agentleFS
serbanghitaserbanghita/plan-critique-skills/skills/critique/SKILL.md

critique

Skill13 starsChanged 2 months agoMITView on GitHub

Critique the user's plan from plan.md

80 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use delight, teach-impeccable and 10 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.
  • HK-hubSkill

    delight

    Add moments of joy, personality, and unexpected touches that make interfaces memorable and enjoyable to use. Elevates functional to delightful. Use when the user asks to add polish, personality, animations, micro-interactions, delight, or make an interface feel fun or memorable.

    Also in 33 projects that use this

  • HK-hubSkill

    teach-impeccable

    One-time setup that gathers design context for your project and saves it to your AI config file. Run once to establish persistent design guidelines.

    Also in 19 projects that use this

  • zwy123zwySkill

    normalize

    将用户随意输入的提示词规范化为结构清晰、便于大模型执行的标准化提示词。适用于需要整理、统一提示词格式时调用。

    Also in 19 projects that use this

  • owshq-mecSkill

    layout

    Improve layout, spacing, and visual rhythm. Fixes monotonous grids, inconsistent spacing, and weak visual hierarchy. Use when the user mentions layout feeling off, spacing issues, visual hierarchy, crowded UI, alignment problems, or wanting better composition.

    Also in 8 projects that use this

  • +8 more from HK-hub

The file

Show its rules and text

We split this skill into the rules it gives an agent

30 rules. 0 are common, 13 are distinctive, and 17 are very specific to this project.

  • 0

    Common

    Most of the 73 edited versions we read keep these.

  • 13

    Distinctive

    This author's own choices. Worth a look.

  • 17

    Very specific

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

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

    Read the working agreement at working-agreement.md before step 1 and follow it for the whole run.

    line 16
  2. Distinctive:

    Check edge cases specific to the area the plan touches: concurrency, nulls, error paths, boundaries.

    line 31
  3. Distinctive:

    Keep re-scanning the plan until two consecutive passes surface nothing new.

    line 33
  4. Very specific:

    Do not edit project code or plan.md during this phase.

    line 36
  5. Very specific:

    Read .claude/plan-critique-config.json, get plansFolder path from settings.

    line 42
  6. Very specific:

    Read the current session's plan from [plansFolder]/.sessions/[sessionPID] if it exists.

    line 49
  7. Very specific:

    If no plan folders exist: Respond with "No plans found.

    line 52
  8. Very specific:

    If sessionPlan exists and matches a plan folder, auto-select it.

    line 54
  9. Very specific:

    Update the session file [plansFolder]/.sessions/[sessionPID] with the selected plan slug (create if needed).

    line 67
  10. Very specific:

    Read the plan file at [plansFolder]/[selected-plan]/plan.md

    line 68
  11. Very specific:

    If plan.md is empty: Respond with "Plan file is empty.

    line 70
  12. Distinctive:

    If CLAUDE.md does not exist in project root: Respond with "Create a CLAUDE.md file in the root of your project."

    line 71
  13. Distinctive:

    If the probe succeeds, LSP is available.

    line 79
  14. Distinctive:

    If the probe fails or the LSP tool is not available for the detected language, inform the user before continuing: "No LSP support detected …

    line 81
  15. Distinctive:

    Do not block the critique.

    line 88
  16. Very specific:

    Read the existing "Iteration: [number]" at [plansFolder]/[selected-plan]/critique.md (if it exists) and determine the current iteration …

    line 90
  17. Very specific:

    If the plan references files that are in the [plansFolder]/[selected-plan]/ folder, review those as well and add them to the context of the …

    line 92
  18. Distinctive:

    When suggesting a split, be specific about which sections should become their own plan.

    line 116
  19. Very specific:

    Write the critique to [plansFolder]/[selected-plan]/critique.md.

    line 117
  20. Very specific:

    When writing the critique, follow the original chapters from plan.md.

    line 118
  21. Distinctive:

    Use the format in critique-format.md.

    line 121
  22. Very specific:

    When critiquing, always analyze codebase structure (existing files, directories, patterns), project standards from CLAUDE.md, the README.md …

    line 125
  23. Very specific:

    When doing the writeup of the critique, in the "Description" area make use of the line numbers from plan.md and reference them so that the …

    line 128
  24. Distinctive:

    Use code intelligence to verify the plan against the actual codebase.

    line 130
  25. Very specific:

    Verify types exist: use LSP go-to-definition.

    line 132
  26. Very specific:

    Check method/function existence: use LSP go-to-definition.

    line 134
  27. Distinctive:

    Verify file paths exist with Glob before referencing them in the critique

    line 139
  28. Distinctive:

    Add the found issues/observations list in the beginning of the critique.md file as a Table of contents

    line 140
  29. Distinctive:

    Always follow the chapters from plan.md as a structure for critique

    line 141
  30. Distinctive:

    Only include current issues: The critique should reflect the current state of plan.md.

    line 146

“Common” is measured: kept by at least 6 in 10 of the 73 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

You are performing an iterative review of the user's execution plan. Critique the plan, code, architecture, system design, and design patterns.

Fixed rules:

Read the working agreement at working-agreement.md before step 1 and follow it for the whole run. If that path does not resolve, look for working-agreement.md in the project root, then in .claude/. If it cannot be found anywhere, tell the user it is missing and apply the rules listed below.

These rules bind this phase and override any step below that conflicts with them:

  1. Be adversarial. Try to refute each finding before you write it down, and report only what survives the attempt. A finding you could not refute is worth more than three you did not test.
  2. Grade confidence per finding. Report a finding as CONFIRMED only when you verified it against the codebase. Label everything else UNVERIFIED and state what you could not check and why.
  3. Prove every finding with evidence: a file:line reference, a command output, or a failing test. Never assert a plausible but unverified conclusion, and never present a guess as a fact.
  4. Cite plan.md line numbers only from the copy you read in this run. Re-read plan.md before writing the critique and check that each line reference points at the text you describe. Merging a critique into the plan shifts every line below the merge point, so numbers carried over from an earlier iteration are wrong.
  5. Check edge cases specific to the area the plan touches: concurrency, nulls, error paths, boundaries. A generic "find bugs" pass over the plan is not enough.
  6. Keep re-scanning the plan until two consecutive passes surface nothing new. Stop at that point, not before.
  7. Require the plan to state its affected files and its verification approach. When a chapter is missing either one, raise it as a finding, because unlisted files are where import and dependency breakage hides.
  8. Do not edit project code or plan.md during this phase. This phase writes critique.md only. The user decides what moves into the plan.
  9. Be brief. No filler, no preamble, no emojis, no em dashes, no bold or italic text.

To do this, follow these steps precisely:

  1. Read .claude/plan-critique-config.json, get plansFolder path from settings. If the file doesn't exist or plansFolder is not set or is an empty string: Respond with "No plans folder configured. Run /plan:create first to set up."
  2. Get the Claude Code process ID by running: echo $PPID. Store this as sessionPID.
  3. Clean up stale sessions: Scan [plansFolder]/.sessions/ for files. For each file named with a PID, check if that process is still running via kill -0 [PID] 2>/dev/null. If the command fails (process not running), delete that session file. This is non-blocking cleanup.
  4. Read the current session's plan from [plansFolder]/.sessions/[sessionPID] if it exists. Store as sessionPlan.
  5. Scan [plansFolder]/ for subdirectories (each subdirectory is a plan). Exclude archived/ and .sessions/ folders and any files, only list plan directories. If no plan folders exist: Respond with "No plans found. Create one with /plan:create".
  6. Select the plan to critique:
    • If sessionPlan exists and matches a plan folder, auto-select it. Inform the user: "Using current session plan: [sessionPlan]"
    • Else if only one plan exists, auto-select it and inform user.
    • Otherwise, ask the user to select a plan from the list. Example:
      Available plans:
      1. add-user-authentication
      2. refactor-database-layer
      3. implement-caching
      
      Which plan would you like to critique? [1-3]
  7. Update the session file [plansFolder]/.sessions/[sessionPID] with the selected plan slug (create if needed).
  8. Read the plan file at [plansFolder]/[selected-plan]/plan.md
  9. Check for errors:
    • If plan.md is empty: Respond with "Plan file is empty. Edit [plansFolder]/[selected-plan]/plan.md"
    • If CLAUDE.md does not exist in project root: Respond with "Create a CLAUDE.md file in the root of your project."
  10. Detect project languages and verify LSP availability. LSP is the required tool for code verification in this phase, so this check is mandatory:
    • Detect the main project languages with Glob: tsconfig.json, *.ts, or *.tsx for TypeScript, composer.json or *.php for PHP, and the equivalent markers for any other language the plan touches.
    • Probe LSP: pick a source file referenced by the plan (or any project source file) and request go-to-definition or hover on a known symbol using the LSP tool.
    • If the probe succeeds, LSP is available. Use it for all code verification in this run, as described in the Notes below. Do not substitute Grep for checks LSP can answer.
    • If the probe fails or the LSP tool is not available for the detected language, inform the user before continuing: "No LSP support detected for [language]. The critique will fall back to Grep, which is less accurate. To enable code intelligence, install the LSP plugin for your language (for example typescript-lsp or php-lsp from claude-plugins-official, check with the /plugins command), or use an LSP setup skill such as https://github.com/github/awesome-copilot/tree/main/skills/lsp-setup to install and configure a language server."
    • Do not block the critique. Proceed with the Grep fallback and record in the critique Summary that LSP was unavailable, so the user knows symbol verification relied on Grep.
  11. Read the existing "Iteration: [number]" at [plansFolder]/[selected-plan]/critique.md (if it exists) and determine the current iteration number. If no critique exists, this is iteration 1.
  12. If the plan references files that are in the [plansFolder]/[selected-plan]/ folder, review those as well and add them to the context of the critique.
  13. Perform a thorough critique of the plan considering:
    • Clarity: Are requirements specific and unambiguous?
    • Completeness: Are all necessary steps included?
    • Order: Are dependencies between steps correctly sequenced?
    • Feasibility: Can each step be executed given the current codebase?
    • Risk: Are there potential side effects or breaking changes?
    • Standards: Does it comply with CLAUDE.md project standards?
    • Scope: Is scope reasonable? Any unnecessary additions?
    • Testability: How is success verified? Does each chapter name the test to write or the command to run?
    • Affected files: Does each chapter list the files it creates, modifies or deletes, and are those paths real?
    • Supporting materials: Are referenced files in the plan folder adequate?
  14. Evaluate whether the plan can be split into independent tasks. If the plan contains multiple features or changes that can be executed separately, strongly recommend splitting it into separate plans. Why this matters:

    • Reduces complexity during execution
    • Limits the context window needed for each plan
    • Makes critique iterations more focused and actionable
    • Allows independent tasks to proceed without blocking each other

    Example: A plan with "Add user authentication", "Refactor database layer", and "Add caching" should be split into three separate plans if these can be implemented independently.

    When suggesting a split, be specific about which sections should become their own plan.

  15. Write the critique to [plansFolder]/[selected-plan]/critique.md. When writing the critique, follow the original chapters from plan.md. The goal is to be able to easily override the plan.md if the user chooses to merge critique.md with plan.md

    Use the format in critique-format.md.

Notes:

  • When critiquing, always analyze codebase structure (existing files, directories, patterns), project standards from CLAUDE.md, the README.md file, dependencies (package.json, requirements.txt, etc.), git state if relevant, whether referenced files/APIs actually exist, and supporting files in the plan folder.
  • When doing the writeup of the critique, in the "Description" area make use of the line numbers from plan.md and reference them so that the user can easily find what text to replace or update.
  • Use code intelligence to verify the plan against the actual codebase. Always use LSP when step 10 confirmed it is available; use Grep only as a fallback when step 10 found no LSP support:
    • Verify types exist: use LSP go-to-definition; fallback is Grep for class, interface, type, or struct definitions
    • Check method/function existence: use LSP go-to-definition; fallback is Grep for function/def/fn declarations in the target file
    • Find usages/references: use LSP find-references; fallback is Grep for the symbol name across the codebase
    • Review diagnostics: use mcp__ide__getDiagnostics to pull current errors/warnings from the IDE for files referenced in the plan
    • Verify file paths exist with Glob before referencing them in the critique
  • Add the found issues/observations list in the beginning of the critique.md file as a Table of contents
  • Always follow the chapters from plan.md as a structure for critique
  • Be direct and constructive in feedback
  • Suggest multiple solutions when appropriate
  • Each critique iteration completely overwrites the previous critique.md file.
  • Discard addressed issues: If an issue from the previous critique has been fixed in plan.md, do not include it.
  • Only include current issues: The critique should reflect the current state of plan.md.
  • New unrelated observations: If new issues appear that don't fit under existing plan.md chapters, add them as new chapters at the bottom of the critique
  • Increment iteration number: Always increment from the previous critique's iteration number

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.