Status: IMPLEMENTATION IN PROGRESS Updated: 2026-04-11
All decisions resolved, released as v1.4.0.
AgentSpec-style runtime constraint DSL. What syntax?
| Option | Pros | Cons |
|---|---|---|
| A) YAML-based rules with predicates (Recommended) | Consistent with existing policy.yaml, zero learning curve | Less expressive than custom DSL |
| B) Custom DSL (Python-like syntax) | Most expressive, AgentSpec paper uses this | Parser complexity, new syntax to learn |
Current plan: A (YAML-based, extends existing format with triggers/predicates/enforcement actions)
| Option | Pros | Cons |
|---|---|---|
| A) HMAC-SHA256 with hash chain (Recommended) | stdlib only (hmac module), simple key management | Shared secret — key compromise = forge all logs |
| B) Ed25519 asymmetric signatures | Strongest guarantees, no shared secret | Requires cryptography package (optional dep) |
Current plan: A (HMAC-SHA256, stdlib only). Ed25519 as optional upgrade path.
| Option | Pros | Cons |
|---|---|---|
| A) JSON files (Recommended) | Zero dependencies, consistent with project philosophy | Slower queries on large datasets |
| B) SQLite | Built into Python stdlib, better querying | File locking issues on some platforms |
Current plan: A (JSON files). SQLite as optional optimization.
| Option | Pros | Cons |
|---|---|---|
| A) Opt-in (warn only by default) (Recommended) | Low adoption friction | Tools can change without notice if user doesn't enable |
| B) Mandatory (block unverified tools) | Maximum security | Breaks if tool definitions update legitimately |
Current plan: A (opt-in with warnings)
| Module | Status | Agent |
|---|---|---|
spec_lang/parser.py |
✅ Done (YAML + fallback, Trigger/Predicate/Enforcement) | Phase 3a agent |
spec_lang/evaluator.py |
✅ Done (RuleEvaluator, custom predicates) | Phase 3a agent |
spec_lang/stdlib.py + defaults.py |
✅ Done (9 built-in predicates, 7 default rules) | Phase 3a agent |
audit/signed_log.py |
✅ Done (HMAC-SHA256, auto key gen) | Phase 3b agent |
audit/chain.py |
✅ Done (SHA-256 hash chain) | Phase 3b agent |
audit/verify.py |
✅ Done (4 checks: sig/chain/seq/time) | Phase 3b agent |
supply_chain/hash_pin.py |
✅ Done (SHA-256 pinning, thread-safe) | Phase 4a agent |
supply_chain/sbom.py |
✅ Done (CycloneDX 1.5, 20 package prefixes) | Phase 4a agent |
supply_chain/verify.py |
✅ Done (known vuln DB: litellm, ultralytics) | Phase 4a agent |
cross_session/store.py |
✅ Done (JSON file-based, path traversal safe) | Phase 4b agent |
cross_session/correlator.py |
✅ Done (4 correlation checks, z-score outlier) | Phase 4b agent |
cross_session/sleeper.py |
✅ Done (3 detection methods, E2E test) | Phase 4b agent |
| Integration + Tests | ⏳ After agents complete | — |
| Codex Deep Review | ⏳ After integration tests pass | — |
| Fix review findings | ⏳ After Codex review | — |
| CHANGELOG + Release | ⏳ After review fixes | — |