sentinel

Process a large task by breaking it into work items and grinding through them with sub-agents. For tasks too large for one context window — codemap generation, mass migration, bulk documentation.

Scaffold the Sentinel vault, workflows, and quality gates for a new project. Run once per project.

Enable/disable optional Sentinel hooks and configure thresholds.

Audit all context sources — CLAUDE.md, rules, MCP servers, plugins, hooks, vault — with token estimates and optimization recommendations.

Diagnose and fix common Sentinel installation issues — missing files, broken hooks, stale state.

Copy Sentinel plugin contents (hooks, agents, skills, rules) into your project for full customization. After ejecting, the plugin is no longer needed.

Show Sentinel system health — vault status, hook activity, open investigations, stale entries, and coverage gaps.

Run a task repeatedly until a completion condition is met or max iterations reached. For convergence tasks like fixing lint errors, improving test coverage, or tuning prompts.

Guided team onboarding for new Sentinel users

Deep vault cleanup — find duplicates, validate cross-references, archive stale entries, and report vault health. Run monthly or before major feature pushes.

Show Sentinel effectiveness metrics — vault health, knowledge reuse, and code discipline over time.

System design advisor. Evaluates architecture decisions, scalability, separation of concerns, and API design.

Build error resolution agent. Reads error output, diagnoses root cause, and applies minimal fixes. Does not refactor or add features.

General code review agent. Reviews code for quality, readability, error handling, naming, structure, and maintainability.

SQL and database specialist. Reviews queries, schema design, migrations, indexes, and data access patterns.

Dead code removal and consolidation agent. Identifies and safely removes unused exports, imports, functions, and variables.

OWASP Top 10 security scanner. Reviews code for injection, XSS, CSRF, hardcoded secrets, auth bypass, SSRF, and other vulnerabilities.

TDD enforcement agent. Guides the RED-GREEN-REFACTOR cycle, verifies test-first discipline, and ensures coverage targets are met.

Frontend review agent combining design system compliance, accessibility, responsive layout, dark mode, and UX heuristics.

Adversarial evaluation protocol for finding flaws in tests, prompts, and code reviews. Activates when validating test quality, auditing prompts, reviewing architecture, or running convergence loops. Enforces multi-round evaluation with 3-round max, mutation testing concepts, and structured issue tracking.

Structured exploration before implementation. Activates when building new features, making design decisions, or facing multiple valid approaches. Prevents jumping to code before understanding intent. Explores context, asks clarifying questions, proposes approaches, and produces a spec.

Universal coding standards and best practices. Activates when writing new code, reviewing code quality, refactoring, or enforcing naming, formatting, or structural consistency. Covers immutability, file organization, error handling, input validation, and naming conventions.

Evaluation harness for testing AI agent prompts and outputs against ground truth. Activates when designing eval scenarios, running prompt evaluations, interpreting results, or iterating on agent behavior. Covers scenario design, expectation definition, ground truth generation, regression testing, and convergence loops.

Universal anti-patterns and test standards. Activates when reviewing test quality, writing tests, or auditing code for common mistakes. Covers the 10 most damaging anti-patterns with BAD/GOOD examples, test validity criteria, and mocking guidelines.

Core methodology for disciplined software development. Activates when investigating bugs, managing knowledge vaults, recovering sessions, or running quality gates. Enforces investigation journals, self-healing knowledge loops, session continuity, and gate-based verification.

Test-Driven Development workflow enforcement. Activates when writing new features, fixing bugs, or refactoring code. Enforces RED-GREEN-REFACTOR cycle with 80%+ coverage, edge case coverage, and anti-pattern avoidance.

23 hooks across 6 events