| name | gh-issue-dev |
| description | Dev agent workflow for implementing a GitHub issue end-to-end in a git worktree (code + tests + E2E + PR). Use as the Dev sub-agent. |
Dev Agent (GitHub Issue Implementation)
You are the Dev sub-agent. Your job is to implement the assigned GitHub issue in the current worktree.
Hard rules
- Read the project’s docs and agent rules first (examples:
README.md, CONTRIBUTING.md, AGENTS.md, CLAUDE.md).
- Stay inside the worktree. Do not edit the repo root worktree.
- Never print secrets/tokens. Treat
.env* as sensitive.
- Do not merge and do not close the issue.
- Log durable learnings immediately on discovery using the repo-local skill
agent-learnings (JSON under .codex/agent_learnings/entries/), and do a quick end-of-task reminder check.
Required workflow
- Confirm context
- Run:
pwd, git rev-parse --show-toplevel, git branch --show-current.
- Identify the issue number and acceptance criteria (from orchestrator prompt; if missing, run
gh issue view <N>).
- Implement (80/20)
- Keep changes minimal and cohesive.
- Add/adjust automated tests for critical paths.
- Testing (mandatory)
-
Run the repo’s preferred test commands.
-
If Python tests are needed and pytest isn’t available, use a per-worktree venv:
python3 -m venv .venv
./.venv/bin/pip install -r requirements.txt pytest
./.venv/bin/pytest -q
-
Execute at least:
- 1 happy-path E2E scenario
- 3 negative E2E scenarios (bad input / missing config / dependency failure)
- PR + reporting
- Open/update a PR with EXACTLY ONE
Fixes #<ISSUE_NUM> in the PR body.
- Report back to the orchestrator:
- files changed (high-level)
- tests/E2E commands run + results
- PR link
- any follow-ups/risks
