agentleFS
a-ghorbania-ghorbani/pocketpal-dev-team/skills/start-task/SKILL.md

start-task

Skill6 starsChanged 25 days agoMITView on GitHub

Start a new PocketPal delivery task from a work-item reference (any tracker) or a free-form description. Resolves the reference per context/issue-tracking.md, then runs the delivery loop through implementation, draft PR, independent review, and review-fix rounds when needed.

16 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use review.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • SentrySkill

    review

    Fixture review skill.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 14 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    DESIGN_INVALIDATED: check the restart triggers under "Going back" in docs/workflows/design-and-restarts.md before dispatching the …

    line 48
  2. Very specific:

    Gates: wait on tools/pr-gates.sh <PR> with a watcher that outlives your turn, not a poll-and-exit.

    line 50
  3. Very specific:

    REQUEST_CHANGES: follow "Independent review loop" in docs/workflows/review-and-finalization.md.

    line 51
  4. Very specific:

    Stop only on the conditions under "Autonomous-run contract" in pipeline.md.

    line 53
  5. Distinctive:

    When the task came from an internal tracker, its work item is the task's status for the PM.

    line 57

“Common” is measured: kept by at least 6 in 10 of the 14 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

Start Task

You are the top-level delivery controller for one PocketPal task. You dispatch every stage role yourself and route between them. Read docs/workflows/pipeline.md first: it is your runbook.

Task: $ARGUMENTS (if that placeholder is not substituted, the task is the argument the user gave with the invocation).

1. Resolve the reference

Resolve the reference by its shape, per context/issue-tracking.md:

  • GitHub issue (#123): gh issue view <n> --repo a-ghorbani/pocketpal-ai --json title,body,labels.
  • Plane (POC-123, …): ./tools/plane show <ref>.
  • Linear (FOU-123, legacy): ./tools/linear.sh issues, then match the identifier.
  • Anything else: a free-form description.

2. Dispatch intake with a self-contained brief

Use pocketpal-intake: <title>

Request:
<the issue / work-item body verbatim, or the description>

Metadata:
- Source: github | plane | linear | description
- GitHub issue: #<n>             # github only
- Tracker ID: <identifier>       # internal: keep out of every public GitHub artifact
- Labels / Priority / Status: <from the tracker, when present>

Repository: ./repos/pocketpal-ai

3. Run the pipeline

Once intake returns TASK_ID, join agistry as <TASK_ID>:orchestrator (the agistry skill), so the operator's replies reach this task even if the session is replaced.

Route on each stage's VERDICT by the routing table in pipeline.md. Beyond the table:

  • Design arena (DESIGN_EXPLORATION=YES): before the architect's first draft, write the handoff block to <STORY_DIR>/handoff.txt, run tools/design-arena.sh run <STORY_DIR> in the background (it takes up to a few hours), then dispatch the architect in merge mode with its exit code.
  • Slice review (SLICE_READY): give pocketpal-reviewer (tracer-slice mode) the slice's commit range.
  • DESIGN_INVALIDATED: check the restart triggers under "Going back" in docs/workflows/design-and-restarts.md before dispatching the architect's amendment.
  • Questions for the operator (AGENTS.md, Asking the operator): send a stage's NEEDS_INPUT questions and its defaulted product calls to operator in one message, each with its default and the research behind it, or to the PM's address (the PM: line in your dispatch) when a PM dispatched this task. Contract-shaping calls wait for the reply; the rest proceed on their defaults. Log each answer in decisions.md as a ruling (in a PM run the PM logs it), then resume.
  • Gates: wait on tools/pr-gates.sh <PR> with a watcher that outlives your turn, not a poll-and-exit.
  • REQUEST_CHANGES: follow "Independent review loop" in docs/workflows/review-and-finalization.md. You write the fix brief into the PR-fix story intake created, and you are the only writer of review dismissals in decisions.md.

Stop only on the conditions under "Autonomous-run contract" in pipeline.md. When a PM session dispatched this task, send stops to the PM's address (PM:) over agistry rather than to the operator. Otherwise finish with APPROVE, and the human reviews and merges.

4. Keep the tracker current

When the task came from an internal tracker, its work item is the task's status for the PM. At each milestone, post one comment and move the state, using that tracker's commands from "Status updates" in context/issue-tracking.md:

MilestoneStateComment
Intake doneIn Progress[started] <complexity> · <TASK-ID>
Spec approved[spec] approved
PR openedIn Review[pr] #<n> · head <sha7>
Review verdict[review] <verdict> · round <R> · head <sha7>
Stopped on a stop condition[blocked] <what is needed, from whom>
Final APPROVE[ready] #<n> · head <sha7> · CI <status>

Each comment is one line: facts only, no reasoning or explanation. Detail lives in the PR and the story. Skip this step for GitHub issues and free-form descriptions.

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.