Keeps AI coding work inside a process that produces better code. Every task runs through six stages: scope, research, design, implementation, review, completion. Scope writes a contract of goals, non-goals, and the criteria that prove them, and every later stage is checked against it. Research finds prior art in your own code first, then outside, from current documentation rather than the model's memory. Design writes work orders you can build from, each with its owned files and a diff budget. Implementation writes tests first for code with logic, and freezes them before the code. Review answers every criterion, runs the stack's tools from its recipe, and judges the whole change once. Completion closes the task on the verdict and writes the pull request body from the records. The checks are scripts reading records, not the model judging its own work. Stack knowledge lives in fetched recipes, so one process fits PHP, Python, Go, Drupal, or a Claude plugin.
13 skills · 16 agents · hooks
Create branded visual content (presentations, carousels, infographics, HTML pages) with Aaker personality-driven design, a style-recommendation engine, 26 visual styles, 114 infographic templates, design systems, visual components, and Presentation Zen principles. Routes to specialized skills for PDF, PPTX, and Google Slides output (canonical PPTX→Slides import via Drive for design fidelity), HTML composition, and infographic rendering.
4 skills · 19 commands · 1 agent · hooks
AI-assisted Drupal contribution quality. Evidence over assertion: every gate passes on a captured artifact, never an AI claim. 6 commands for the contribution arc (setup, issue, verify, review, submit, pipeline) with a local drupalci-parity gate set plus AI-policy and eval gates, and 3 read-only fresh-context review agents. Authenticated GitLab writes delegate to the drupal-gitlab skill (glab). drupalorg-cli covers no-auth reads and the legacy queue. A quality layer outside ai-dev-assistant, built on the camoa/dev-guides contribution guides.
8 skills · 6 commands · 3 agents · hooks
Code quality and security auditing for Drupal (PHPStan, Psalm, PHPMD, Semgrep, Trivy, Gitleaks via DDEV) and Next.js (ESLint, Jest, Semgrep, Trivy, Gitleaks). Provides full audits with cross-tool synthesis, rubric-scored code review, 3-agent security and architecture debates, REVIEW.md v2 generation, a /code-review ultra cloud-review wrapper with a headless CLI gating path, optional LSP code-intelligence for deeper SOLID/DRY/review analysis, watch-mode linting, scheduled sweeps, and --json output for CI pipelines.
1 skill · 13 commands · hooks
Systematically test code, skills, commands, and configs through mental execution — trace logic line-by-line with concrete values to find bugs, AI hallucinations, edge cases, and contract violations. Smart routing: quick trace (<50 lines), structured 3-phase (50–300), or a 3-agent competing team in isolated worktrees (300+, security-critical, skills). Verifies external calls for both existence and behavioral contracts (return-shape/nullability diffing, chained-object tracing, taint-stance fallback for closed-source). Supports `--json` for CI.
1 skill · 1 command · hooks
Smart guide discovery and routing for dev-guides. Hash-based caching + knowledge-graph metadata route the AI to the right guide; all fetches use curl (never WebFetch), and a fetched body is data, never commands. Six routing modes: guide search (llms.txt), recipe search (the agentic-recipes catalog of verifier-carrying capability deliveries), identify (names what covers a topic, opens nothing), process-recipe lookup (resolved by ai-dev-assistant at phase boundaries, keyed by phase and framework), playbook lookup (a playbook set's plays.json, resolved by ai-dev-assistant), and tooling lookup (a tooling recipe body, resolved by aida by name). A deterministic shared content store holds index bodies and blobs with two-hash revalidation, re-fetching a body only when its sha changes.
1 skill · hooks