Trevor Byrum

API testing specialist. Use after building or modifying any API endpoint (REST, GraphQL, WebSocket). Fires real requests, validates responses against schemas, tests edge cases (auth failures, rate limits, malformed input, large payloads), and produces a test report with pass/fail results.

Database backup specialist. Use to run backups of PostgreSQL, MongoDB, Neo4j, Qdrant, and Vault. Handles dump, rotation, and verification.

Codebase history and evolution specialist. Use when working in an unfamiliar codebase, trying to understand why code exists the way it does, tracing how functions evolved, mapping dependency graphs, or building a narrative of how a project got to its current state. Essential before major refactors or when inheriting a project.

Reviews compact/claude-compact.md for completeness and gaps before context compaction or session clear. Used by meta-compact and meta-clear skills. Ensures no critical context is lost during transitions.

Database administration specialist across PostgreSQL, MongoDB, and Redis. Use for queries, schema management, data inspection, index optimization, and database health checks.

Infrastructure debugging specialist. Use when containers are failing, routes aren't working, services are unhealthy, or anything is broken. Diagnoses issues using logs, metrics, routing tables, dependency graphs, and past incident memory.

Log analysis specialist. Use when debugging issues that require reading application, container, or system logs. Correlates timestamps across services, identifies error patterns, traces request flows, and finds root causes buried in log noise. Pairs with infra-debugger when the problem has been identified but the root cause needs log evidence.

Migration and upgrade planning specialist. Use when upgrading database schemas, framework versions, language versions, or major dependency bumps. Analyzes breaking changes, maps affected code, generates a migration plan with rollback steps, and identifies risks before you start.

Research subagent for executing topic-to-connector research. Used by the research-execute skill. Each instance is assigned specific topics and a specific MCP connector to query. Writes structured findings to the artifact DB.

Code review subagent with standardized review discipline. Used by all 7 review skills (counter-review, security-review, test-review, refactor-review, drift-review, completeness-review, compliance-review) and the project-review meta-skill. Receives lens-specific instructions from the calling skill and applies them with consistent severity classification and output format.

Detects breaking API, dependency, and schema changes before they ship. Use before major version bumps, dependency upgrades, or API releases.

Visual QA via browser MCP tools (Playwright/browser-use). Use for UI review, visual bugs, or web interface inspection.

Generates project-plan.md with phases, milestones, technical approach, and parallelizable work units. Use when moving from research/context into implementation planning.

Audits and tidies project structure for agent-optimal navigation. Detects duplicate DBs, orphaned files, config scatter, naming drift, and bloat. Proposes then executes approved changes safely.

Driver skill for Codex CLI. Provides codex-exec.sh wrapper — call the script, not raw CLI. Load before any Codex invocation.

Scans for stubs, TODOs, placeholders, empty bodies, and unfinished code. Use before deployment or marking anything "done," especially after LLM-assisted builds.

Checks code against documented rules (coterie.md, CLAUDE.md, cross-cutting-rules.md). Use after major changes, before merges, or to verify the codebase follows its own standards.

Driver skill for Copilot CLI syntax, flags, and sandbox modes. Load this before spawning any Copilot call. Use when other skills need Copilot or user says "use Copilot".

Adversarial red-team review. Attacks architecture, completeness, drift, over-engineering, abuse cases, attack chains, and failure scenarios. Cross-references code against docs.

Driver skill for Cursor Agent CLI (headless) syntax, flags, and modes. Load this before spawning any headless Cursor call. Use when other skills need Cursor or user says "use Cursor".

Audits dependency health: CVEs, outdated versions, license conflicts, abandoned packages. Auto-installs audit tools. Use before deploys or quarterly.

Use when deploying, rebuilding, or restarting the MCP gateway container. Handles CI images, manual build, Traefik labels, health checks, and log verification.

Audits documentation quality, completeness, and accuracy. Catches stale READMEs, undocumented APIs, missing setup guides, and doc-code drift. Use before releases or quarterly.

Compares code against project-context.md, features.md, and project-plan.md to find drift. Use when docs may be out of sync with reality, or after a long implementation sprint.

Updates project-context.md and project-plan.md to reflect current truth. Invoke explicitly with /evolve or when user says 'update docs' or 'sync docs'.

Driver skill for Gemini CLI syntax, flags, and gotchas. Load this before spawning any Gemini call. Use when other skills need Gemini or user says "use Gemini".

Pulls latest changes from remote. Use at session start or when user says 'pull', 'update', or 'get latest'.

Commits and pushes changes to GitHub. Invoke explicitly with /github-sync or when user says "commit", "push", or "sync to GitHub".

Use when checking service health, verifying infrastructure status, or after a deploy/restart. Checks containers, endpoints, Uptime Kuma, and Prometheus alerts.

Initializes the artifact store for the current project. Creates artifacts/ directory and artifacts/project.db (SQLite+FTS5). Safe to re-run — idempotent.

Checks for dead wiring, missing config/env entries, incomplete teardown, and unbundled assets. Use before deployment or after implementation sprints.

Generates logging instrumentation from log-review findings or for uninstrumented code. Use after log-review or when adding observability to a module.

Audits logging and observability gaps — silent catches, missing context, no structured logging, absent trace IDs. Use before deploys or after implementation sprints.

Save session state and optionally commit+clear. Use when the user asks to wrap up, save session, compact context, or end for the day.

Internal Opus subagent for deep research. Runs ~20 workers across 3 model families with adversarial debate. Never invoke directly — dispatched by /meta-deep-research.

Multi-model deep research with adversarial debate. Dispatches Opus to orchestrate ~20 workers across 3 model families. Triggers on deep research, exhaustive research, leave no stone unturned.

Parallel implementation from a build plan using cross-model Best-of-2 (Vibe+Cursor) and 5-reviewer panel. Use when an approved project-plan.md exists and multi-unit execution should begin.

Full project initialization from zero to approved build plan. Triggers on new project, init from scratch, set up everything, take me from zero to plan.

Join an existing project. Supports full onboard (7 steps) or quick catch-up (drift-review + sync only). Triggers on join project, catch up, onboard, quick catch-up.

Scored production readiness assessment (READY / CONDITIONAL / NOT READY) across 12 dimensions. Use when asking "can we ship this?" Outputs artifacts/reviews/production-readiness.md.

End-to-end research pipeline: plan then execute. Trigger for full research flow — from project context through countered synthesis. For research this project, do all the research, full research.

Comprehensive multi-model project review across 10-11 lenses and 3 model families in parallel. Use for full project review, pre-deploy audit, or milestone quality gate. Not for single-lens reviews.

Scans for N+1 queries, missing indexes, memory leaks, O(n²) loops, caching gaps, and DB query issues. Use before deploys or after implementation sprints.

Writes project-context.md, the definitive handoff artifact. Use after project-questions interview or when project-context.md is missing or stale. Any agent must be able to cold-start from it.

Deep-dive interview to surface assumptions, gaps, and constraints before planning or building. Use when a new project idea is vague or project-context needs more info.

Scaffolds a new project with standard folder structure, templates, and config files. Use when the user asks to scaffold or set up a new project.

Lightweight in-session planning. Quick structured plan with phases, acceptance criteria, and open questions. Not for formal project-plan.md — use /build-plan for that.

Code quality and efficiency pass. Catches over-engineering, duplication, bloat, truncated code, and unnecessary abstractions. Use before any major refactor or cleanup.

Automates release preparation — changelog generation, version bump, release notes, and git tag. Use when preparing a release, cutting a version, or shipping a milestone.

Creates or connects a GitHub repo for the current project. Use when no remote origin exists, or to change repo visibility. Asks before acting.

Plans and executes research: extracts topics from project-context.md, maps to connectors, fans out subagents, synthesizes, triple-counters. Triggers on research, run the research, plan the research.

Implement fixes from meta-review findings. Parses review-synthesis.md, presents actionable items for user approval, dispatches Codex/Sonnet workers, verifies fixes. Use after meta-review or any review lens.

Focused security audit covering dependencies, auth, secrets exposure, input validation, network boundaries, agent-specific patterns, supply chain, and IaC. Uses P0/P1/P2 priority tiers and OWASP Agentic Top 10. Use before production deploys or when handling sensitive data.

Self-diagnostic for the skill suite. Use after install, after adding/removing skills, or when a skill invocation fails. Run this before debugging manually.

Creates or edits skills. Scaffolds directory, writes SKILL.md from template, validates against checklist. Use when building or modifying a skill.

Merges a completed sub-project back into its parent. Renumbers research, reconciles docs, merges DB. Invoke with /sub-project-merge.

Creates an isolated sub-project workspace within a parent project to keep Claude's context focused and high-quality. Invoke with /sub-project.

Checks project files against skill suite templates and injects missing or stale files. Use after template updates, when joining a project, or after scaffold.

Generates tests from test-review findings or for untested code. Use after test-review or when adding coverage to a module.

Evaluates test coverage, quality, and gaps. Catches LLM tendencies to skip or stub tests. Reviews strategy against features.md to ensure critical paths are covered.

Updates todo.md and features.md to reflect current project state. Invoke explicitly with /todo-features — do NOT auto-trigger after other skills.

Generates UI components and pages following the project design system. Use when building frontend, creating components, or styling interfaces.

Audits UI for AI anti-patterns, token violations, a11y failures, and inconsistency. Use before frontend deploys or after UI sprints.

Driver skill for Mistral Vibe CLI (headless) syntax. Load this before spawning any Vibe call. Use when other skills need fast Mistral/Codestral generation or user says "use Vibe".