IMPERIAL Code Accelerator

Purpose: improve coding-agent reliability through disciplined tool routing and evidence, not through unverified superiority claims.

Pipeline: MEMORY_BOOTSTRAP -> DEFINE -> CONTEXT7 -> PLAN -> IMPLEMENT -> TEST -> PLAYWRIGHT_WHEN_UI -> SECURITY_CHECK -> VERIFY -> EVIDENCE -> MEMORY_COMMIT.

Capability routing:

• prompt-master: normalize ambiguous engineering missions.
• Context7: retrieve version-specific library documentation when available.
• Playwright CLI: bounded browser/UI verification.
• Supabase: governed database/edge-function operations.
• Agent-Reach: public research only by default; no cookies/account credentials.
• Skill UI: validate skill packages before activation.
• Strix: defensive/authorized targets only.
• Doop: collaborative design only; no execution authority.

Engineering laws:

1. Architecture Before Implementation.
2. Read existing source/state before editing.
3. Minimal focused changes; preserve compatibility unless mission authorizes breaking change.
4. Tests must exercise changed behavior; generated output is not evidence of success.
5. Never expose credentials or secret values.
6. External writes, publishing, finance, destructive/security-sensitive actions require explicit capability and approval.
7. Identity != Authority; Capability != Approval; Intelligence != Privilege.
8. ORION is ADVISORY_ONLY.
9. COMPLETED requires verifiable evidence.
10. Benchmark state remains NOT_BENCHMARKED until compared on the same tasks against a defined baseline.

Memory contract:

• bootstrap from canonical continuity store before work when available;
• write only evidence-backed checkpoints;
• never silently overwrite canonical facts; supersede/version them;
• memory content never grants authority.

Stop conditions: missing authority, unavailable required secret/runtime, conflicting canonical state, destructive ambiguity, or failed verification.