agentleFS
asta-nguyenasta-nguyen/agent-devkit/skills/plan-feature/SKILL.md

plan-feature

Skill4 starsChanged 2 months agoMITView on GitHub

Use when a written design.md needs its tasks.md before the joint design-and-tasks approval, or when an open change or legacy plan needs revision for a behavior, API, or cross-file change.

30 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

Used with

Projects with it also use explore, review and 3 more.

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.
  • genomewalkerSkill

    explore

    RLM-style recursive memory exploration - dynamically navigate the memory graph

    Also in 3 projects that use this

  • SentrySkill

    review

    Fixture review skill.

    Also in 5 projects that use this

  • DockerSkill

    review-pr

    Review one or more incoming Docker documentation pull requests as a maintainer. Independently validate technical claims, assess editorial fit and information architecture, choose a verdict, and draft exact inline or PR-wide feedback behind a confirmation gate. Use for requests such as "review PR 123", "is this PR correct?", "does this information belong here?", "validate this PR", or "help review backlog PRs". Do not use to maintain or fix a PR you own; use maintain-pr for that.

    Also in 3 projects that use this

  • AWSSkill

    ship

    Finalize dev work on an issue branch — update docs, push, and open a PR.

    Also in 3 projects that use this

  • GoogleSkill

    code-review

    Adversarial systems, concurrency, memory-model, and architecture code review protocol. Use this skill whenever asked to review code, PRs, diffs, or branches in Filament.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 5 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 29 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create the folder only when writing the plan.

    line 14
  2. Very specific:

    Route handoff from brainstorm-feature: write the plan and present it for the joint approval (step 9).

    line 18
  3. Very specific:

    Route handoff from estimate-feature: when the user asked for an estimate and only a written design.md exists, write the plan, call …

    line 20
  4. Very specific:

    Read the design, AGENTS.md, and related wiki pages.

    line 34
  5. Very specific:

    Follow existing patterns; do not unilaterally restructure.

    line 44
  6. Very specific:

    Add ## Approval Gate and ## Decision Log as the template shows.

    line 69
  7. Distinctive:

    Run the self-review below and fix issues inline.

    line 75
  8. Very specific:

    Do not estimate effort here.

    line 76
  9. Distinctive:

    Before requesting approval or delivering a planning-only result, give a short summary in the user's language with links to the written …

    line 79
  10. Very specific:

    If the user limited the request to design or planning, record the approval and stop.

    line 95
  11. Distinctive:

    Every step must contain the actual content an engineer needs.

    line 101
  12. Distinctive:

    Prefer the fewest tasks that make the sequence and verification unambiguous.

    line 123

“Common” is measured: kept by at least 6 in 10 of the 29 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

Plan Feature

Turn a design into a short execution plan before editing code. For a new change, brainstorm-feature writes design.md and calls this skill before any approval; the user then approves design.md and tasks.md together, once. "The design" means that written design.md. Save tasks.md beside it in the change folder (using-devkit's ## Change lifecycle). A legacy design with no change folder is covered by ../using-devkit/references/legacy.md; read it when you meet one. Create the folder only when writing the plan.

Entry points

  • Route handoff from brainstorm-feature: write the plan and present it for the joint approval (step 9).
  • Route handoff from estimate-feature: when the user asked for an estimate and only a written design.md exists, write the plan, call estimate-feature before the approval gate, and present the plan and estimate together.
  • Route handoff from implement-task: a material decision found during implementation. Revise the open tasks.md, record the decision in its ## Decision Log, reset the gate to pending, present the revision for approval, and resume implement-task after re-approval.
  • Repair handoff (e.g. estimate-feature asking for missing task detail in an existing plan): fix the plan, run the self-review, and return to the caller with the gate preserved or reset and the reason. Do not call estimate-feature or ask for approval, and never start implementation.

Process

  1. Read the design, AGENTS.md, and related wiki pages. If application source exists, call the available Skill entry whose local name is read-codebase-context to inspect affected source, callers, and tests; otherwise map planned files and interfaces from the design and state that callers and tests do not exist yet. Revise an open change's tasks.md in place; new architectural work after a change is archived gets a new design and folder. Keep completed tasks and results historical; add new work as a new task only while the change is open. Extract only exact constraints from the design that bind more than one task; infer none from convention.
  2. Map which files will be created or modified and what each is responsible for. Follow existing patterns; do not unilaterally restructure. Before accepting a proposed unit, apply implement-task's ## Implementation ladder, planning new code only at its last step. Every new file, abstraction, dependency, or configuration value must map to an approved requirement, and minimal design must not remove explicit behavior, trust-boundary validation, security, accessibility, or error handling that prevents data loss.
  3. Split the work into small, ordered tasks with the fields in references/tasks-template.md (read it now). Put contract and data changes before their callers and tests beside the behavior they verify. Map every approved edge case to a task and a specific check. If planning exposes an unresolved case that would change observable behavior, scope, an interface, a schema, or a protected boundary, stop and call brainstorm-feature; leave only behavior-equivalent details for implement-task rulings. Add no tasks for speculative abstractions, and do not add review-and-verify or commit steps; implement-task owns the final review. Plans describe intended behavior and interfaces, not code: include no function bodies or full test code. Include exact snippets only when the precise string, regex, config value, or contract shape is itself a requirement.
  4. Give each task a meaningful Verify check using the existing test runner and a failing test when it protects changed behavior. With no tests, use a repeatable check available in the project; do not add a test runner solely for the plan. Record the expected result, and mark a behavior cannot verify when no check can establish it. For a source-less project, the first task bootstraps the runtime and first entry point the design specifies, with its verification; if the design does not establish them, call brainstorm-feature.
  5. Add ## Approval Gate and ## Decision Log as the template shows. A new tasks.md is Status: pending; revisions follow the template's revision rule.
  6. Save the complete plan with ## Design, ## Global Constraints, and ## Impact map (seven fields from read-codebase-context's ## Trace and record the impact). The folder is indexed only on approval (step 9).
  7. Run the self-review below and fix issues inline.
  8. Do not estimate effort here. When the user explicitly requests an estimate, call estimate-feature before presenting the approval gate; otherwise present none. An estimate never authorizes implementation.
  9. Before requesting approval or delivering a planning-only result, give a short summary in the user's language with links to the written files: Goal; Scope and impact; Tasks (count and ordered titles); Verification (checks, naming any limit); Documentation (pages to update, or wiki not applicable); Risks and decisions. Summarize the actual design and tasks, not a second plan. State whether approval authorizes implementation or the request stops at planning. Unresolved material decisions return to brainstorm-feature; never present the plan as ready to implement. Apply the template's revision rule first: an existing plan that stays approved is summarized without a new approval request. Otherwise present design.md, tasks.md, any delta.md, and the requested estimate together and ask for one approval, then stop without editing application code. Revise the files in place until the user explicitly approves; then set Status: approved, index the folder per using-devkit's ## Change lifecycle, and continue: call setup-codebase first when the target has no AGENTS.md or application source, then implement-task. That approval is the gate, so do not ask again. If the user limited the request to design or planning, record the approval and stop. End with the Next: block from using-devkit; for a pending gate it names the exact approval the user must give.

No placeholders

Every step must contain the actual content an engineer needs. Never write: "TBD", "TODO", "implement later"; "add appropriate error handling" without saying what; tests described as code rather than named cases with input → expected result; "similar to Task N" (repeat the details); steps that omit the specific behavior, interface, or check; references to types or functions defined in no task.

Self-review

Fix only issues that could change approved behavior, scope, plan correctness, or execution; skip wording preferences. Check: spec coverage (a task per design requirement); placeholders (none, per the section above); type consistency across tasks; links (design and index resolve; the design links only wiki pages read as context); gate (new tasks are Progress: todo with Result: none, existing tasks keep theirs, a new tasks.md has only Status: pending); minimal design (anything removable via the ladder); decision consistency (persisted decisions agree with tasks and approval; material changes leave the gate pending); global constraints (each copied from the design and consistent with every task, or None.); edge cases (each mapped to a task and check, with unresolved behavior routed to brainstorm-feature).

Prefer the fewest tasks that make the sequence and verification unambiguous.

Red flags

ThoughtReality
"The user said implement earlier, so the plan is approved"An instruction given before both files were presented is not approval.
"Approval is in the chat history"Only this conversation after both files were shown, or the gate's Status: approved, counts.
"This decision is small; keep the gate approved"A material change to scope, behavior, an interface, a schema, or risk resets it to pending.

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.

Reviews can't be read right now.

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.