Skip to content

Orchestration And Auto Detection

github-actions[bot] edited this page Jun 25, 2026 · 3 revisions

Orchestration and Auto-Detection

Zero-config by construction — the harness reads a project's shape to wire itself, then surfaces memory at session start and phase boundaries without being asked.

Two things happen automatically so you never hand-configure them: the harness detects a project's shape (language, layout, CI) to wire the right phase tooling, and it orchestrates memory around your work — a session-start briefing and idle-time chains that push what's relevant, plus phase-boundary recall/save that pulls and persists context. Both are tuned to inform, never to nag.

How it works

Detection and orchestration are separate surfaces over the same vault:

Surface Trigger Does
Auto-detect install / register reads project shape → writes .harness/project.json config
Push (orchestration) session start · idle briefing + idle chains surface over-threshold signals
Pull (auto-context) phase boundary recalls relevant memory in, saves phase state out

Every emission is gated — thresholds, cooldowns, and per-emission toggles in auto-orchestration-config.md — so a signal fires only when it crosses a bar and its cooldown allows. Nothing emits on every session by default; the config is yours to tune, and a re-seed never clobbers your edits.

How it fits

  • AgentMemory — the store the push/pull surfaces read and write. Orchestration is the when; AgentMemory is the what.
  • Phases — auto-context injects at phase boundaries, so each phase opens with the memory it needs and closes by persisting what it learned.
  • Device-Wide Substrate — detection writes the on-host/project config that the substrate resolves state against.

See also

Detail:

Architecture · Explanation · Home

Orchestration and Auto-Detection

Clone this wiki locally