v0.3.0

Update CHANGELOG.md

v0.2.0

feat: SENTINEL v0.2.0 — The Agent-Safety Release

13 new features transforming SENTINEL from a controlled-execution kernel into a
complete agent safety toolkit:

1. Agent Firewall (firewall/): deterministic, framework-agnostic tool-call
   scanning with 32 built-in detection patterns (injection, exfiltration,
   privilege escalation, credential access, destructive, persistence). Ships
   with strict/balanced/permissive policy presets and runtime-loadable custom
   rules. Pure regex, sub-millisecond. CLI: sentinel-fw.

2. Deterministic Replay Engine (replay/): hash-chained execution log with
   record-and-replay, non-determinism detection, time-travel seek/step/rewind,
   stateAt reconstruction, and JSON export.

3. Kill Switch + Forensics (exec/killswitch.ts): transactional kill with
   graceful (safe-checkpoint window) and hard modes, compensation-based
   rollback in reverse order, full forensic snapshot, and recover() post-mortem
   plan.

4. Context Guardian (info/context.ts): real-time context-window health with
   token budget/utilization, normalized Shannon entropy, pollution scoring,
   untrusted source boundaries, lost-in-the-middle detection, and trust-aware
   compaction.

5. Memory Integrity Layer (memory/): signed, append-only, hash-chained memory
   ledger with cryptographic provenance, tamper detection, temporal-decay trust
   scoring, and HMAC-SHA256 signing.

6. Multi-Agent Trust Graph (bridge/trust-graph.ts): delegation depth limits,
   permission narrowing (subset-only, escalation rejected), per-hop trust decay,
   HMAC-SHA256 inter-agent message signing with timestamp-bound nonce replay
   protection, and anomaly detection. Exports Mermaid/Graphviz DOT.

7. OpenClaw Security Bridge (bridge/openclaw.ts): wraps MEMORY.md-style
   plaintext files with the Memory Integrity Layer for sealing, out-of-band
   tamper detection, and authenticated write-through.

8. OWASP ASI 10/10 Dashboard (compliance/owasp.ts): real-time coverage scoring
   of all 10 OWASP ASI Top-10 risks from enabled capabilities. CLI:
   sentinel-compliance.

9. Observable Agent Protocol (observe/): OpenTelemetry-style span/trace emitter
   with agent semantic conventions and OTLP-compatible JSON export.

10. Prevented Futures TUI (tui/): branching timeline rendering of blocked vs.
    allowed actions with risk scores, triggering rules, justifications, and
    safety bar. CLI: sentinel-tui.

11. A2A Identity Verifier v2: cryptographic per-message identity verification
    (HMAC-SHA256 + replay-bound nonces) with JIT, per-task permission narrowing
    via the Trust Graph.

12. Temporal Branching v2: 6-dimensional auto-normalizing scoring (confidence,
    safety, minimality, completeness, speed, reversibility).

13. New CLIs: sentinel-fw, sentinel-compliance, sentinel-tui.

Bug fixes:
- B1: version consistency across package.json, spec, and CLI
- B2: expanded package.json keywords for discoverability
- B4: HTTP API authentication enforced with constant-time comparison
- B5: session tokens now HMAC-signed with timingSafeEqual verification

Zero heavy dependencies. 215 tests passing across 28 test files.

v0.1.0

Add MIT License to the project