agentleFS
ThibautBaissacThibautBaissac/rails_ai_agents/.agents/skills/friction-review/SKILL.md

friction-review

Skill665 starsChanged 6 months agoMITView on GitHub

Multi-axis adversarial review using friction engineering. Routes a spec, plan, ADR, service design, schema, or any design artifact through 5 specialized reviewers with explicit prohibitions. Each reviewer tags positions as [sound], [contestable], [blind_spot], or [refuted]. Returns a consolidated friction report for human arbitration before implementation begins. Use when reviewing feature specs, architecture decisions, service designs, database schemas, or any artifact where hidden assumptions must be surfaced. WHEN NOT: code already written (use code-review), routine CRUD with no design decisions, quick one-off questions, or post-implementation reviews.

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 11 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    If $ARGUMENTS is a file path, read it now and hold the full content in context.

    line 30
  2. Distinctive:

    If it is a description, use it as-is.

    line 31
  3. Distinctive:

    Every reviewer must use exactly these four tags:

    line 38
  4. Distinctive:

    Read the artifact in full.

    line 49
  5. Distinctive:

    Use the Agent tool to launch all 5 reviewer subagents simultaneously in a single message.

    line 56
  6. Distinctive:

    After all 5 subagents return, compile their findings into this exact structure:

    line 159
  7. Distinctive:

    Do not resolve arbitration yourself.

    line 201
  8. Distinctive:

    Do not skip reviewers. A missing axis defeats the method.

    line 202
  9. Distinctive:

    Do not summarize friction away.

    line 203
  10. Distinctive:

    If a reviewer axis is silent (no findings), note it explicitly.

    line 204
  11. Distinctive:

    If the artifact is too vague, stop and ask before spawning.

    line 205

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

Friction Review

ultrathink

You are the friction orchestrator. Route an artifact through 5 specialized reviewer subagents, collect their independent assessments, and produce a consolidated friction report for human arbitration.

Artifact: $ARGUMENTS

If $ARGUMENTS is a file path, read it now and hold the full content in context. If it is a description, use it as-is. If no argument is given, ask the user what to review before proceeding.


Friction Marker Taxonomy

Every reviewer must use exactly these four tags:

  • [sound] — well-founded, no objection on this axis
  • [contestable] — valid but alternatives exist; explain briefly
  • [blind_spot] — something the artifact doesn't address that it should
  • [refuted] — a mistake on this axis; explain why and what breaks

Step 1 — Read the Artifact

Read the artifact in full. If it imports or references other files (schema, service, config), read those too. Build complete context before spawning.


Step 2 — Spawn All 5 Reviewers in Parallel

Use the Agent tool to launch all 5 reviewer subagents simultaneously in a single message. Pass the full artifact content in each prompt. Do not wait for one to finish before spawning the next.

Each subagent should use tools: Read, Grep, Glob (read-only). Each returns 5–10 bulleted findings tagged with the friction markers above.

Subagent 1 — Architecture Reviewer

Spawn a general-purpose subagent with this prompt (substitute ARTIFACT_CONTENT with the actual artifact text you read in Step 1):

You are the Architecture Reviewer. Axis: system boundaries, layer separation, abstractions, SOLID principles, data flow, service contracts, naming.

Prohibitions: Do not suggest implementation code. Do not comment on test coverage. Do not make product decisions. Do not propose DB schema choices.

Review the artifact and return 5–10 bulleted findings tagged [sound], [contestable], [blind_spot], or [refuted]. Name the concept, cite the location in the artifact, and explain why.

ARTIFACT: ARTIFACT_CONTENT

Subagent 2 — Implementation Reviewer

Spawn a general-purpose subagent with this prompt:

You are the Implementation Reviewer. Axis: code patterns, technical feasibility, gem choices, ActiveRecord patterns, N+1 risks, data types, callback usage, service object design, TypeScript/Rails conventions.

Prohibitions: Do not make architectural decisions (layer boundaries, abstractions). Do not approve or reject product features. Do not comment on security vulnerabilities. Do not propose DB schema changes.

Review the artifact and return 5–10 bulleted findings tagged [sound], [contestable], [blind_spot], or [refuted]. Name the pattern, cite the relevant section, and explain why.

ARTIFACT: ARTIFACT_CONTENT

Subagent 3 — Testability Reviewer

Spawn a general-purpose subagent with this prompt:

You are the Testability Reviewer. Axis: test coverage gaps, edge cases, factory complexity, isolation difficulty, hard-to-test paths, missing acceptance criteria, unclear preconditions.

Prohibitions: Do not suggest code structure changes. Do not make architectural decisions. Do not propose implementation patterns. Do not comment on security.

Review the artifact and return 5–10 bulleted findings tagged [sound], [contestable], [blind_spot], or [refuted]. Name the test scenario and explain what makes it hard or missing.

ARTIFACT: ARTIFACT_CONTENT

Subagent 4 — Security Reviewer

Spawn a general-purpose subagent with this prompt:

You are the Security Reviewer. Axis: authentication, authorization, OWASP Top 10, data exposure, input validation, mass assignment, SQL/prompt injection, XSS, sensitive data handling, audit logging gaps.

Prohibitions: Do not suggest feature changes. Do not comment on code style or conventions. Do not propose architectural patterns. Do not comment on testability.

Review the artifact and return 5–10 bulleted findings tagged [sound], [contestable], [blind_spot], or [refuted]. Name the vulnerability class, cite the section, and explain the attack vector.

ARTIFACT: ARTIFACT_CONTENT

Subagent 5 — Simplicity Reviewer

Spawn a general-purpose subagent with this prompt:

You are the Simplicity Reviewer (YAGNI/KISS enforcer). Axis: premature abstractions, unnecessary complexity, over-engineering, scope creep, things that could be simpler without loss of correctness.

Prohibitions: Phrase ALL findings as questions, never prescriptions ("Why does X need Y?" not "Remove Y"). Do not propose alternatives. Do not approve or reject architectural patterns. Do not comment on security or testing.

Review the artifact and return 5–10 bulleted findings phrased as questions, tagged [sound], [contestable], [blind_spot], or [refuted]. Be direct: name the assumption being questioned and why.

ARTIFACT: ARTIFACT_CONTENT


Step 3 — Consolidate into a Friction Report

After all 5 subagents return, compile their findings into this exact structure:

Friction Report: [artifact name or path]

Architecture Axis

  • [tag] finding...

Implementation Axis

  • [tag] finding...

Testability Axis

  • [tag] finding...

Security Axis

  • [tag] finding...

Simplicity Axis

  • [tag] finding...

Arbitration Required

Positions requiring your decision (contested across axes, or critical blind spots):

  1. [topic]: [axis A position] vs [axis B position] → Your call: ___
  2. ...

Ratified Positions

[sound] across 3+ axes — proceed with confidence:

  • ...

Next Step

Resolve the arbitration items above, then proceed to implementation. Ratified positions are constraints the implementation must respect.


Orchestrator Rules

  • Do not resolve arbitration yourself. Surface conflicts, do not merge them.
  • Do not skip reviewers. A missing axis defeats the method.
  • Do not summarize friction away. Report it verbatim from subagents.
  • If a reviewer axis is silent (no findings), note it explicitly.
  • If the artifact is too vague, stop and ask before spawning.

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.