Cyber-Team v0.3.15

Highlights

• Hardened autonomous mandate output with an explicit typed role-state vocabulary and a single separately traced schema-repair attempt.
• Prevented unvalidated mandate prose from creating role gaps through the legacy heuristic path.
• Replayed the two live Observer failures: one completed with grounded claims and one correctly failed closed on an authoritative-state conflict.
• Activated and proved a bounded Operations advisory canary with live non-side-effectful tools, indexed memory, zero approvals, and no ERPNext or communication mutation.
• Preserved explicit company unknowns during idempotent evidence and strategy reconciliation.

Verification

• 363 backend tests and 25 frontend tests passed.
• Ruff, compileall, Alembic offline SQL, real PostgreSQL migration rehearsal, dependency audits, secret scan, Google Cloud isolation, FOSS/resource policy, Compose smoke, observability validation, and Trivy image scans passed.
• Staging was promoted backup-first and remains ready on application version 0.3.15.
• Exact-head push CI: https://github.com/Hyper-AI-Lab/cyber-ai-team/actions/runs/30973910819
• Exact-head manual CI: https://github.com/Hyper-AI-Lab/cyber-ai-team/actions/runs/30973925166

Scheduled exact-head proof remains pending the next GitHub cron by definition.

Cyber-Team v0.3.14

Highlights

• Replaces brittle free-text role-state circuit inputs with typed role-state-claims-v1 assertions bound to exact agent and role-gap IDs.
• Prevents negated, unrelated, historical, hypothetical, latent, and future role-gap language from falsely pausing autonomous domains.
• Preserves fail-closed behavior for genuine current-state contradictions and malformed claim contracts.
• Updates Next.js to 15.5.22 and PostCSS to 8.5.25, resolving the scheduled CI advisory failure.

Verification

• 362 backend tests and 25 frontend tests passed.
• PostgreSQL upgrade rehearsals, dependency/secret/FOSS/GCP-isolation checks, Compose smoke, and both image scans passed.
• Staging recovery and three Mistral-backed Knowledge canaries passed with zero side effects; readiness is ready.
• Exact-head push CI and full manual CI are green.

Cyber-Team v0.3.13

Bounded autonomous work recovery

• Enforces mid-execution domain pause mediation.
• Bounds generated work by depth, domain backlog, semantic deduplication, and cooldown.
• Adds audited generated-backlog stabilization and grounded circuit-breaker recovery.
• Surfaces backlog saturation and recovery state in readiness and the Owner Console.
• Passed 350 backend tests, 25 frontend tests, PostgreSQL migration rehearsals, Compose smoke, image scans, staging canaries, and exact-head push/manual CI.

Staging is healthy and ready; Knowledge and Governance recovered with zero external side effects.

Cyber-Team v0.3.12

Corrective autonomy lifecycle release: concurrency-safe Memory Steward remediation, canonical-inactive procedural-memory filtering, provider recovery evidence, bounded workflow-failure benchmarking, and verified staging Governor/Observer acceptance. Staging is running application SHA 6d9f5d5.

Cyber-Team v0.3.10

Highlights

• Hydrates semantic vector hits from canonical PostgreSQL memory before agent reasoning.
• Uses prior agent outcomes instead of replaying stale task prompts.
• Fails closed to deterministic PostgreSQL recall when embeddings are unavailable.
• Preserves specific-gap grounding and pre-invocation stale-memory quarantine.

Verification

• 335 backend tests and 24 frontend tests passed.
• PostgreSQL migration rehearsals and Docker Compose smoke passed.
• Core and UI image scans reported zero findings.
• Staging Security/Compliance canary completed with 0.98 confidence, passed grounding, and no external side effects.

Cyber-Team v0.3.5

Full Changelog: v0.3.0...v0.3.5

Cyber-Team v0.3.0

Full Changelog: https://github.com/Hyper-AI-Lab/cyber-ai-team/commits/v0.3.0