agentleFS
shawnclyborshawnclybor/clybor-claude-tooling/.claude/skills/task-plan/SKILL.md

task-plan

Skill6 starsChanged 28 days agoMITView on GitHub

Decompose a PRD into a sequenced checkbox task list. Each task is small enough to complete in one focused pass and tied to one or more PRD success criteria. Use after the PRD is accepted and before implementation begins. Triggers — "plan from the PRD", "break this into tasks", "task list for [feature]", "decompose [PRD]", "give me a plan".

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 9 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When picking up an existing PRD that lacks a plan

    line 13
  2. Distinctive:

    When re-planning after a Stage 6 evaluation surfaces structural issues

    line 14
  3. Distinctive:

    If a task fails any of these, split it.

    line 38
  4. Very specific:

    After the flat task list is drafted, invoke task-distributor to produce the parallelizable-group section.

    line 42
  5. Distinctive:

    No shared write target between parallel tasks

    line 44
  6. Distinctive:

    No read-after-write dependencies between parallel tasks

    line 45
  7. Distinctive:

    No contended shared resources (DB row, rate limit, file lock)

    line 46
  8. Distinctive:

    When the work is a fixed sequence of steps with a human review gate between each (research → draft → final.

    line 64
  9. Distinctive:

    Before writing the plan:

    line 73
  10. Distinctive:

    If a task is large, have I split it before adding to the plan?

    line 80

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

Task Plan

Turn an accepted PRD into a flat, sequenced task list with checkboxes. The plan is the contract between the PRD and the implementation. Tasks too big = invisible progress; tasks too small = ceremonial overhead. Aim for one-focused-pass per task.

When to use

  • Right after a PRD is written and accepted
  • When picking up an existing PRD that lacks a plan
  • When re-planning after a Stage 6 evaluation surfaces structural issues

Input

  • prd_path — path to the PRD (e.g., docs/PRDs/oauth-token-refresh.md)

Output

docs/PRDs/<feature_slug>-plan.md, copied from templates/plan-template.md and filled in. Sections:

  1. Tasks — checkbox list with Touches, Verifies, Check, Notes per task
  2. Parallelizable groups — which tasks can run concurrently
  3. Risks per task

Sizing rules

Each task should pass these checks:

  1. One-pass completable — ≤ 2 hours of focused work
  2. Single responsibility — advances one concern; does not bundle unrelated changes
  3. Independent where possible — runs in parallel with siblings when safe
  4. Tied to a criterion — advances at least one PRD success criterion
  5. Testable in isolation — there is a check (unit test, integration test, smoke) that confirms this task alone is done

If a task fails any of these, split it.

Parallelizable-group analysis

After the flat task list is drafted, invoke task-distributor to produce the parallelizable-group section. It applies the safety rules:

  • No shared write target between parallel tasks
  • No read-after-write dependencies between parallel tasks
  • No contended shared resources (DB row, rate limit, file lock)

Take the task-distributor output verbatim into the plan's "Parallelizable groups" section. The ralph-implement skill at Stage 4 reads this section to spawn parallel work.

Common decomposition patterns

PRD shapePlan shape
New featureSchema → API → client → tests → docs
RefactorAdd new path → migrate callers in batches → remove old path
IntegrationContract / types → happy path → error paths → retries → observability
Bug fixReproduce → narrow → fix → regression test
MigrationBackup → write new state alongside old → verify equivalence → switch reads → drop old

Starting points, not formulas. The plan must match the actual work.

Stage-pipeline plans

When the work is a fixed sequence of steps with a human review gate between each (research → draft → final; data → eval → report), the plan scaffolds the execution structure instead of only listing tasks:

  • Copy templates/stage-pipeline/ from the master tooling repo to where the work lives; rename/renumber stages (01-, 02- — zero-padded prefixes encode execution order; reordering = renaming folders).
  • Each stage folder carries a CONTEXT.md contract — Inputs (source / where / why), Process (numbered steps), Outputs (artifact / location / format) — under 80 lines, plus an output/ folder for the handoff artifact.
  • A stage reads ONLY its declared Inputs (previous stage's output/ + named references). Everything else is do-not-load.
  • Tasks then map 1:1 to stages; each stage's DONE check = its Outputs exist and satisfy the contract. The human inspects output/ before the next stage runs — every output is an edit surface.

Pre-flight checklist

Before writing the plan:

  1. Have I read the full PRD, not just the title?
  2. Does every PRD success criterion appear as at least one task's Verifies: field?
  3. Is every task one-pass completable?
  4. Has task-distributor run on the flat task list to produce parallelizable groups?
  5. Have I flagged task-level risks separately from PRD-level risks?
  6. If a task is large, have I split it before adding to the plan?

Anti-patterns

  • One-line tasks ("write tests") that hide a day of work — split
  • Mega-tasks ("implement auth") that span the whole feature — split
  • Tasks that do not tie to any PRD criterion — either find the criterion or remove the task
  • Hand-rolled parallel grouping — let task-distributor do it; the safety rules are easy to skip when humans estimate
  • Plans that re-state the PRD — the plan is a decomposition, not a restatement

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.