CI / verify (push) Failing after 40s

Details

docs(prospector): retire model-boss for @prospector/ai-harness; add DRAFT mode + alignment gate

- Replace every model-boss/coordinator reference with the @prospector/ai-harness
  story (direct vLLM client + classify/draft/judge/orchestrate task registry +
  on-demand GPU lifecycle + CoT-workflow runner + cost meter) across ai-first-v4,
  draft-engine, model-eval-pipeline, and PROSPECTOR.md; GPU_INFERENCE_URL is the
  canonical inference contract. Note ai-harness is promotable to a shared @ct
  package (onlyfans carries a parallel src/engine/classifier.ts).
- Fix migration collision: ai-first-v4 actor-attribution renumbered 0007 -> 0016
  (0007_tasks.sql exists; tree at 0013).
- Add the three missing pieces from the plan: a formal DRAFT runner mode distinct
  from PAUSE + DRAFT->GO graduation (new control-modes.md); a runtime per-draft
  alignment gate (deterministic facts/policy + GPU judge; spec_conflict/
  policy_conflict holds) in draft-engine's pipeline; and the facts/mission config
  schema (src/specs/, 0014_specs.sql) in ai-system-plan §5 + draft-engine.
- Index control-modes.md and the ai-harness rename in features/README.md.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-06-30 11:06:30 -04:00

3.1 KiB

Raw Permalink Blame History

Feature documentation — grouped by feature

Per-feature docs for prospector, grouped by area. Start with ../PROSPECTOR.md (the unified definition) and ai-system-plan.md (the AI master plan); everything below drills down.

🤖 AI system & control surface

The v4 "AI-first" effort — making the app operable by an LLM agent.

Doc	What it covers
`ai-system-plan.md`	Master plan. Layered architecture (trained-stable vs mission-volatile), the model facts (size/storage/GPU-vs-CPU/accuracy), UI/UX gaps, tuning levers. Read this first for the AI system.
`ai-first-v4.md`	The three control planes (operation/observation/autonomy) + the AI-roles taxonomy (orchestrator vs classifier vs message-generator).
`ai-orchestrator.md`	The streaming agent (`@packages/mcp-orchestrator/`) that watches the system, reports, and nudges the operator.

✍️ The engine (classify → generate)

How an inbound becomes a reply.

Doc	What it covers
`draft-engine.md`	OSS-on-GPU draft engine (the `@prospector/ai-harness` inference layer) + the CoT workflow builder; the `template` vs `do-gpu-<model>_<build>` engine selection; the runtime per-draft alignment gate + facts/mission config.
`control-modes.md`	The runner modes GO / PAUSE / AWAY + the new DRAFT trust-ramp mode (stage-for-review) and the DRAFT→GO graduation criteria.

🎓 Training & evaluation

Turning Quinn's message history into a tuned, hardened model.

Doc	What it covers
`training-loop.md`	CoT-labeled corpus → LoRA → eval-gated build flip; the matcher+generator hybrid; the move taxonomy.
`model-eval-pipeline.md`	The Claude-advisor / OSS-worker bake-off; per-role scoring; the candidate roster.
`../../tooling/eval/README.md`	The runnable pipeline — extract → sweep → rationalize → run → score, plus `gpu.py`. (Code, not just design.)

💻 GPU, cost & ops

Doc	What it covers
`gpu-cost-control.md`	Presence-driven warm-up, live cost meter, pause/teardown.
`../../tooling/eval/README.md`	`gpu.py` — on-demand provisioning with the self-reaping auto-teardown (no secret on the droplet).
`deploy.md`	Prod backend deploy on the DO droplet.

🔌 Interfaces

Doc	What it covers
`mcp.md`	The `@packages/mcp-prospector` agent interface (one MCP tool per REST endpoint).
`ai-orchestrator.md`	The proactive streaming orchestrator (see AI system above).

3.1 KiB Raw Permalink Blame History