This release adds 2 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Summary
AI summaryAdded SessionObjective helper and AutoCheckpointSignal with configurable thresholds.
Full changelog
Bundle 5: A6 objective memory + M5 auto-checkpoint signal
A6 — SessionObjective
Pure deriveSessionObjective() helper produces a structured session objective from the task string, recent messages, and optional checkpoint nextStep continuation. It attaches to WorkingMemoryBrief so every attend surface can see what this session is actually about. Pairs with A1 to form session executive function.
Confidence tiers:
1.0— action verb + signal tokens0.95— checkpoint continuation (nextStep ≥ 6 chars)0.6— one of verb/signals0.3— neither
Primary cap: 120 chars.
M5 — AutoCheckpointSignal
Pure detectAutoCheckpointTrigger() helper emits a soft recommendation signal when pressure builds, mirroring M3 writeNudge. It is a signal, not auto-persistence — hosts stay in control.
Priority:
drift_detected(critical)turns_without_write(strong)tool_cost_threshold(suggested)
Exported thresholds: AUTO_CHECKPOINT_TOOL_COST_THRESHOLD = 15, AUTO_CHECKPOINT_TURN_THRESHOLD = 3. Threaded through all three attend return paths; gated off post-response. The message embeds sessionObjective.primary so the host knows what to checkpoint about.
Tests
27/27 green in tests/attendant/run_b5_objective_autocheckpoint_tests.ts:
- 10 A6 pure unit
- 8 M5 pure unit
- 2 A6 integration
- 7 M5 integration
Regression gate
9/10 suites clean: b4-skip-drift, tool-call-retrieval, mid-turn-attend, m2-tool-result-autowrite, file-change-recall, a4-plan-proposal, user-operating-rules, compliance-task-relevance, artifact-capture. memory-attribution is a pre-existing local DB schema drift (validFrom column missing from local DB), unrelated to B5 — confirmed by git stash baseline test.
Backlog progress
5 of 8 bundles shipped. Remaining: Bundle 6 (A3 + A5), Bundle 7 (S5 + S6), Bundle 8 (M6).
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About nfemmanuel/iranti
Persistent shared memory for AI coding agents. Stores facts as `entity/key/value` triples with hybrid semantic search, task checkpoints, and conflict resolution — shared across Claude Code, Codex CLI, and GitHub Copilot.
Related context
Related tools
Beta — feedback welcome: [email protected]