EasyClaudeCode

Agent Trust Score

69/100

Scanned 4/7/2026

Agent Safety (40%)

66

Code Security (25%)

93

Cost Governance (20%)

65

Compliance (15%)

40

Found 5 keyword-based write guard(s) without intent classification

Fix: Replace keyword checks with intent classification. Add write_intent_classifier to middleware.

Missing: max iterations/turns

Fix: Add resource ceiling env vars, rate limiting middleware, and max_turns config.

Missing: identity re-validation

Fix: Rotate session tokens on SessionStart. Validate identity on every new session.

Behavioural probe required — static analysis cannot verify social engineering resistance

Fix: Run behavioural probes to test multi-turn social pressure scenarios.

No safety coordination logging found (design credit — not penalised in grade override)

Fix: Add safety_coordination_log table. Wire agent refusal events to log.

2/3 governance controls present (expiry: true, hash: true, rotation: false)

Fix: Hash tokens at rest. Add expiry. Track rotation.

No model routing or tier governance detected

Fix: Use model routing table. Route lightweight tasks to Haiku.

No spend alerting or daily tracking

Fix: Add spend tracking. Alert at configurable thresholds.

Insufficient PII controls detected

Fix: Identify PII fields. Log access. Define retention policy.

Audit logging with input hashing detected

Fix: Ensure audit_log table is insert-only (no UPDATE/DELETE).

No data residency documentation or AU region configuration found

Fix: Document data storage locations. Ensure AU data stays in AU Supabase region.