Anti-Ceremony

Keep work tied to the caller-visible outcome and the proof still needed for it. The default guard is a single, artifact-free judgment before RPI Plan; the full honesty audit runs only on an explicit "full anti-ceremony audit" request.

Prompt

Quick guard. Outcome: "ao gate check lists the probe-coverage row". Proposed
process work: a coverage dashboard and a second audit pass. Remaining proof:
the gate row, green in CI. Stop condition: that gate passes on main.

It's working if

• The response is the YAML shape below and nothing else: no file appears under .agents/, and no tracker write is issued.
• reason: is one sentence, and parked_process_work: names the dashboard-shaped items.
• A full audit appears only after the caller wrote full anti-ceremony audit.

Quick guard

1. Freeze the caller-visible outcome in one short statement.
2. Park proposed process work that has no concrete consumer, gated subject or release decision, observed defect, or retirement condition.
3. Name the implementation or independent proof still missing.
4. Name the condition that ends this traversal.
5. Return STOP when the traversal would only create control artifacts or the frozen outcome is already sufficiently proved; otherwise return CONTINUE.

Progress needs evidence tied to acceptance or a named blocking uncertainty. Digest movement, finding counts, and additional reviews alone earn no credit. Informative red can justify a different experiment inside an explicitly selected outer goal's unchanged acceptance and allowance; repeated no-information work cannot. Keep required independent proof in remaining_proof, including every necessary unresolved finding. Apply Validate's effect-based risk rule to the cost of proof; never weaken the exact-subject or all-acceptance PASS bar.

Perform the guard exactly once in memory. Parking is a field in the response, not a tracker or delivery mutation. Return this shape:

decision: CONTINUE | STOP
reason: <exactly one sentence>
frozen_outcome: <nonempty string>
parked_process_work: [<work item>, ...]
remaining_proof: [<proof item>, ...]
stop_condition: <nonempty string>

The two lists contain strings and may be empty. Done when every field is present and the decision explains whether one RPI traversal earns its cost.

Full honesty audit

Run this mode only when the caller explicitly requests it. Examine each named control artifact or ceremony step for its consumer, gated subject or release decision, observed defect, and retirement condition; return concise keep, park, or drop findings in the response.

Boundary

Before returning the decision, read boundaries.md in the rpi skill's references directory. On malformed or missing inputs, return STOP with the missing field named in the one-sentence reason.