GPT Review — Claude Code Pack 1 Dispatch Prompt
GPT Review — Claude Code Pack 1 Dispatch Prompt
Date: 2026-05-04 Scope: review Opus dispatch prompt for IU-0 Pack 1 after User C10 approval.
Verdict
APPROVE DISPATCH — with one operational clarification to Opus before sending to Claude Code.
The prompt matches 09b Pack 1 execution scope: preflight, backup, vocab insert, two tables, L1/L2 birth gates, smoke T1–T14, cleanup, final verify, report, hard stop. No Qdrant/outbox/legacy vector/Pack 2 scope is introduced.
Required operational clarification
Before Opus sends the prompt, add one sentence near Preflight/Step 1:
If
dot_configschema differs from the assumed columns (config_key,config_value,description,created_at) or Step 1 INSERT fails because of column/type mismatch, Claude Code must STOP and report the actual schema. Do not adapt the INSERT, do not add columns, and do not proceed.
Reason: this preserves HB-1 and avoids Claude Code silently modifying existing config schema.
Checks
- Scope: Pack 1 only — PASS.
- Law/constitution conflict: prior GPT + Opus law reviews PASS; no new conflict introduced.
- Preflight: includes PG access, gen_random_uuid, dot_config, table existence, backup — PASS.
- Safeguards: rollback, smoke tests, cleanup, hard stop — PASS.
- Smoke tests: T1–T14 cover L1, L2, deferred positive, soft-delete, duplicate canonical, vocab negative — PASS.
- Boundary: no Qdrant, no outbox, no legacy mutation, no Directus integration — PASS.
Decision
Opus should dispatch Claude Code after adding the operational clarification above, then wait for the actual KB execution report and verify it from Agent Data before any Pack 2 discussion.