stempeck

architecture-docs

Generate or refresh `/docs/architecture/` — the current-state source of architectural truth for this codebase, grounded in code and git history (NOT in .designs/). Produces an anchor-dense corpus (`invariants.md`, `idioms.md`, `trust-boundaries.md`, `seams.md`, `subsystems/*.md`, `history.md`, `gaps.md`) plus a high-level overview layer (C4 L1/L2 diagrams, sequence flows, Nygard ADRs). Every claim is anchored to a file:line or commit SHA, or labelled literally `"unknown — needs review"`. A bundled `validate.sh` mechanically enforces citation density and phase-completion gates. Used to ground future design reviews so they cannot silently drift away from established idioms.

architecture-elevation

Validate the architectural ALTITUDE of a stated root cause or proposed fix before peer review or implementation. Interrogates the frame itself — whether the concern exists because of an abstraction that could be removed — before mapping enforcement sites within the concern. Tests whether a different boundary could DELETE the concern's substrate, not merely reorganize its enforcement. Invoked as a gate from /rootcause-all (after synthesis), /design-v5 (between Phase 3 and Phase 4), and /rootcause-review (at intake). Can also be invoked standalone.

rootcause-review

Perform scientific peer review of a rootcause_analysis.md document, validating all claims with independent verification. Appends review findings without modifying original content.

formula-create

Create a new agentfactory formula from a description or SKILL.md file. Generates a properly structured TOML formula with steps, dependencies, inputs, and iteration mechanisms. When given a SKILL.md, preserves phase gates as separate formula steps with enforcement language.

rapid-implement

Lean, adaptive implementation skill that classifies task complexity and scales process accordingly. Uses native Claude Code sub-agents and checkpoints for speed and token efficiency while maintaining quality gates.

agentic-skill-eval

Evaluates a skill library to identify which SKILL.md files would produce valuable autonomous agents when converted via /formula-create. Use when assessing a repository of skills for agent creation candidates, prioritizing which skills to convert to formulas, or auditing an existing agent roster for gaps.

documentation-update

Audits a documentation file line-by-line against the actual codebase, proving every factual claim with source file and line number citations. Produces a structured evidence table, applies corrections for inaccuracies, then verifies corrections are themselves accurate. Use when a user asks to review, audit, refresh, or update a documentation file, or says a doc is outdated.

github-issue

Creates well-documented GitHub issues (or comments on existing ones) from problems discussed in conversation. Investigates the codebase to map affected layers, files, and data flow, then writes up findings with acceptance criteria — without prescribing fixes. Use when a conversation has identified a problem that needs a GitHub issue, when the user wants to document a bug or problem for an implementer, or when dispatching work. Triggers on "create a GitHub issue", "write up an issue for this", "file a bug", or "document this problem".

improve-agent

Improve an agent's formula TOML based on post-execution learnings. Use after a formula run required manual intervention, produced incorrect artifacts, or left cleanup work for the operator. Classifies the failure type, scans for sibling vulnerabilities, selects the appropriate fix pattern, and surgically inserts corrective steps into the formula.

six-sigma-challenge

Challenges the quality of any completed review or assessment to determine what would achieve six-sigma (99.9999%) quality outcome. Works with scored reviews (X/10), qualitative reviews (PASS/FAIL), or unstructured assessments. Identifies gaps, independently discovers missed issues, stress-tests each proposed improvement for feasibility including logical impossibilities, and appends a "Six-Sigma Caveats" section. Use when a review is less than perfect, when the user asks "what would make this perfect," or after any quality gate completes.