What this skill IS

This is a quality review of both creative work and assurance, not just an artifact checklist. Check:

• Did you design the problem or just react to symptoms?

• Did you explore alternatives or jump to the first solution?

• Did evidence feed back into problem reframing?

• Are there stepping stones worth preserving?

Procedure

Restate objective — what was requested, what "done" means

Check creative pipeline (problem factory):

• Was the problem designed proactively or reactively?

• Does a goldilocks assessment exist? Is the problem feasible-but-hard?

• Were alternatives to the problem framing considered?

• If evidence refuted claims: was the problem reframed?

Check solution pipeline (solution factory):

• Was SoTA surveyed (for architectural work)?

• Was a method family bet made explicitly (STRAT-*)?

• Were ≥3 genuinely distinct variants generated?

• Was selection explicit (policy before applying)?

• Were stepping stones preserved?

• Were weakest links identified for variants/selected solution (WLNK)?

• If the solution adds complexity, was MONO justification provided?

Check artifacts — tier-appropriate:

• T2+: ANOM/PROB-* if investigative, EVID-* if claims made

• T3+: SPORT-* if options evaluated, SEL-* if choice made, DRR-* if irreversible

• T4: SOTA-* and STRAT-* if architectural

• All: worklog updated

Check feedback loop:

• Did any evidence show "refuted"? If yes → was PROB-* updated?

• Did stepping stones get recorded for future work?

Check Factory 3 (process improvement):

• Was there friction in the FPF workflow itself? (gates too strict/loose, templates too heavy, missing skills)

• Did any hook block you incorrectly? Did any template feel wrong for the task?

• Process friction is a Factory 3 problem — record it, don't just tolerate it

• If friction found → write .fpf/anomalies/PROC-${CLAUDE_SESSION_ID}--<slug>.md with:

Process Friction Record

• ID: PROC-... Created: YYYY-MM-DD Factory: 1|2|3

Friction observed

(what happened, which gate/skill/template)

Impact

(time lost, workaround used, creative quality affected)

Proposed improvement

(what would fix it — skill change, hook adjustment, template redesign)

• Check .fpf/anomalies/PROC-*.md from previous sessions — are any still unresolved?

Run minimum verification — smallest test that reduces risk

Report — completed items, evidence links, remaining risks, open threads (including Factory 3 observations)

Output

• Update .fpf/worklog/session-${CLAUDE_SESSION_ID}.md with final summary

• Report to user: what was done, what's verified, what's left

Sentinel cleanup

After review:

• Write .fpf/.review-done (allows stop hook to pass)

• Delete .fpf/.session-active