Agent-Skills.md

Agent Skills: MCAF: Human Review Planning

Plan a human review 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.

UncategorizedID: managedcode/MCAF/mcaf-human-review-planning

Repository

managedcode/MCAF
managedcodeLicense: CC-BY-4.0
526

Install this agent skill to your local

pnpm dlx add-skill https://github.com/managedcode/MCAF/tree/HEAD/skills/mcaf-human-review-planning

Skill Files

Browse the full folder contents for mcaf-human-review-planning.

Download Skill

Loading file tree…

skills/mcaf-human-review-planning/SKILL.md

Skill Metadata

Name
mcaf-human-review-planning
Description
"Prioritize human review of a large generated code area by tracing real flows and ranking risky files. Use when a reviewer cannot inspect every line and needs a practical review sequence."

MCAF: Human Review Planning

Read root/local AGENTS.md and the requested area. Keep this pass read-only unless the user requests a saved plan or implementation.

Workflow

  1. Trace the actual user or system lifecycle from entry points through validation, permissions, persistence, integrations, side effects, and completion. Include immediate callers and contracts needed to understand it.
  2. Compare new behavior with established architecture, analogous entities, and existing invariants. Prioritize state corruption, authorization bypass, irreversible effects, retry/idempotency mistakes, and integration drift.
  3. Produce two separate lists: the ordered review flow, and priority files/modules. For each item, name the concrete risk and what the reviewer should verify.
  4. Distinguish inspected evidence from assumptions and unexamined paths. A priority plan is not proof that the code has no defects.
  5. Present both lists in chat. Save only when requested, at the user's exact path or the nearest relevant docs/ area.

Do not invent a universal signup/create/update sequence; follow the actual system. Do not replace human prioritization with a generic checklist or claim completed verification.

References