loop-readiness

Loop readiness review

This is a readiness review, not a gate. It never produces a pass/fail verdict or a single overall score — it reports what's present, what's missing, and what to consider per dimension, so the person deciding whether to start the loop has the full picture.

A "loop" here means any agent/process that runs multiple iterations with no human checking every step — an autopilot, a scheduled job, a long-running agent workflow, or anything similar. The review looks at the plan for that loop, not the loop itself — nothing gets executed during this review.

Step 1: Get the plan

• If a plan was given — as text or a file path — use it.
• If not, use the grill-design skill if available. Otherwise, interview the user one question at a time to build up an understanding of the plan — what it's trying to do, and how it's supposed to know if it's working. One question per turn, never ask a compound question.

Step 2: Review each dimension

Walk through all four. For each, note what's present, what's missing, and how deep the coverage is — there's no fixed checklist to fill in mechanically; use judgment about how much depth this specific loop actually needs.

2.1 Definition of Ready (DoR)

Is there something written down that defines what this loop is for and where its edges are? Depending on the loop, this could include: a goal statement, explicit non-goals / out-of-scope items, a contract (data shapes, API boundaries), a UI mock or diagram — go as deep as makes sense for the stakes involved. A one-line goal can be enough for a small loop; a loop that touches production data probably needs more.

2.2 Feedforward

Feedforward is guidance given to the loop before or during each iteration, to raise the odds it does the right thing in the first place — conventions docs, skills, contracts, bootstrap scripts, code-mod tooling, anything the loop reads or is constrained by while acting. Note whether this guidance is:

• Computational — a deterministic tool/script/generator it must go through
• Inferential — a doc/skill it reads and applies judgment to
• Both, or neither