plan.
  • Respect existing code patterns and architecture boundaries.
  • Do not fix unrelated bugs inline. Log them in docs/exec-plans/known-bugs.md with date, location, symptom, root cause if known, and workaround.
  • Log intentional shortcuts or structural gaps in docs/exec-plans/tech-debt-tracker.md.
  • If the plan is wrong or incomplete, use judgement, note the deviation in the final report, and comment on the Issue.
  • Never revert user changes or unrelated worktree changes.

Trivial Frontend Mode

When the caller says Flow: trivial-frontend:

  • Work from the Issue body and comments.
  • Do not invent an execution plan.
  • Keep changes self-contained to the frontend.
  • Leave lint, build, and relevant tests green.
  • Touch product docs only when behavior visibly changes.

Output

Report:

  • Issue number and title
  • What changed
  • Tests added or updated
  • Verification commands and pass/fail results
  • Docs updated
  • Plan deviations, blockers, or concerns

Search skills

Search the agent skills registry