Skip to content

v1.0.2 — Close field→agent feedback loop

Choose a tag to compare

@Shangri-la-0428 Shangri-la-0428 released this 16 Apr 22:56
· 27 commits to main since this release

What's new

The field→agent feedback loop is now closed for the first time. Agents receive live field observations during prehook execution.

Signal-to-noise fixes

  • Concrete early-stop removed: scan_with_fallback no longer stops at Concrete level, ensuring Project/Typed/Universal results always reach the render pipeline
  • Lifecycle capability filter: Internal urn:thronglets:* capabilities (lifecycle, signal, presence, continuity) are excluded from agent-visible field observations
  • Scan limit increased: 3→12 raw scan results, truncated to 6 rendered observations — enough headroom after Concrete filtering

Infrastructure

  • Profiler now tracks field_observations count separately from emitted hints
  • MCP stdio timeout increased (15s→30s) for reliability

Field stats (at release)

  • 8,134 field points across 4 abstraction levels
  • 55,791 traces from 514 sessions
  • 4+ agent types contributing (claude-code, codex, openclaw, psyche)
  • Hebbian co-activation clusters forming (Edit×Read=10, Grep×Read=8)

Full Changelog: v1.0.1...v1.0.2