Skip to content

gpsaggese/helpers / .claude/skills/auto_task.create_plan_for_feature

auto_task.create_plan_for_feature

Verified

Break a complex feature into an ordered PR plan, with a spec per complex PR

Goal

  • Given a request to implement a complex feature, break it into an ordered sequence of small, reviewable PRs
  • Write a full spec for each PR complex enough to need one
  • Do not implement any code

Input

  • The feature/component description, and optionally an existing plan file to extend

When to Use This Skill

  • Use it when one request needs several independently reviewable changes, and at least one has a design decision worth writing down
  • If the request fits in one PR, skip to "Write a Spec for a PR" with that PR as the whole request

Workflow

Understand the Request

  • Read the related code and the existing code in the target directory
  • If the user references a roadmap file (e.g., plan.<TOPIC>.md or docs/04-plan.md per .claude/skills/project.rules.md), read it for the exact PR scope
    • Use the plan ids (e.g., PR_M10) for the PR<NUM> bullets and spec files, per .claude/skills/project.rules.md section "Name the PRs"
  • Read the architecture doc of the target directory, if one exists
  • If the problem or solution is unclear, follow .claude/skills/auto_task.rules.md section "Ask for Clarification Before Executing an Unclear Plan" before decomposing

Decompose into PRs

  • Break the feature into an ordered sequence of small, cohesive PRs, each reviewable and testable alone
  • Prefer PRs that change whole files over splitting one file's logic across PRs
  • State each PR's dependencies explicitly
  • Merge low-risk, low-complexity PRs instead of over-fragmenting

Decide Which PRs Need a Full Spec

  • A PR needs a full spec if it adds a new interface (class, function, CLI flag), makes a non-obvious design decision, or carries meaningful risk (shared/core code, hard to reverse, many call sites)
  • Small, mechanical, low-risk PRs (rename, config tweak, adding a test) need no spec: the plan bullet is enough
  • For each PR that needs a spec, write it with "Write a Spec for a PR" below, using <TOPIC>_PR<NUM> as the topic and the plan file below
    • This produces spec.<TOPIC>_PR<NUM>.md

Write a Spec for a PR

  • Create a markdown document following .claude/templates/specs.template.md
  • Follow .claude/skills/markdown.rules.md and .claude/skills/text.rules.md
  • Save it as spec.<TOPIC>.md in the current directory
    • E.g., to spec PR_P2b from plan.md, save spec.PR_P2b.md
    • In a project docs/ dir, add the slot number: save 06-spec.PR_P2b.md and read 04-plan.md, per .claude/skills/project.rules.md section "Name the Docs with a Slot Number"
    • Do not name it plan.<TOPIC>.md: that name is reserved for the roadmap
  • Describe the architecture only. Do not implement any code
    • Short illustrative snippets (interface, schema, signature) are fine
  • Distinguish facts (from reading existing code) from decisions
  • Reference actual file, class, and function names, not generic paraphrases
  • Keep the spec under 100 lines, wrapped at 85 characters
  • Update the architecture file, if one exists, per .claude/skills/readme_file.write_architecture/SKILL.md
    • Add only what shows how the pieces work together, not how the new component works internally

Write the Plan

  • Create or update the plan file (e.g. tasks.md, or 05-tasks.md in a project's docs/ dir) following .claude/templates/auto_task.template.md
  • Fill in the * Repo: checklist per .claude/skills/auto_task.rules.md section "Multi-Repo Issues, Branches, and PRs"
  • For each PR<NUM> bullet, link its spec when one was written: - [ ] PR<NUM>: <Goal> (see spec.<TOPIC>_PR<NUM>.md)
  • Write bullets per .claude/skills/markdown.rules.md and .claude/skills/text.rules.md, with minimal text
  • Do not change any code: only propose the plan and specs

Rank the PRs

  • Order by dependency first: a PR never precedes a PR it depends on
  • Among independent PRs, order by increasing complexity, starting with the highest design confidence

Conventions

  • Follow .claude/skills/auto_task.rules.md for plan format, task list format, and issue/branch/PR naming
  • Follow .claude/skills/architecture.rules.md for layering and interface design
  • Follow .claude/skills/coding.rules.md and .claude/skills/testing.rules.md for illustrative snippets
  • Say in each spec file how it was generated, right after the title:
    - Generated by `.claude/skills/auto_task.create_plan_for_feature/SKILL.md`
    
  • Say in the plan file how it was generated, at the top of the file:
    <!-- Generated by `.claude/skills/auto_task.create_plan_for_feature/SKILL.md` -->
    

Constraints

  • Do not implement any code
  • Do not create the GitHub issue or branch: the auto_task.execute_* skill does that later
  • Distinguish facts from decisions made while decomposing

Verification

  • The plan follows .claude/templates/auto_task.template.md
  • Every PR<NUM> bullet states a standalone goal and explicit dependencies
  • Every complex PR has a linked spec.<TOPIC>_PR<NUM>.md (not plan.<TOPIC>.md)
  • Every spec section in .claude/templates/specs.template.md is filled in or marked "Not applicable"
  • No spec contains full implementation code, only interfaces/illustrative snippets
  • Every spec's architecture doc was updated, if one exists
  • PRs are ordered by dependency, then by decreasing confidence and increasing complexity
  • Only the plan and spec files were written; no code was changed
  • Unclear problems or solutions were raised with the user before finalizing
  • Each spec file and the plan file say they were generated by .claude/skills/auto_task.create_plan_for_feature/SKILL.md
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 192kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 192kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 192kUpdated Claude CodeCodexVerifiedInstall