Mace-Enterprise

codex-sparring

Get a second opinion from Codex (a different model) on an idea, plan, or code question — adversarial sparring to catch blind spots a same-model review misses. Use deliberately for weighty decisions. Read-only; costs the user's Codex credits.

experience-capture

Capture reusable development learnings from the current session into the knowledge wiki, after work has passed its verifier. Use when the user runs /learn, or at the end of a verified task to file what was learned. Gates on a green verify, then dispatches the librarian agent to write and commit.

karpathy-init

Scaffold the Karpathy 3-layer development contracts into a repository — SPEC (specs/), VERIFIER.md, ENVIRONMENT.md — and wire them into AGENTS.md so every agent (Claude, GPT/Copilot, Codex) follows them. Use when a repo has no specs/VERIFIER.md/ENVIRONMENT.md yet, or when the user runs /karpathy-init. Generic across stacks (.NET, Node, Python, Go, Rust, other).

retro

Run a team retrospective at a milestone or session end: read the outcome-log + recent diffs and turn finished work AND user corrections into concrete, owner-gated improvement deltas (process gate / skill edit / guardrail / role-brief / team-composition). Use when the user runs /retro, asks for a retrospective / "what did we learn", or criticizes a recurring behavior. Implements the team-coached-retrospective-loop methodology in the wiki.

skill-smith

Create and improve our own skills — a verified, token-frugal discipline. Use when a repeated workflow should become a reusable skill, or to refine an existing skill after a better way proved out. Gates on a real success; keeps skills lazily loaded (progressive disclosure).

brainstorming

Turn a rough idea into a validated design BEFORE any code or spec — Socratic questions (one at a time), 2–3 explored alternatives, incremental sign-off. Use when the partner describes a feature/project idea and the shape isn't settled yet.

codex-as-role

Enact ANY tool-neutral read-only role brief AS Codex (a different model) for a fresh-model perspective — analyst, security-auditor, researcher, reviewer, etc. Read-only. Use for a deliberate cross-model take on a specific task.

condition-based-waiting

Kill flaky async tests/automation by waiting for the actual condition instead of a guessed sleep/timeout. Use when tests have arbitrary delays, race conditions, or pass/fail inconsistently (especially under load or in CI).

experience-recall

Look up prior development learnings in the knowledge wiki before starting non-trivial work, so past patterns, gotchas, and verifier/environment recipes are reused instead of rediscovered. Use before designing or implementing a change, or when the user asks "have we solved this before / what do we know about X".

peer-review

Get a cross-model PEER REVIEW of a concrete change/diff from Codex (a different model), read-only. Use before merging a non-trivial change when an outside, blind-spot-catching review adds value. Costs the user's Codex credits.

session-search

Search your own past sessions/transcripts to recall what was already done or decided, instead of re-deriving it. Use for "did we already do/discuss X?", or to find a prior decision, command, fix, or path across past conversations.

subagent-driven-development

Execute a plan by dispatching a FRESH subagent per task with a code-review gate between tasks — fresh context each time, issues caught early. Use when a plan has mostly independent tasks and you want continuous progress with quality gates.

systematic-debugging

Find the root cause with a disciplined 4-phase process instead of shotgun fixes. Use when a bug is non-obvious, intermittent, or a quick fix didn't hold (e.g. a timezone/after-midnight calculation that's wrong only sometimes).

using-git-worktrees

Create an isolated git worktree for feature work — smart directory choice, a .gitignore safety check, and a clean test baseline. Use before parallel/independent work that should not disturb the current workspace.