agentleFS
jabbott-iiijabbott-iii/Tidepool/.claude/skills/next-task/SKILL.md

next-task

Skill1 starsChanged 10 days agoApache-2.0View on GitHub

Pick the next unblocked task from docs/roadmap.md (or the task ID given) and carry it through the project's definition of done, leaving changes uncommitted for review.

10 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

24 rules. 0 are common, 12 are distinctive, and 12 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 12

    Distinctive

    This author's own choices. Worth a look.

  • 12

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If a task ID was given, use it.

    line 15
  2. Very specific:

    If the task has a Decide first line, ask each question in it before writing code, unless docs/decisions.md already records the answer:

    line 21
  3. Distinctive:

    Read the docs that AGENTS.md → "Read before working" lists for the crates the task touches, starting with each crate's README.md.

    line 39
  4. Distinctive:

    Read the existing code in those crates before writing new code.

    line 41
  5. Distinctive:

    If the spec and the code disagree, or the spec is ambiguous in a way that changes a public API, a file format, a protocol or the dependency …

    line 44
  6. Very specific:

    Do exactly what the task's Do lines and acceptance criteria require.

    line 50
  7. Distinctive:

    Use the subagent that owns the crate (table in CLAUDE.md).

    line 51
  8. Very specific:

    New third-party crate needed?: Stop and run /evaluate-dependency.

    line 52
  9. Very specific:

    OS bindings needed?: Follow /add-ffi-binding.

    line 53
  10. Very specific:

    A new or changed outside tool?: Follow /add-managed-tool.

    line 54
  11. Very specific:

    A design decision changed?: Run /record-decision and update the affected docs.

    line 55
  12. Distinctive:

    Write the tests the acceptance criteria call for, beside the code.

    line 57
  13. Distinctive:

    Start every new source file with the license header (ADR-0021).

    line 58
  14. Very specific:

    Ask security-reviewer to review when the change touches sys, platform, toolchain, mobile, agent, or any parser.

    line 59
  15. Distinctive:

    If the change spreads beyond three crates, pause and tell the user the task may be mis-scoped.

    line 61
  16. Very specific:

    Run every command in the task's Verify line.

    line 66
  17. Distinctive:

    Do everything possible here.

    line 73
  18. Distinctive:

    Leave the box unchecked.

    line 74
  19. Very specific:

    Add an indented ⏳ needs <OS/tool>: <exact remaining check> line under the task.

    line 75
  20. Distinctive:

    Check the box only if every acceptance criterion is met and every verification command ran and passed in this session.

    line 80
  21. Very specific:

    Add brief indented Note: lines under the task where useful: pinned versions, measured numbers, tool facts checked and their sources.

    line 82
  22. Very specific:

    Update the README.md of every crate whose built state changed: its ## Status names the task (ADR-0023).

    line 84
  23. Distinctive:

    Update docs only where behavior or design actually changed.

    line 86
  24. Distinctive:

    Leave all changes uncommitted.

    line 90

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Work the next roadmap task

Requested task: $ARGUMENTS (empty means "pick the next unblocked task").

1. Choose the task

  1. Open docs/roadmap.md.
  2. If a task ID was given, use it. If its Depends tasks are not all checked, stop and tell the user which ones block it.
  3. Otherwise, take the first unchecked task whose Depends are all checked.
    • Skip a task whose remaining work is marked ⏳ needs <OS/tool> when this session can't provide it. Mention any task you skipped and why.
  4. If the task has a Decide first line, ask each question in it before writing code, unless docs/decisions.md already records the answer:

    • for a Proposed ADR, present its decision and alternatives in a few sentences and ask whether to accept it;
    • for any other question, lay out the options the line names, with their costs, and recommend one.

    Record each answer with /record-decision (accept, supersede, or a new or extended ADR) before writing code. If the user is not there to answer, stop after the preparatory reading and say what is waiting.

  5. Tell the user in one or two sentences which task you're starting and how you plan to approach it. Create a task list whose last item is "Verify and report".

2. Load context, and only the context you need

  • AGENTS.md is already loaded. Re-read its "Non-negotiable rules" if unsure.
  • Read the docs that AGENTS.md → "Read before working" lists for the crates the task touches, starting with each crate's README.md.
  • Read the existing code in those crates before writing new code. Follow established patterns and naming. Where the task says to port from Nadir, read the matching Nadir code first and port it with a note.
  • If the spec and the code disagree, or the spec is ambiguous in a way that changes a public API, a file format, a protocol or the dependency matrix: stop, describe the conflict, propose the safest option, and ask.

3. Implement the smallest complete change

  • Do exactly what the task's Do lines and acceptance criteria require.
  • Use the subagent that owns the crate (table in CLAUDE.md).
  • New third-party crate needed? Stop and run /evaluate-dependency.
  • OS bindings needed? Follow /add-ffi-binding.
  • A new or changed outside tool? Follow /add-managed-tool.
  • A design decision changed? Run /record-decision and update the affected docs.
  • Write the tests the acceptance criteria call for, beside the code.
  • Start every new source file with the license header (ADR-0021).
  • Ask security-reviewer to review when the change touches sys, platform, toolchain, mobile, agent, or any parser.
  • If the change spreads beyond three crates, pause and tell the user the task may be mis-scoped.

4. Verify honestly

  • Run every command in the task's Verify line.
  • Then run /verify (cargo xtask ci).
  • Fix failures by fixing the code. Never weaken a lint, delete or #[ignore] a test, raise a tolerance, or re-bless goldens to make a check pass, unless the user explicitly agrees.
  • Criteria that need another OS, a real tool, a device or manual observation:
    1. Do everything possible here.
    2. Leave the box unchecked.
    3. Add an indented ⏳ needs <OS/tool>: <exact remaining check> line under the task.

5. Update the roadmap and docs

  • Check the box only if every acceptance criterion is met and every verification command ran and passed in this session.
  • Add brief indented Note: lines under the task where useful: pinned versions, measured numbers, tool facts checked and their sources.
  • Update the README.md of every crate whose built state changed: its ## Status names the task (ADR-0023).
  • Update docs only where behavior or design actually changed.

6. Do not commit

Leave all changes uncommitted. Do not commit, amend, rebase, tag or push.

7. Report

  • Task: ID and title, and whether its box is now checked.
  • Changes: what changed, and the files affected.
  • Validation: each command run, with its real result.
  • Not run: checks that weren't run, and why. Include ⏳ items.
  • Assumptions you made.
  • Unresolved issues and residual risks.
  • Commit message: a proposed message in the repository's style: the area first, then the summary and the task ID, as in text: encodings and atomic save (M1.3) (CONTRIBUTING.md → "Commits and pull requests").
  • Next: the next unblocked task ID.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.