Software Development Skills

Agentic Implementation Optimizer — audit existing agentic AI implementations for reliability/cost/architecture issues, walk through creating new agentic implementations from scratch, or extend existing implementations with new capabilities. Use when: '/aio', 'audit my agent', 'build an agent', 'agentic review', 'extend this agent', 'agent architecture', 'optimize my agent'. Covers architecture patterns, reliability engineering, guardrails, cost optimization, observability, escape-hatch/abstention design, gap-signal capture, and production readiness.

Use when the user wants to learn a topic deeply — asks for a curriculum, learning path, self-study plan, study guide, syllabus, or says teach me X. Generates a structured, adaptive curriculum directory of markdown modules (ELI5 → Core Concepts → Deep Dive → Supporting Material → Understanding Check) plus an append-only JSONL assessment log that adapts future modules to the learner's answers. Tailors depth and emphasis to the learner's long-term goal.

Audit a personal open-source repo for public-readiness before going public or pushing to its public git history — secrets/PII in history, licensing, README & community-health files, code quality, tests/CI, deps & releases. Produces a scored report (SHIP IT / NEEDS POLISH / NOT READY) with exact fix commands. Use when: '/debut', 'is this repo ready to go public', 'oss readiness audit', 'public-ready check', 'audit my repo before open-sourcing', 'is this presentable', 'pre-publish audit'. Adaptive: full-repo readiness sweep or lighter pre-push diff mode. Draft-only, flag-only — never commits, never auto-fixes, never rewrites history.

Use when a database query or query-backed endpoint is slow — this query is slow, the page takes forever, DB is pegged, find the slow queries, why is this endpoint slow, N+1 suspicion, query timeouts, high p99 latency, EXPLAIN or plan analysis, a slow Snowflake/dbt model, or a query that used to be fast. Routes to one of five diagnostic modes (analyze a known query, locate an unknown offender, triage a live incident, audit proactively, or chase a regression), runs the right toolkit (EXPLAIN ANALYZE, pgstatstatements, pgstatactivity, APM/DBM tracing, Snowflake Query Profile), and produces a root-caused fix with before/after verification.

Self-paced run-loop dispatcher that drives the Foundry V0 autonomous-fix pipeline ONE unit per iteration against a selected repo (repo-agnostic via repos.json profiles). Each tick it calls the deterministic foundrytick.sh engine to decide the next action, then spawns a capable Worker to implement a unit or an independent capable Integrator to review-and-route a finished one — integrate-before-implement, strictly serial (cap=1), no auto-retry, and NEVER pushing to a remote. Stops when the bucket is drained and prints an end-of-run summary for you to review and push by hand. Use when - "/foundry-run", "run the foundry loop", "drive the foundry pipeline", "start the fix loop", "kick off foundry V0", or whenever a Foundry bucket should be drained one unit at a time.

Use when reviewing or auditing a knowledge graph, ontology, or schema-for-an-ontology for structural health before bad modeling corrupts inference. Triggers — /ontology-review, 'review the ontology/graph', 'check my schema for an ontology', 'audit the memory graph structure', or deciding whether a set of concepts is modeled correctly. Runs a structured seven-axis review (orthogonality, granularity, taxonomic hygiene, identity and rigidity, relationship semantics, competency questions, inference safety), grounded in OntoClean meta-properties, the OOPS! pitfall catalogue, Grüninger and Fox competency questions, and Gómez-Pérez consistency/completeness/conciseness dimensions. Produces a severity-ranked findings list led by inference-corrupting issues — each with the smell, the offending concept/edge, why it is a problem, and a concrete fix.

Review database schemas AND in-code data structures for correctness, integrity, performance, scalability, evolvability, simplicity, and design quality. Audits Ecto schemas, Postgres DDL/migrations, dbt models, and application types (Elixir structs/typespecs, TypeScript types). Use when: '/schema-review', 'review this schema', 'review this migration', 'review this data model', 'is this schema sound', 'audit this table design', 'review these types', 'is this over-engineered', 'will this schema scale', 'data structure review', 'check this migration is safe'. Adaptive execution: single-pass for a handful of objects, parallel multi-agent fan-out when the review surface exceeds a few line items. Draft-only — never auto-posts.

Use when rebuilding a scoped section of legacy code clean-room — when the existing implementation is polluting attempts to improve it and a from-scratch rewrite is wanted. Distills the legacy code to pure functional requirements, builds a firewalled greenfield reimplementation that a separate agent designs WITHOUT ever seeing the legacy code, verifies behavioral parity, and optionally cuts over. Triggers on /strangler-fig, clean-room rewrite, rebuild this module from scratch, strangle this legacy code, the old code keeps leading the rewrite astray.

Review Claude Code skills, hooks, CLAUDE.md files, scripts, or workflow configurations for design flaws, unclosed loops, stale references, side effects, security risks, and portability issues. Use when: 'review this skill', 'audit this hook', '/util-review', 'check this config', or when evaluating any supplementary Claude Code artifact for quality and correctness.