reaper

When adding a new user-invocable command:

Surface changes since CLAUDE.md last touched down.

Validate .reaper.yml against the schema and check for drift between configured and detected values.

Chart work into flight-ready issues with dependencies mapped.

Detect runners, trackers, and conventions and write a validated .reaper.yml at the repo root.

Subject matter expert on writing, auditing, and optimizing LLM prompts for maximum performance with minimal token usage. Analyzes prompts for anti-patterns across all major frontier models (Claude, GPT, Gemini, Llama, DeepSeek, Mistral) and recommends model-specific token optimization strategies. Examples: <example>Context: User has a bloated system prompt costing too much and wants it optimized. user: "Our main orchestrator prompt is 8000 tokens and our API costs are through the roof - can you optimize it without losing quality?" assistant: "I'll use the ai-prompt-engineer agent to audit the prompt for token waste, redundancy, and anti-patterns, then produce an optimized version with measurable token reduction while preserving critical instructions." <commentary>The user needs prompt token optimization with quality preservation, which is the core expertise of ai-prompt-engineer.</commentary></example> <example>Context: User is building a new agent and wants the prompt written well from the start. user: "I need to write a system prompt for a code review agent - what techniques should I use to get the best results from Claude?" assistant: "I'll deploy the ai-prompt-engineer agent to draft a prompt using Claude 4.x best practices including XML semantic tags, proper instruction specificity, technique selection, and output format control." <commentary>The user needs expert prompt authoring guidance tailored to Claude models, so use ai-prompt-engineer for technique selection and prompt drafting.</commentary></example> <example>Context: An existing agent is producing inconsistent or low-quality output. user: "The security-auditor agent keeps hallucinating vulnerabilities that don't exist - something is wrong with its prompt" assistant: "I'll use the ai-prompt-engineer agent to analyze the security-auditor prompt for anti-patterns like vague instructions, missing grounding constraints, or over-aggressive directives that may cause hallucination." <commentary>The user has a prompt quality problem manifesting as hallucination, which requires prompt-level diagnosis from ai-prompt-engineer.</commentary></example>

Designs REST and GraphQL APIs with OpenAPI specifications, API contracts, versioning strategies, and integration patterns. Examples: <example>Context: User needs to plan a new user management API before development begins. user: "I need to design a user management REST API with authentication, role-based access, and proper versioning strategy" assistant: "I'll use the api-designer agent to create a comprehensive OpenAPI 3.0 specification, define endpoint contracts with request/response schemas, establish versioning and backward compatibility strategy, and plan integration patterns with authentication systems." <commentary>The user is in strategic planning phase needing API design before implementation, so use the api-designer agent to create detailed contracts and versioning strategy.</commentary></example> <example>Context: Team wants to restructure their API with breaking changes. user: "Plan an API versioning strategy for deprecating v1 endpoints and migrating users to v2 with breaking changes" assistant: "Let me use the api-designer agent to analyze the current API surface, design a comprehensive migration strategy, create versioning roadmap with deprecation timelines, and establish backward compatibility guidelines for smooth user migration." <commentary>Since this involves strategic API design decisions and versioning trade-offs, use the api-designer agent for architectural planning.</commentary></example>

Git operations, worktree management, and repository cleanup with safety protocols. Use for branch operations, worktree setup/teardown, safe merging, and repository maintenance. Examples: <example>Context: User needs to start work on a new feature with proper isolation. user: "I need to implement OAuth authentication for PROJ-456" assistant: "I'll use the branch-manager agent to set up an isolated session worktree environment at .claude/worktrees/PROJ-456-oauth with a feature branch, install dependencies, and validate the setup before proceeding with implementation." <commentary>Since the task requires proper git workflow and worktree isolation, use the branch-manager agent to handle all git operations and environment setup before code implementation begins.</commentary></example> <example>Context: After all quality gates pass, code needs to be committed and consolidated for review. user: "Quality gates passed - commit the authentication changes to the review branch" assistant: "I'll use the branch-manager agent to commit the validated changes in the agent worktree, fast-forward merge the agent branch into the session branch, clean up the agent worktree, and report success." <commentary>The branch-manager executes git operations as directed by the orchestrator after quality gates pass.</commentary></example>

Diagnoses and fixes bugs using Test-Driven Development methodology with systematic reproduction and minimal fixes. Examples: <example>Context: User reports a bug where login fails for users with special characters in their email addresses. user: "Users can't log in when their email has a plus sign, like '[email protected]'" assistant: "I'll use the bug-fixer agent to reproduce this issue by writing a failing test case for special character emails, then implement the minimal fix to handle email validation correctly." <commentary>Since this involves reproducing and fixing a specific bug, use the bug-fixer agent to follow TDD methodology: write failing test, implement minimal fix, ensure test passes.</commentary></example> <example>Context: User encounters an error where API responses are occasionally missing data fields. user: "Sometimes our API returns incomplete data - the 'metadata' field is missing randomly" assistant: "Let me use the bug-fixer agent to reproduce this intermittent bug by writing tests for the API response structure and then identify and fix the root cause." <commentary>The user has a bug that needs systematic reproduction and fixing, so use the bug-fixer agent to diagnose the issue with proper test coverage.</commentary></example>

Designs, creates, and refactors specialized agents with clear purposes, proper structure, and adherence to design principles. Serves as quality control and design authority for all agent development. Examples: <example>Context: User needs a new agent for API integration work. user: "Create an agent that handles third-party API integrations with webhooks and event handling" assistant: "I'll use the claude-agent-architect agent to design a comprehensive integration-engineer agent with proper model selection, workflow stage assignment, and complete specifications following the format template." <commentary>Since the user needs a new specialized agent created with proper design standards, use claude-agent-architect to ensure it follows all principles and patterns.</commentary></example> <example>Context: Existing agent has unclear purpose and poor structure. user: "The database-specialist agent is confusing and doesn't follow our standards - can you refactor it?" assistant: "I'll deploy claude-agent-architect to analyze the current agent, identify structural issues, and refactor it with clear sections, proper model assignment, and adherence to design principles." <commentary>The user needs an existing agent refactored to meet standards, so use claude-agent-architect for systematic improvement following established patterns.</commentary></example>

Universal Gate 2 code review protocol for Reaper's quality gate pipeline. Guides a reviewer through plan verification, scope creep detection, completeness checks, and structured JSON output.

Use when working with Beads issues in a Beads-tracked repository. TRIGGER when a Beads ID is referenced (slug-hash form like "myapp-a3f" or child form "myapp-a3f.2"); creating or updating issues; adding dependencies; walking the dependency tree; querying ready or blocked work. Returns the correct dependency semantics, child-issue ID conventions, and ready-queue queries on the first try. SKIP for GitHub Issues, Jira, Linear, or markdown plan files.

Use when working with GitHub Issues or GitHub Projects. TRIGGER when a GitHub issue is referenced (e.g., "#42", "issue 42", "GH-42"); creating, editing, closing, or listing issues; linking sub-issues; updating GitHub Projects v2 fields; resolving labels or assignees on a repo. Returns the correct command shape, parent/child linking pattern, and field-update workflow on the first try. SKIP for pull request operations, repo/release/auth tasks, or work targeting Jira, Linear, Beads, or markdown plan files.

Use when working with Jira issues and work items. TRIGGER when a Jira key is referenced (e.g., "APP-123", "APP-7"); creating, transitioning, updating, linking, or querying parent/child hierarchy of Jira work items; searching across projects with JQL. Returns the correct field mapping, transition IDs, and link-type semantics — direct calls without it routinely produce malformed updates or wrong status transitions. SKIP for GitHub Issues, Linear, Beads, or markdown plan files.

Use when working with Linear issues. TRIGGER when a Linear issue is referenced (e.g., "TEAM-123", "issue TEAM-25"); creating, transitioning, updating, linking, or querying parent/child hierarchy; resolving team, status, or label IDs needed to save an issue. Returns the right ID-resolution sequence (team → status → label) and the correct hierarchy mechanism — direct calls without it routinely create orphan issues or rejected updates. SKIP for GitHub Issues, Jira, Beads, or markdown plan files.

1 hook across 1 event