Instructions file CodexOpenCode
Instructions for Justinmendezai/The-Adam-Repo, covering agents.md — adam index, identity, drive the chat (all hosts, especially codex / chatgpt), read before every response (calibrated projects) and pods (logical — not separate chat personas unless fan-out).
Instructions file
Instructions for Justinmendezai/The-Adam-Repo: Read AGENTS.md first. Same harness, Claude-native paths.
Skill Claude CodeCodex
Copy Adam foundation rules and folder contract into a target project, and verify global MCP servers. Use during setup-adam, after foundation updates, or when adopting Adam in an existing repo.
Skill Claude CodeCodex
Watch in-flight Composer 2 builder subagents, redirect on failure or blocker, collect commits, and triage. Auto-retries before escalating; human items queue to agent-control/human-queue.md instead of blocking the run.
Skill Claude CodeCodex
Strategy-first mode — zoom out, no code, deterministic-over-inference, decisions before implementation. Use for brainstorm, zoom out, don't jump to code, or strategy not implementation.
Skill Claude CodeCodex
First-run interview that generates adam/context profile files and then sets up the project. Use when starting with Adam, onboarding a new founder, when context files are missing, or when the user just opened Adam and does not know what to do next.
Skill Claude CodeCodex
Render project status in a Cursor Canvas — sprint, slices, blockers, main vs branch state. Use for canvas project, visual status, or project diagram.
Skill Claude CodeCodex
Compact the current high-level orchestrator session into a primer doc so the next agent (or you, after a refresh) can resume cleanly. Prefer keeping one canonical agent-control/next-orchestrator-brief.md in sync. Use when token usage is approaching 50-60%, when the user asks to hand off, or before a long pause.
Skill Claude CodeCodex
Run an engineering council pass — INPUT, seven perspectives, synthesis, final-output. Use for council run, run a council, perspectives pass, or council pass on a topic.
Skill Claude CodeCodex
Path-unblocker for broken dev launches. Complete the main E2E happy path with the fewest safe changes; defer non-blockers to a ledger. Use when bringing up a freshly-pulled or freshly-installed dev environment, when the app boots but the main happy path breaks, or when the temptation to refactor would slow E2E…
Skill Claude CodeCodex
Disciplined diagnosis loop for hard bugs and performance regressions. Reproduce, minimize, hypothesize, instrument, fix, write a regression test. Use when chasing a bug whose root cause isn't obvious, or when a perf regression appears.
Skill Claude CodeCodex
Launch a single Composer 2 subagent to implement one slice. Use when dispatching a single ready slice or running a sequential slice that depends on another in-progress slice. For multiple independent slices at once, use dispatch-parallel.
Skill Claude CodeCodex
Tier-2 per-slice manager. Reads exactly one slices/ /SPEC.md, constructs a tightly-scoped worker prompt (reproducing the verifier verbatim), dispatches the worker subagent, runs the verifier once, and returns a single-line summary to the Tier-1 orchestrator. Owns sub-task fan-out for oversized slices. Use when…
Skill Claude CodeCodex
Fan out N independent slices to N Composer 2 subagents in parallel, each in its own worktree branch. Use when multiple slices in slices/ are marked ready to dispatch and have no unresolved dependson edges between them.
Skill Claude CodeCodex
Fan out a research pod to land findings in scratch/research. Use for dispatch research, research brief, or bring back info on a topic.
Skill Claude CodeCodex
Final end-to-end acceptance pass after all slices are merged: full tests-first suite, Playwright CLI when the packet defines e2e, extra verification.layers commands (API/integration/contract), code-review-graph integrity, and browser MCP walks for UI success criteria — without using Playwright for backend depth. Use…
Skill Claude CodeCodex
Proceed with the last proposed plan without re-explaining — minimal ack, execute. Use when user says go, yes, yep, keep going, do it, go ahead, proceed, approved, lgtm, or sounds good. On Codex/ChatGPT this is ordinary yes — never ask them to type /go.
Skill Claude CodeCodex
Get relentlessly interviewed about a plan, design, or decision until every branch of the decision tree is resolved. Use when the user wants to think hard before acting, when a plan feels hand-wavy, or when "grill me" is requested directly.
Skill Claude CodeCodex
Grill the operator on open decisions first, then run council on resolved INPUT. Use for grill me first then council or make council pass worth it.
Skill Claude CodeCodex
Grilling session that lands in CONTEXT.md and ADRs. Sharpens shared language with the agent and documents hard-to-explain decisions. Use when starting a non-trivial change, when the codebase has rich domain language, or when the user wants the grilling output durable.
Skill Claude CodeCodex
Write a handoff doc plus a paste-ready orchestrator kickoff prompt, optionally switch workspace and continue in the same chat. Use when the user says handoff-prompt, wants handoff and prompt, kick off next agent, or one handoff file and a prompt to pass.
Skill Claude CodeCodex
Compact the current conversation into a handoff document so another agent can continue the work. Use when switching agents mid-task, when the conversation is getting heavy, or when the user explicitly asks for a handoff. Always paste the kickoff prompt in chat and name the target repo.
Skill Claude CodeCodex
Process a raw idea or brainstorm doc into plan updates — critical review, gap analysis, packet or plan edits. Use for idea intake, process ideas, or critical review before building. Follow from natural language; do not ask for /intake.
Skill Claude CodeCodex
Merge approved slice branches then output operator manual test checklist. Use for merge then manual, merge branches and test list, or land slices and give me QA steps.