feat(governance): add policy-as-code — .llm-security/policy.json for distributable hook configuration

New policy-loader.mjs reads .llm-security/policy.json with deep-merge against
defaults that exactly match existing hardcoded values. Integrated into all 7 hooks:
- pre-prompt-inject-scan: injection.mode (env var still takes precedence)
- post-session-guard: trifecta.mode, window_size, long_horizon_window
- pre-edit-secrets: secrets.additional_patterns
- pre-bash-destructive: destructive.additional_blocked
- pre-write-pathguard: pathguard.additional_protected
- pre-install-supply-chain: supply_chain.additional_blocked_packages
- post-mcp-verify: mcp.volume_threshold_bytes, mcp.trusted_servers

Backward compatible: no policy file = identical behavior to v5.1.0.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Kjell Tore Guttormsen 2026-04-10 13:37:02 +02:00
commit 8ec320f40c
9 changed files with 300 additions and 13 deletions

View file

@ -43,18 +43,19 @@ import { createHash } from 'node:crypto';
import { extractMcpServer } from '../../scanners/lib/mcp-description-cache.mjs';
import { jensenShannonDivergence, buildDistribution } from '../../scanners/lib/distribution-stats.mjs';
import { writeAuditEvent } from '../../scanners/lib/audit-trail.mjs';
import { getPolicyValue } from '../../scanners/lib/policy-loader.mjs';
// ---------------------------------------------------------------------------
// Constants
// ---------------------------------------------------------------------------
const WINDOW_SIZE = 20;
const WINDOW_SIZE = getPolicyValue('trifecta', 'window_size', 20);
const STATE_PREFIX = 'llm-security-session-';
const STATE_DIR = tmpdir();
const CLEANUP_MAX_AGE_MS = 24 * 60 * 60 * 1000; // 24 hours
// Long-horizon monitoring (OpenAI Atlas, Dec 2025)
const LONG_HORIZON_WINDOW = 100;
const LONG_HORIZON_WINDOW = getPolicyValue('trifecta', 'long_horizon_window', 100);
const SLOW_BURN_MIN_SPREAD = 50;
const DRIFT_THRESHOLD = 0.25;
const DRIFT_SAMPLE_SIZE = 20;
@ -62,8 +63,9 @@ const DRIFT_SAMPLE_SIZE = 20;
// Sub-agent delegation tracking (DeepMind Agent Traps kat. 4, v5.0 S4)
const DELEGATION_ESCALATION_WINDOW = 5; // calls after input_source
// Rule of Two enforcement mode: block | warn | off (default: warn)
const TRIFECTA_MODE = (process.env.LLM_SECURITY_TRIFECTA_MODE || 'warn').toLowerCase();
// Rule of Two enforcement mode: block | warn | off (env var takes precedence over policy)
const policyTrifectaMode = getPolicyValue('trifecta', 'mode', 'warn');
const TRIFECTA_MODE = (process.env.LLM_SECURITY_TRIFECTA_MODE || policyTrifectaMode).toLowerCase();
// Volume tracking thresholds (cumulative bytes per session)
const VOLUME_THRESHOLDS = [