Skip to content

Release v1.4.0

Latest
Latest

Choose a tag to compare

View all tags
@philngt philngt released this 20 Aug 04:50
· 19 commits to main since this release
v1.4.0
cdec73c
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

vNEXT Implementation — Synapse lifecycle and focused context packs

Synapse lifecycle and focused context packs

This release makes contextd more lifecycle-aware, more efficient at materializing context, and more useful when developers need help deciding what to do next.

Highlights

Lifecycle-aware synapse graph

Synapse projections now represent:

  • Typed relationships between knowledge nodes
  • Fresh and outdated lifecycle states
  • Runtime memory
  • Active context memory
  • Long-term memory

Outdated knowledge is preserved as a traceable node marked outdated instead of being silently removed. This keeps provenance and historical relationships explainable without treating
stale data as current guidance.

More efficient context materialization

Context materialization no longer performs a redundant full-workspace synapse rebuild solely to detect drift.

The runtime now reuses safe build results and reports context budgets separately as:

  • Referenced context
  • Static context
  • Total materialized context

This reduces repeated workspace scanning and prepares contextd to scale more predictably as the knowledge corpus grows.

Component-scoped pack knowledge

Manifest v3 introduces component-scoped retrieval through knowledge.md and pack.yaml#retrieval.

Instead of loading an entire pack, runtime retrieval can now load:

  1. Global Principles
  2. Only the component sections relevant to the current task

In the operator-steering pilot, the estimated pack payload decreased from approximately 4,307 to 1,386 tokens — about a 68% reduction — while preserving essential decision, escalation,
and stopping guidance.

Packs that support human judgment

Pack guidance has been refreshed to help users:

  • Identify missing knowledge before acting
  • Challenge assumptions and premature solutions
  • Understand what should happen next
  • Decide when more investigation is justified
  • Recognize when to stop, defer, or escalate
  • Avoid delegating all product and technical judgment to AI

The goal is not only to guide agent behavior, but also to help developers recover direction when long-running AI-assisted work becomes unclear.

Stronger pack validation

Pack validation is now manifest-version aware.

This release adds or updates:

  • Manifest-v3 schema support
  • Component retrieval validation
  • Canonical knowledge checks
  • Golden evaluation coverage
  • Artifact schema validation
  • Release readiness checks

Documentation

README, onboarding, pack authoring guidance, runtime documentation, schemas, and release checks have been updated to describe the new lifecycle and retrieval model.

Compatibility and migration

  • Existing manifest-v2 packs remain supported.
  • No immediate pack migration is required.
  • Manifest v3 can be adopted incrementally.
  • Existing workspace-isolation behavior is preserved.
  • No knowledge is shared or inferred across workspaces.
  • Legacy v0.x filenames and configuration adapters remain supported.

Pack authors can migrate to manifest v3 when they want component-scoped retrieval and canonical knowledge.md guidance.

Validation

  • Runtime tests: 46/46 passed
  • Golden evaluations: 8/8 passed
  • Pack validation: 15 packs, 0 errors, 0 warnings
  • Repetition checks: 17/17 passed
  • Wiki lint, pattern-index, atomic-write, and artifact-schema checks passed

Upgrade notes

No mandatory migration is required for existing users.

To benefit from smaller pack payloads, pack authors may adopt manifest v3 and divide canonical guidance into Global Principles and task-focused component sections.

Assets 13
Loading