Agent Claude Code
Grades work against the plan's acceptance criteria from a separate context window. Use in the evaluate stage. The anti-self-praise firewall — must never be the agent that built the work. Verifies by running tests/app, returns pass/fail per criterion with evidence.
Agent Claude Code
Disposable read-only fan-out. Use whenever a task needs broad searching or large-file reading that would blow the caller's 40% budget. Burns its own context, returns a short distillate, and dies. The context firewall — keeps the caller in the smart zone.
Agent Claude Code
Entropy garbage-collector. Use periodically/in background to scan for drift — stale docs, dead always-loaded context, code smells, convention drift — and open small, green, one-move fixes. Keeps the repo coherent and the smart zone smart for future runs.
Agent Claude Code
Implements one issue at a time, test-first. Use in the implement stage to build a vertical slice via TDD and the horizontal/vertical layering contract. Produces green, clean code; hands the slice to the evaluator. Never grades its own work.
Agent Claude Code
Expands a goal into an executable plan. Use in the prd/issues stages to turn a raw requirement into a committed PRD and atomic, testable issues. Runs the founder funnel for new projects. Returns a plan; does not write production code.
Skill Claude CodeCodex
Write a committed handoff checkpoint so a fresh session can resume with zero loss. Use when crossing the 40% context line, on every stage transition, and at session end. Writes .trace/checkpoints/ - .md while the agent is still sharp.
Skill Claude CodeCodex
The forward code-quality constraint (Clean Code). Use while writing code in the implement stage and as a review lens in evaluate. Intention-revealing names, small single-responsibility functions, no duplication, comments that explain why not what.
Skill Claude CodeCodex
Grade work against the plan's acceptance criteria from a SEPARATE context — tiered by risk (L0 self-check, L1 lightweight reviewer [default], L2 full evaluator). Use in the evaluate stage. The anti-self-praise firewall — the agent that built the work must never be the one that grades it. Returns pass/fail per…
Skill Claude CodeCodex
Apply Elon Musk's Five-Step Algorithm to any requirement or plan before building. Use in the intake/plan stages to question, delete, simplify, accelerate, and automate — in that order. Prevents the expensive mistake of optimizing or automating something that should have been deleted.
Skill Claude CodeCodex
Greenfield-only intake gate. Apply founder/startup best-practice (from the Founder's Playbook) before any PRD on a NEW project — who is the user, the smallest valuable slice, the riskiest assumption, the build-measure-learn loop. Skip for existing projects.
Skill Claude CodeCodex
Entropy garbage-collection. Periodically scan for drift — stale docs, dead "always-loaded" context, smells, mismatched conventions — and open small targeted fixes. Use to keep the smart zone smart and the repo coherent for future agent runs. Runs orthogonally to the main lifecycle.
Skill Claude CodeCodex
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
Skill Claude CodeCodex
Fan out the horizontal expansion of the implement stage to parallel generators. Use AFTER the vertical walking skeleton passes evaluate, to build independent features concurrently. The main agent dispatches one generator per issue; parallelism is gated on disjoint file footprints. Vertically related, horizontally…
Skill Claude CodeCodex
Drive a long-running "work → check → repeat" loop until an exit condition is met, max iterations hit, or a human-judgment flag is raised. Use for autonomous long tasks where an agent does work and an evaluator decides whether it passes (e.g. "keep fixing until all reviewers pass"). Backed by bin/ralph.sh.
Skill Claude CodeCodex
The recovery code-quality constraint (Refactoring). Use to improve code structure without changing behavior — smell → named refactoring, always under green tests, one move at a time. Used by the generator (in the tdd refactor step) and the gardener (entropy cleanup).
Skill Claude CodeCodex
Cut a versioned release of harness-mini — bump VERSION, roll the CHANGELOG, tag, and publish a GitHub release. Use when shipping a new version. Wraps bin/harness.sh release; covers the human-judgment steps (semver choice, changelog curation) the script can't make.
Skill Claude CodeCodex
View and advance the lifecycle stage of every requirement. Main-agent only — the single authority that moves a plan through intake→prd→issues→implement⇄evaluate→checkpoint→done. Use to see what stage work is in, route simple vs complex requests, and promote a plan (no sub-agent may self-promote).
Skill Claude CodeCodex
Implement code test-first via the red→green→refactor loop. Use in the implement stage for every issue. Never write implementation before a failing test; never refactor on red. The generator's core working rhythm.
Skill Claude CodeCodex
Decompose a PRD into atomic, testable issues — the executable work units the generator implements one at a time. Use in the issues stage. Each issue is independently verifiable and maps to one vertical slice or a step within one.
Skill Claude CodeCodex
Turn a raw idea or requirement into an executable PRD document — a versioned artifact the harness treats as a first-class source of truth. Use in the prd stage after founder-check/five-step. Output is agent-readable, testable, and committed.
Instructions file CodexOpenCode
Instructions for harness-mini/harness-mini, covering harness-mini — agent map, start here — routing gate (precedence), the one rule that governs everything: the 40% line, core constraint: mini (when changing the harness itself) and lifecycle (the state machine you live inside).
Cursor rule Cursor
Write a committed handoff checkpoint so a fresh session can resume with zero loss. Use when crossing the 40% context line, on every stage transition, and at session end. Writes .trace/checkpoints/ - .md while the agent is still sharp.
Cursor rule Cursor
The forward code-quality constraint (Clean Code). Use while writing code in the implement stage and as a review lens in evaluate. Intention-revealing names, small single-responsibility functions, no duplication, comments that explain why not what.
Cursor rule Cursor
Grade work against the plan's acceptance criteria from a SEPARATE context — tiered by risk (L0 self-check, L1 lightweight reviewer [default], L2 full evaluator). Use in the evaluate stage. The anti-self-praise firewall — the agent that built the work must never be the one that grades it. Returns pass/fail per…