AUDIT-drift-detector

AUDIT-drift-detector

Purpose

The federation has multiple registries (tool_registry.json, agent cards, SKILL_ALIAS_TABLE, affordances.yaml). Drift between them causes routing failures, orphaned skills, and silent capability loss.

Drift Dimensions

1. Build vs Runtime Manifest Drift — Canonical drift check via arifOS/runtime/manifest.py (build_manifest vs runtime_manifest). This is the primary drift detection mechanism post-KSR Epoch 1+2.
2. Tool Manifest Drift — Live MCP tools vs registered tools vs agent card references
3. Skill Registry Drift — SKILL_ALIAS_TABLE vs actual directories vs agent card skill IDs
4. Agent Card Drift — Card skill IDs vs existing skill directories
5. Schema Drift — Tool input schemas vs documented schemas
6. Floor Drift — Declared floor_scope vs actual floor enforcement
7. Verdict Taxonomy Drift — Verdict emissions vs closed 6-value set (OBSERVE_ONLY|SEAL|SABAR|VOID|HOLD|888_HOLD)

Detection Pipeline

1. Snapshot — Capture current state of all registries
2. Compare — Diff against saved baseline (or last-known-good)
3. Classify — Each mismatch: CRITICAL (breaks routing), WARNING (orphan), INFO (cosmetic)
4. Report — Structured drift report with fix recommendations
5. Escalate — CRITICAL drift → 888_HOLD before any SEAL operation

⚠️ Runtime-Injected Files (2026-07-19)

Some organ services modify files at runtime. Known patterns:

• WELL index.html: WebMCP adapter injected on service start → dirty after commit
• arifOS session-state: Runtime state files that change during operation

When dirty after clean commit: check if injected content was already committed → if yes, re-commit; if no, actual drift.

Baselines

• Canonical drift check: arifOS/runtime/manifest.py — build_manifest vs runtime_manifest (post-KSR Epoch 1+2)
• Tool registry: /root/arifOS/tool_registry.json
• Agent cards: /root/AAA/a2a-server/agent-cards/
• Skill alias: /root/AAA/skills/SKILL_ALIAS_TABLE.json
• MCP surface: Live tools/list from each organ
• Verdict taxonomy: arifOS/runtime/verdict.py — closed 6-value set

Floors

• F2 TRUTH: Report only what is actually observed. No inference without evidence.
• F4 CLARITY: Drift report must be actionable, not noise.
• F11 AUDITABILITY: Every drift check logged with timestamp and findings.