Skip to content

nfemmanuel/iranti

v0.3.27 Feature

This release adds 2 notable features for engineering teams evaluating rollout.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Summary

AI summary

ToolCallGuidance now includes structured skip verdict fields and DriftSignal adds task drift detection.

Full changelog

Bundle 4 — Mid-turn executive function (M1 + M4)

M1: skipTool with teeth

Turns A2's advisory toolCallGuidance into a structured skip verdict. ToolCallGuidance now carries shouldSkip, skipConfidence (mean fact confidence / 100), and skipReasonCode so hosts can gate tool calls programmatically instead of string-matching note text.

Skip reason codes:

  • memory_not_needed — attend early-returned
  • no_entities_derived — no entity hints from tool args
  • no_facts_for_target — entities derived but no matching facts
  • facts_cover_target — facts match via entityKey.startsWith(hint + '/')

shouldSkip=true only fires on facts_cover_target.

M4: Task drift detection

Adds DriftSignal on AttendResult. detectTaskDrift() is a pure Jaccard overlap between the current message/context and the declared (inferred) task type, with stopword filtering. Fires when overlap < DRIFT_JACCARD_MIN=0.15. Gated on phase !== 'post-response'.

Tests

25 new B4 tests in run_b4_skip_drift_tests.ts:

  • 6 skip verdict pure unit tests
  • 8 drift pure unit tests
  • 5 skip integration tests
  • 6 drift integration tests

Backward-compat verified on 10 suites including A2 tool-call-retrieval (critical since ToolCallGuidance shape changed).

Co-Authored-By: Claude Opus 4.6 [email protected]

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

Track nfemmanuel/iranti

Get notified when new releases ship.

Sign up free

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.

All releases →

Beta — feedback welcome: [email protected]