Add test-writing task to Solo Dev Starter Suite

[Jordak]

Goal

Add a test-writing task to the Solo Dev Starter Suite.

Task shape

The task should ask the agent harness to add or update tests for existing behavior without requiring a production-code change, unless a tiny fixture change is necessary.

Acceptance criteria

• The task uses a real external repository if possible.
• The repo and commit are public and pinned.
• Deterministic graders verify the new tests exist and pass.
• The task has a verified reference_artifact.
• The task card is generated and readable by humans and AI assistants.
• Expected failure modes include risks such as context_miss, test_gap, over_edit, and resource_inefficient where appropriate.
• agentlab task verify-reference <task-bundle> passes.

Notes

Prefer a task where the intended test coverage is clear from existing behavior and not merely snapshot churn.

[Jordak]

This was generated by AI during triage.

Triage Notes

Recommendation: enhancement / ready-for-agent.

Agent brief:

• Add a test-writing task to the Solo Dev Starter Suite.
• The task should require tests only. Production-code changes are forbidden for the intended solution.
• Prefer a real public repository where the missing coverage is clear from existing behavior, not snapshot churn.
• The reference artifact should change only test files, except for narrowly justified fixture metadata if unavoidable.
• Graders should verify both behavior and coverage shape: the targeted test command should pass after the reference patch, and a robust structural check should confirm the new tests exercise the intended behavior or edge case.
• Avoid brittle exact-text checks when an AST, importable test module, or targeted test invocation can prove the same point.
• Include a generated task card, expected failure modes, and passing reference verification.

[Jordak]

Implemented Issue #5 as a pushed task slice.

• Commit: fa43bb1 on origin/codex/issue-5-test-writing-task
• Task: click-should-strip-ansi-tests-001 at tasks/starter/click-should-strip-ansi-tests-001
• Repo and pinned commit: https://github.com/pallets/click.git @ 8bd8b4a074c55c03b6eb5666edc44a9c43df38a2
• Upstream reference: Add parametrized tests for should_strip_ansi pallets/click#2731, reference commit cf0c36d33734af1de9ecd5d3305970eb26ebba7f
• Reference verification: passed with files_changed=1; report at tasks/starter/click-should-strip-ansi-tests-001/reference-report.md
• Codex smoke trial: passed, trial id 20260508-211730-click-should-strip-ansi-tests-001-codex-c5484e14; changed only tests/test_compat.py
• Tests and validation run: python3 -m agentlab task verify-reference tasks/starter/click-should-strip-ansi-tests-001; python3 -m agentlab task verify-reference --no-write-artifacts tasks/starter/click-should-strip-ansi-tests-001; python3 -m agentlab run --agent codex --task tasks/starter/click-should-strip-ansi-tests-001 --trials 1 --jobs 1; python3 -m agentlab task validate tasks; python3 .agents/skills/task-card/scripts/render_task_cards.py tasks --check; python3 -m unittest tests.test_tasks tests.test_task_cards tests.test_reference tests.test_cli; git diff --cached --check; commit hook task-card and task validation checks
• Limitations and follow-up: no larger batches were run, per the single-smoke workflow. Codex adapter metadata still reported model and cost as unknown, though token counts were captured in the trial result.

[Jordak]

Reopening because the resolving commit was pushed to a feature branch but had not yet landed on main. I will merge the branch to main before closing this again.

[Jordak]

Now correctly integrated before closure.

• Reopened because the first closure only had the resolving commit on a feature branch.
• Merged origin/codex/issue-5-test-writing-task into main with merge commit b173255.
• Pushed main: origin/main now contains the Issue Add test-writing task to Solo Dev Starter Suite #5 task slice and the original resolving commit fa43bb1.
• Updated AgentOS closure-discipline wording so a pushed feature branch is explicitly not enough; the resolving commit must be on the remote integration branch, usually main, or merged via PR before an implementation issue is closed.