mega-security

Plan a security check for an LLM agent BEFORE running mega-security. Profiles the product (domain, users, input/output surface, sensitive data), maps the threat surface to active categories and compliance overlays, and writes MEGA_SECURITY_PLAN.md as a "what to check" anchor consumed by mega-security. Adapt mode only.

Internal report writer. Auto-invoked by agent-optimize at loop completion; not a user-facing entry point. Reads loop history from .mega_security/feedback/ and writes .mega_security/MEGA_SECURITY.md (final audit-grade hardening report) plus .mega_security/meta/security-learnings.md.

Simulation-optimization loop that iteratively hardens an LLM agent against an automated Red Team without Blue Team (usability) regression. Dual-axis Pareto acceptance: Red Team block rate (DSR) must improve AND Blue Team usability loss (FRR) must stay within budget, else auto-revert via git. Forked from mega-data-eval-optimize; specialised for security-eval mode (stratified Seed-Epoch, fixed-target calibration, security failure-trace pre-tagging, native countermeasure-pattern reference injection, asymmetric-saturation architectural-pivot trigger).

Internal baseline-check orchestrator. Auto-invoked by agent-check at the end of its planning phase; not a user-facing entry point. Drives research → threat-simulation tier classification → Red Team / Blue Team suite curation → dual-axis evaluation → baseline measurement → MEGA_SECURITY_CHECK.md. Hardening is a separate skill (agent-optimize), invoked by the user when the check report recommends it.

Lightweight security check for a single chat system prompt (no agent loop, no tools, no RAG). Runs 100 attack tests across 4 attack types (prompt injection, jailbreak, PII disclosure, system prompt leak) as the held-out scoring set, plus a parallel tuning set of 100 used only by the optimizer. The attacks ship with the tool as a vetted set — each one previously got past or barely held against a capable baseline AI, so meaningful differences between models actually surface. Plus 16+16 legitimate-use tests for over-blocking detection. Writes MEGA_PROMPT_CHECK.md with block rate per attack type, failure examples, and weakness analysis.

Iteratively rewrite a chat system prompt to fix failures found by prompt-check. Reuses the check's scoring-set baseline (no re-measurement); measures the tuning set once at start; per iteration, the orchestrator (the main Claude Code session running this skill) rewrites the prompt against the TUNING-SET failures only — guided by a cumulative cheat_map of which rules previous iters accepted vs rejected — then re-measures on the tuning set as a search signal; only candidates that pass the tuning gate get a scoring-set generalization check. The scoring set is held back from the rewriter at all times. The final candidate is shown as a diff with three explicit choices — Auto-apply (overwrite the discovered source file), Select changes (pick a subset of accepted rules to apply), or Discard (leave source untouched). Adapt mode only.