Skill Claude CodeCodex
Tier-1 set-and-forget driver. Walk the slice dependency graph, dispatch per-slice managers (Tier 2) for every newly-unblocked slice, independently re-run each slice's verifier, flip status in agent-control/slice-status.md, and fan out the next wave. Use when the build orchestrator has slices ready and wants one lean…
Skill Claude CodeCodex
Output a paste-ready build-orchestrator prompt from agent-control brief and orchestrate-build driver. Use for orchestrate, kick off build orchestrator, or tier-1 prompt for next wave.
Skill Claude CodeCodex
Validate a adam project packet, run a required upfront grilling pass with the operator, and produce intake notes. The grilling session is the front-loaded gate that makes the rest of the build autonomous. Use at the start of any adam run, when a user hands over a packet folder or PACKET.md, or when asked to "ingest…
Skill Claude CodeCodex
Build a throwaway prototype to flesh out a design question. Either a runnable terminal app for state/business-logic questions, or several radically different UI variations toggleable from one route. Use when the right design isn't obvious and TDD is premature.
Skill Claude CodeCodex
Read-only audit of git vs project docs — main HEAD, floating branches, stale plan rows, operator punch list. Use for repo-truth, project sync, what landed on main, floating branches, or punch list.
Skill Claude CodeCodex
Fan out parallel explore subagents to research the codebase and domain, then fold their findings into plan.md, CONTEXT.md, and ADRs. Use after packet-intake completes, or when asked to "research and plan" for a adam project.
Skill Claude CodeCodex
Runtime acceptance check for a slice using the project browser MCP (Playwright MCP or cursor-ide-browser) and logger MCPs. Spin up the dev server, optionally run slice-scoped Playwright CLI tests from the packet, drive the UI, capture logs, validate ACs against actual behavior. Use after review-via-graph passes, or…
Skill Claude CodeCodex
Structural review of a builder subagent's branch using the code-review-graph MCP. Use after dispatch-builder or dispatch-parallel returns a green slice, before merging, or when you want to review without reading every line of the diff.
Skill Claude CodeCodex
Read-only security inventory of a repo or multi-repo ecosystem — auth surfaces, secrets, webhooks, dependencies, CI gaps — outputting scratch/security-audit/.json and .md for planning agents. Use before security refactor, tests-first security slices, or security-acceptance.
Skill Claude CodeCodex
Lightweight context steward — compress the latest orchestrator run into orchestration-runs/ and refresh agent-control/next-orchestrator-brief.md. Does not write product code or dispatch subagents. Run between bounded build orchestrator sessions.
Skill Claude CodeCodex
One-time per-repo setup for the Adam orchestration loop. Use when a project does not yet have packet/, plan/, slices/, or agent-control/, or when starting a new build for the first time.
Skill Claude CodeCodex
Commit and push current workspace changes with safe git defaults. Use when they say ship, commit, commit and push, or save this. Follow this skill yourself — never tell them to type /ship.
Skill Claude CodeCodex
Decompose an approved plan into independently shippable vertical slices, each with file paths, acceptance criteria, dependency edges, and a stub for the failing test suite. Use after research-and-plan is approved, or when asked to "slice the plan" or "break this into tasks".
Skill Claude CodeCodex
Sync project docs after a build wave — agent-control, orchestration-runs, next brief. Alias for session-steward in operator vocabulary. Use for steward, sync docs, or refresh next brief.
Skill Claude CodeCodex
Red-green-refactor TDD loop for direct work. Builds features or fixes bugs one vertical slice at a time. Use when the user asks for TDD, when a change has clear behavior to test, or when fixing a bug that should not regress.
Skill Claude CodeCodex
Write the failing test suite for every slice before any Composer 2 subagent touches code. The high-level agent owns the tests; subagents only make them green. Use after slice-to-tasks completes, or when asked to "write the failing tests" or "set up the red suite".
Skill Claude CodeCodex
Break a plan, spec, or PRD into independently grabbable issues using vertical slices. Issues are written either as GitHub issues, Linear tickets, or local files depending on project convention. Use after to-prd or research-and-plan, when scope is settled.
Skill Claude CodeCodex
Synthesize the current conversation into a PRD without re-interviewing the user. Use after a grilling session, after a long planning conversation, or when the user says "write up what we just discussed" or "turn this into a PRD".
Skill Claude CodeCodex
Triage issues through a state machine of triage roles. Move issues from inbox to ready-for-work via a series of narrow per-role passes. Use when an issue tracker has a backlog of fresh issues that need labels, scoping, and decisions.
Skill Claude CodeCodex
Break down a question or statement into the simplest calibrated terms — plain definition, pros/cons, supporting context, and open-source examples for agents. Respects adam/context technical level and preferences. Use when they ask what, break this down, explain simply, what does this mean, or ELI5. Do not tell them to…
Skill Claude CodeCodex
Ask for higher-level context on an unfamiliar section of code instead of diving straight in. Use when the user asks "what is this for?", when about to modify code in a subsystem you don't yet understand, or when the user explicitly says "zoom out".