File Classification for Review Skill

File Classification for Review

Invoke before deciding SKIP on any prose file. Three tiers determine which SKIP rules apply.

Tier Decision

flowchart TD
    File([Changed file]) --> Q1{"Is it prose?<br>(markdown, text, config with prose fields)"}
    Q1 -->|"No — pure code"| Code["Standard review — apply perspective normally"]
    Q1 -->|Yes| Q2{"Does any part describe or influence<br>a process or function?"}
    Q2 -->|"No — pure reference data,<br>changelogs, release notes"| T1["Tier 1 — Documentation Only<br>SKIP permitted for all perspectives"]
    Q2 -->|"Yes — guides behavior, defines<br>constraints, describes a workflow"| Q3{"Primary executor or<br>audience is an AI / LLM / agent?"}
    Q3 -->|"No — human-facing<br>CONTRIBUTING.md, ADRs, runbooks"| T2["Tier 2 — Process Documentation<br>SKIP with explicit skip_reason only"]
    Q3 -->|"Yes — agent files, SKILL.md,<br>CLAUDE.md, rules/*.md, prompts"| T3["Tier 3 — LLM Prompt Engineering Artifact<br>See rules below"]

Tier Rules

Tier 1 — Documentation Only

Changelogs, release notes, README sections describing completed features. SKIP is valid for all perspectives with no applicable checks.

Tier 2 — Process Documentation

CONTRIBUTING.md, ADRs, runbooks, README workflow sections. These are behavioral contracts for human contributors. SKIP is permitted only with an explicit skip_reason explaining why the change has no impact in this reviewer's scope.

Tier 3 — LLM Prompt Engineering Artifacts

The markdown content IS the executable. These files are the product, not descriptions of it.

Tier 3 file patterns:

agents/*.md — plugin agent instruction files
skills/*/SKILL.md — skill instruction files
skills/*/references/** — skill reference files (any depth)
CLAUDE.md — session instruction files
.claude/rules/*.md — scoped behavioral rules

Per-perspective SKIP rules for Tier 3:

| Perspective | SKIP | Required check | |---|---|---| | Security | PROHIBITED | Prompt injection surfaces — see §2.5.1 in verdict-schema.md | | Quality | PROHIBITED | Behavioral correctness: contradictions, ambiguous constraints, missing edge cases | | Performance | Permitted | No applicable performance check | | Accessibility | Permitted | No applicable accessibility check |

For the full classification rule definition and the prompt injection surface checklist (§2.5.1), read: verdict-schema.md §2.5

Agent Skills: File Classification for Review

Install this agent skill to your local

Skill Files