Skip to content

LightAgent v0.9.6

Choose a tag to compare

View all tags
@wxai-space wxai-space released this 30 Jul 14:43
v0.9.6
2ea8917
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Highlights

LightAgent v0.9.6 adds production observability, deterministic evaluation, durable human review, and stronger shared-memory security boundaries.

  • Add TraceSummary, normalized usage/cost estimates, model/tool latency, retry/error metrics, generic trace exporters, and JSONL audit export.
  • Add LightEvaluator, EvaluationCase, and aggregate reports for output, tool choice, trace-event, recovery, latency, usage, cost, and custom checks.
  • Add HumanApprovalHook, approval requests/decisions, in-memory and JSON review stores, exact-context approval reuse, batches, timeout handling, and human feedback.
  • Add durable LightFlow approval checkpoints with approve, reject, edit, respond, resume, and fresh approval on rerun.
  • Add fail-closed require_write_admission, sanitized admission failures, and memory_retrieve_filter audit events for shared memory.
  • Add an adversarial cross-user fake Graph Memory regression for issue #39 and document the remaining real Mem0 Graph validation boundary.
  • Include Python executor AST blocklist hardening from #85 and roadmap/issue-priority updates from #86.
  • Update all localized READMEs, tracing, evaluation, human-review, LightFlow, memory-security, and roadmap documentation.

Compatibility

  • Existing agent.run("hello"), structured results, and streaming calls remain compatible.
  • Existing LightFlow approval handlers remain compatible.
  • Ordinary memory writes remain compatible unless require_write_admission=True is explicitly enabled.
  • Internal reflection, delegation, and other derived memory remains non-injectable until explicitly promoted.
  • No new required core dependency is introduced.

Security Scope

The #39 tests validate LightAgent's framework boundary against a destructive fake shared-graph backend. They do not establish an affected-version range or claim that every Mem0 Graph deployment is fully remediated. Production users should run the documented opt-in matrix against their exact backend version and storage configuration.

Validation

  • GitHub CI passed on Python 3.10, 3.11, 3.12, and 3.13.
  • Each CI environment completed the full tracked suite with 161 tests.
  • Local controlled validation: 161 passed.
  • python -m compileall -q LightAgent, git diff --check, and wheel build passed.
  • Built wheel: lightagent-0.9.6-py3-none-any.whl.

Full changes: v0.9.4...v0.9.6

Assets 2
Loading