agentleFS
Postpartum-genushyacinthus29Postpartum-genushyacinthus29/dotnet-skills/skills/dotnet-mcaf-human-review-planning/SKILL.md

dotnet-mcaf-human-review-planning

Skill13 starsChanged 8 months agoMITView on GitHub

Apply MCAF human-review-planning guidance for a large AI-generated code drop by reading the target area, tracing the natural user and system flows, identifying the riskiest boundaries, and prioritizing the files a human should inspect first. Use when the codebase is too large to review line-by-line and you need a practical review sequence plus a prioritized file list.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 9 are distinctive, and 5 are 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.

  • 5

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    leave reusable project context so future tasks are faster and safer

    line 22
  2. Distinctive:

    Read the nearest AGENTS.md and confirm scope and constraints.

    line 38
  3. Very specific:

    Run this skill's Workflow through the Ralph Loop until outcomes are acceptable.

    line 39
  4. Distinctive:

    Read enough of the target area and its immediate boundaries to understand the generated code before planning review.

    line 44
  5. Distinctive:

    Use that flow to derive the most efficient human review sequence.

    line 52
  6. Distinctive:

    Use the reviewer's domain knowledge as a force multiplier:

    line 53
  7. Distinctive:

    look for places where the new feature should behave like nearby existing flows

    line 55
  8. Very specific:

    If the user asks for a durable artifact, save the plan to the exact docs path they requested.

    line 67
  9. Distinctive:

    Use the Ralph Loop for every task, including docs, architecture, testing, and tooling work.

    line 87
  10. Distinctive:

    write a detailed execution plan from the chosen direction

    line 95
  11. Distinctive:

    Update the plan after each iteration.

    line 100
  12. Very specific:

    If a dependency is missing, bootstrap it or return status: not_applicable with explicit reason and fallback path.

    line 102
  13. Very specific:

    read references/review-plan-format.md for the output shape

    line 117
  14. Very specific:

    read references/risk-signals.md when deciding what deserves human attention first

    line 118

≈ “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

MCAF: Human Review Planning

Trigger On

  • a large AI-generated code drop needs a human review plan
  • the reviewer cannot inspect every line and needs prioritization
  • the user asks which files are highest risk before doing manual review
  • the user names a generated folder and wants a saved review plan for it

Value

  • produce a concrete project delta: code, docs, config, tests, CI, or review artifact
  • reduce ambiguity through explicit planning, verification, and final validation skills
  • leave reusable project context so future tasks are faster and safer

Do Not Use For

  • normal small pull-request review
  • automated bug finding without creating a human review sequence

Inputs

  • the target folder, feature area, or bounded context under review
  • the main user journeys or operational flows involved
  • any known architecture context, adjacent entities, or existing system rules
  • any exact output path the user wants for the saved plan

Quick Start

  1. Read the nearest AGENTS.md and confirm scope and constraints.
  2. Run this skill's Workflow through the Ralph Loop until outcomes are acceptable.
  3. Return the Required Result Format with concrete artifacts and verification evidence.

Workflow

  1. Read enough of the target area and its immediate boundaries to understand the generated code before planning review.
  2. Map the natural flow of operations first:
    • sign up or authentication
    • create
    • update
    • register or configure
    • execute primary business action
    • complete, archive, or finalize
  3. Use that flow to derive the most efficient human review sequence.
  4. Use the reviewer's domain knowledge as a force multiplier:
    • compare the generated code against known architecture and existing entities
    • look for places where the new feature should behave like nearby existing flows
    • prioritize boundaries where generated code may drift from established system rules
  5. Identify high-risk review zones:
    • entry points and orchestration layers
    • persistence and state transitions
    • cross-boundary integrations
    • permissions, validation, and invariants
    • side effects such as email, payments, jobs, or notifications
  6. Produce two separate outputs:
    • prioritized review flow
    • prioritized files or modules to inspect
  7. Present both outputs in chat.
  8. If the user asks for a durable artifact, save the plan to the exact docs path they requested; otherwise use docs/AREA/HUMAN_REVIEW_PLAN.md.

Deliver

  • a prioritized human review sequence
  • a prioritized list of files or modules to inspect first
  • both sections presented separately in chat
  • a saved HUMAN_REVIEW_PLAN.md when requested

Validate

  • the plan is grounded in actual code reading, not only the folder names
  • the review order follows actual user or system flows
  • high-risk files are explained, not only listed
  • priorities account for likely mismatch against existing architecture or analogous entities
  • the plan helps a human skip low-value line-by-line review
  • the saved plan is readable without extra chat context

Ralph Loop

Use the Ralph Loop for every task, including docs, architecture, testing, and tooling work.

  1. Brainstorm first (mandatory):
    • analyze current state
    • define the problem, target outcome, constraints, and risks
    • generate options and think through trade-offs before committing
    • capture the recommended direction and open questions
  2. Plan second (mandatory):
    • write a detailed execution plan from the chosen direction
    • list final validation skills to run at the end, with order and reason
  3. Execute one planned step and produce a concrete delta.
  4. Review the result and capture findings with actionable next fixes.
  5. Apply fixes in small batches and rerun the relevant checks or review steps.
  6. Update the plan after each iteration.
  7. Repeat until outcomes are acceptable or only explicit exceptions remain.
  8. If a dependency is missing, bootstrap it or return status: not_applicable with explicit reason and fallback path.

Required Result Format

  • status: complete | clean | improved | configured | not_applicable | blocked
  • plan: concise plan and current iteration step
  • actions_taken: concrete changes made
  • validation_skills: final skills run, or skipped with reasons
  • verification: commands, checks, or review evidence summary
  • remaining: top unresolved items or none

For setup-only requests with no execution, return status: configured and exact next commands.

Load References

  • read references/review-plan-format.md for the output shape
  • read references/risk-signals.md when deciding what deserves human attention first

Example Requests

  • "Plan a human review for this 40K-line AI-generated feature."
  • "I cannot review every file. Tell me what to inspect first."
  • "Trace the signup-to-completion flow and save a HUMAN_REVIEW_PLAN.md."
  • "Look through the generated folder, give me two separate prioritized review lists, and save them under docs for this area."

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.