laiff-claude-marketplace
Personal Claude Code plugin marketplace — research-driven plugins for code review, agent orchestration, and developer workflows.
Installation
/plugin marketplace add Laiff/claude-marketplace
Then install individual plugins:
/plugin install code-review@laiff-claude-marketplace
Plugins
code-review v4.0.0
Evidence-driven code review with multi-agent orchestration, feedback-grounded quality guards, and YAML-structured inter-agent communication.
Skill: /code-review — triggers on: review PR, check PR, code review, audit pull request
Pipeline:
Phase 0 Preflight gate (haiku)
Phase 1 Context gathering: 3 agents in parallel (haiku)
Phase 2 Review: 3 agents in parallel (sonnet)
Phase 3 Evidence verification (sonnet)
Phase 4 Three-layer dedup + verdict (haiku)
Phase 5 Output composition + GitHub posting (sonnet)
11 Agents:
| Agent | Model | Role |
|---|
| preflight | haiku | Gate check — should this PR be reviewed at all |
| context-collector | haiku | Gather CLAUDE.md, conventions, project context |
| pr-summarizer | haiku | Structured PR summary with intent, scope, risk |
| comment-scanner | haiku | Scan existing comments to prevent re-posting |
| convention-checker | sonnet | CLAUDE.md compliance audit |
| bug-detector | sonnet | Logic bugs, type errors, null derefs, resource leaks |
| security-reviewer | sonnet | Security vulnerabilities and architecture |
| evidence-verifier | sonnet | Validate claims, cross-check between agents |
| dedup-orchestrator | haiku | Three-layer dedup, batch guard, budget cap |
| output-composer | sonnet | Final review output, inline comments, PR posting |
| feedback-learner | sonnet | Learn from reactions, minimize future noise |
4 Protocols:
| Protocol | Purpose |
|---|
| agent-communication | YAML-structured inter-agent data exchange |
| finding-schema | Finding object definition with confidence scoring |
| quality-guards | 10 guards preventing noise (G1-G10) |
| review-verdict | APPROVE / REQUEST_CHANGES / COMMENT decision logic |
Quality Guards:
| Guard | Prevents |
|---|
| G1 Noise filter | Trivial or empty comments |
| G2 Scope guard | Flagging unchanged code |
| G3 Dedup guard | Re-posting same finding |
| G4 Batch guard | Same comment on N files |
| G5 Evidence grounding | False external claims |
| G6 Exception clauses | Repo-specific false positives |
| G7 Security calibration | Over-escalating internal services |
| G8 Comment budget | Developer fatigue (max 10 comments) |
| G9 False positive filter | General low-signal noise |
| G10 Verdict calibration | Over-aggressive REQUEST_CHANGES |
Repository Structure
.claude-plugin/
marketplace.json Marketplace catalog
plugins/
code-review/
.claude-plugin/
plugin.json Plugin manifest
agents/ 11 specialized agents
protocols/ 4 inter-agent protocols
skills/
code-review/
SKILL.md Main skill entry point
Author
Laiff
