Claude Mythos Scaffold
Pattern-based scaffold for systematic AI agent work, with honest capability boundaries. Inspired by observed Claude Mythos Preview behaviors.
Guard 20
The write-time companion to audit-20: the same twenty failure classes, applied as rules WHILE writing code instead of hunted afterwards. Load before touching code on any coding task (feature, fix, refactor, generated code). Not an audit; spawns no agents and no fleet, costs no extra tokens; its job is to make the final audit-20 run come back mostly CLEAN. Trigger: writing code, implement, fix, re…
Mythos Scaffold
Activate Mythos scaffold. Fable-distilled working patterns (decomposition, verification, next-action decision) plus a discipline framework (mode, tool-stack, context-priming, decomposition, agent-loop, verification, failure-recovery) and domain modes (research, migration) that push Opus-class and smaller models toward Mythos/Fable-grade agentic behavior. Use when starting complex multi-step work,…
Audit 20
Twenty-pass verification gate for any code change. Covers injection, authentication, authorization/IDOR, secrets exposure, error handling, concurrency, resource leaks, data access/N+1, algorithmic hot paths, unbounded memory growth, external-call resilience, idempotency, transaction consistency, config hardening, supply chain, observability, API contracts, cross-module contracts, test gaps, and a…