foundry

Documentation specialist for writing docstrings, API references, and README files. Use for auditing missing docstrings, writing Google-style docstrings from code, creating or updating README content, and finding doc/code inconsistencies. NOT for CHANGELOG entries or release notes (use oss:shepherd for lifecycle/format decisions, /oss:release skill for automated generation), NOT for linting code examples (use foundry:linting-expert), NOT for implementation code (use foundry:sw-engineer).

Static analysis and tooling specialist for Python. Use for configuring ruff rules, mypy strictness, pre-commit hooks, fixing lint/type violations, adding missing type annotations to Python source files, and defining the lint/type tool content of quality gates. Handles final code sanitization before handover. NOT for CI pipeline structure, runner strategy, or workflow topology (use oss:ci-guardian), NOT for writing test logic (use foundry:qa-specialist), NOT for implementation fixes beyond annotation/style (use foundry:sw-engineer).

Performance engineer for profiling and optimizing CPU, GPU, memory, and I/O bottlenecks. Use for profiling Python/ML workloads, identifying DataLoader bottlenecks, applying mixed precision, vectorizing loops, and tuning PyTorch throughput. Profile-first — always measures before changing. NOT for general code refactoring (use sw-engineer), NOT for architectural redesign (use solution-architect).

QA specialist for writing, reviewing, and fixing tests. Use for writing new pytest tests, analyzing test coverage gaps, building edge-case matrices, fixing failing tests, and integration test design. Writes deterministic, parametrized, behavior-focused tests. NOT for linting, type checking, or annotation fixes (use linting-expert), NOT for production implementation (use sw-engineer).

Claude Code configuration quality reviewer and improvement coach. Scope: Claude config markdown files only — agents, skills, rules (*.md). Use after editing any agent or skill file to audit verbosity, duplication, cross-reference integrity, structural consistency, content freshness, and agent-roster overlap. Reviews whether roles are still distinct enough to keep, should gain sharper boundaries, or should be merged/pruned. Returns a prioritized improvement report with file-level and roster-level recommendations. Runs on opusplan for best reasoning quality. NOT for hook files (*.js) — those belong to sw-engineer.

System design specialist for ADRs, API surface design, interface specs, migration plans, component diagrams, and hypothesis architectural feasibility assessment. Use for evaluating architectural trade-offs, designing public API contracts, planning deprecation strategies, and filtering AI-generated hypotheses against codebase constraints — reads code and produces specs only. NOT for writing implementation code (use sw-engineer), NOT for release management (use oss:shepherd).

Senior software engineer for writing and refactoring Python code. Use for implementing features, fixing bugs, TDD/test-first development, SOLID principles, type safety, and production-quality Python for OSS libraries. NOT for writing docstrings or docs content (use doc-scribe), configuring ruff/mypy/pre-commit (use linting-expert), system design decisions (use solution-architect), test quality analysis (use qa-specialist), performance profiling and optimization (use perf-optimizer), implementing methods from ML papers / designing ML experiments (use research:scientist), or editing .claude/ config files — agents, skills, hooks, settings, CLAUDE.md (use self-mentor).

Fetches web pages, API docs, and external package/release information for use by orchestrators and other agents. Specializes in package version lookups, GitHub release extraction, and documentation scraping. NOT for code analysis or implementation (use sw-engineer), NOT for ML paper analysis or experiment design (use research:scientist), NOT for writing or auditing docstrings (use doc-scribe), NOT for dependency upgrade lifecycle decisions (use oss:shepherd).

Full-sweep quality audit of .claude/ config — cross-references, permissions, inventory drift, model tiers, docs freshness. Two mutually exclusive action modes: 'fix [high|medium|all]' auto-fixes at the requested severity level; 'upgrade' applies docs-sourced improvements with correctness verification and calibrate A/B testing for capability changes.

Iterative brainstorming skill for turning fuzzy ideas into approved tree documents. Diverges into branches, deepens and prunes them over many rounds, saves a tree doc. Run breakdown on the tree to distill it into a spec via guided questions.

Calibration testing for agents and skills. Generates synthetic problems with known outcomes (quasi-ground-truth), runs targets against them, and measures recall, precision, and confidence calibration — revealing whether self-reported confidence scores track actual quality.

One-time snapshot that extracts patterns from work history and accumulated lessons, then distills them into concrete improvements — new agent/skill suggestions, roster quality review, memory pruning, or consolidating lessons and feedback into rules and agent/skill updates.

Post-install setup for foundry plugin. Merges statusLine, permissions.allow, and enabledPlugins into ~/.claude/settings.json; symlinks rules and TEAM_PROTOCOL.md into ~/.claude/.

Systematic diagnosis for unknown failures — local environment, tool setup, CI vs local divergence, hook misbehavior, and runtime anomalies. Gathers signals broadly, ranks hypotheses, uses adversarial Codex review for ambiguous cases, probes each, and reports root cause with a recommended next action. NOT for known code bugs (/develop debug) or config quality (/audit).

Create, update, or delete agents, skills, rules, and hooks with full cross-reference propagation. Non-trivial writes (agent/skill content-edits and creates) are delegated to foundry:self-mentor subagents; hook content-edits are delegated to sw-engineer; large cross-ref fan-outs (> 3 files) also delegate. The parent orchestrates and handles MEMORY.md, README, audit, and the final report. Also manages settings.json permissions atomically with permissions-guide.md.

Session parking lot — automatically parks diverging ideas and unanswered questions to project-scoped memory; /session resume shows pending items, /session archive closes them, /session summary gives a session digest

17 hooks across 14 events