Consult. Reads memory and scans your repo before asking anything.

Interview. Asks only the decisions that actually steer the outcome.

Spec. Turns the brief into a rubric of exact, checkable commands.

Arm. Gates the loop to a goal file that survives restarts.

Work. The rubric comes back every turn, so the agent can't stop early.

Verify. A fresh-context agent re-runs every check and tries to refute the claims.

Rubric-gated loop. Progress is blocked until every item is checked with evidence from a real command, never vibes.

Fresh-context verifier. An independent subagent that never saw the worker's reasoning re-runs the checks and tries to prove the work wrong.

Persistent goals. State lives in a git-committed file, so a goal survives sessions, /clear, and days away.

Memory that compounds. Verified facts, working patterns, and dead ends are distilled after every goal. Every run starts smarter.

Optimization that can't regress. Baselines the metric, keeps only real improvements, and reverts anything that makes it worse.

Migrations, end to end. Survives /clear and restarts, and won't call it done until every call site is gone.

Bugs fixed for real. The fix has to survive a fresh-context verifier trying to refute it.

Backlogs while you sleep. Works turn after turn unattended within a set budget, gated so it can't fake progress.