Skip to content

feat: add persona-review — multi-persona adversarial review skill#1420

Open
YuhaoLin2005 wants to merge 8 commits into
anthropics:mainfrom
YuhaoLin2005:feature/persona-review
Open

feat: add persona-review — multi-persona adversarial review skill#1420
YuhaoLin2005 wants to merge 8 commits into
anthropics:mainfrom
YuhaoLin2005:feature/persona-review

Conversation

@YuhaoLin2005

Copy link
Copy Markdown

What

A new skill for multi-persona adversarial review using real experts with documented, verifiable philosophies — not abstract roles.

Why

Existing review skills use generic roles ("Security Auditor", "New Hire"). Abstract roles produce generic feedback ("consider adding error handling"). Named personas with known principles produce specific, actionable findings.

This was tested on a real PR review (alirezarezvani/claude-skills#866): 3 rounds, 21 findings. The fixed pool found 18 issues. A random-pool web-search round caught 3 issues both fixed-pool rounds missed. The maintainer merged with Co-authored-by attribution and noted it "filled a real gap."

What it adds

Feature Detail
Fixed pool 9 sourced personas (Hickey, Carmack, Norman, Schell, Wardley, Few, Abramov, Sierra, Tam)
Random pool Web-search for perspectives outside fixed coverage
Anti-fabrication Every attribution carries confidence level; unverifiable claims rejected
Tier system T1 Quick (1-2 reviewers) / T2 Standard (3-4) / T3 Deep (multi-round)
Manager curation McCord (code/docs) or Catmull (creative/strategy) assembles panels

Integration

Complements — does not replace — existing skills:

  • code-reviewer: adds multi-perspective blind-spot coverage
  • adversarial-reviewer: adds named personas with verifiable philosophies
  • security-review: adds architecture-level security perspective

Structure

Follows repo conventions: skills/persona-review/SKILL.md, MIT license, kebab-case naming.

1 file, 188 lines. No dependencies.

YuhaoLin2005 and others added 8 commits June 27, 2026 08:01
…ead See-Also, clarify Hard Constraints, expand Verification
…h verifiable philosophies

A review methodology that uses real experts with documented philosophies
(Hickey, Carmack, Norman, Schell, etc.) instead of abstract roles. Manager
persona assembles panels. Cross-review catches what single reviewers miss.

Key features:
- Fixed pool: 9 sourced personas with verifiable principles
- Random pool: web-search for perspectives outside fixed pool coverage
- Anti-fabrication discipline: every attribution carries confidence level
- 3-tier depth system (T1 Quick / T2 Standard / T3 Deep)
- Manager-curated panel composition with round rotation

Tested on alirezarezvani/claude-skills PR anthropics#866: 3 rounds, 21 findings.
Fixed pool found 18 issues. Random pool caught 3 issues both fixed-pool
rounds missed. Merged with Co-authored-by attribution.

MIT licensed. Complements code-reviewer and adversarial-reviewer.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant