Ait

A Staff-QA-level audit of the why-ait.md value-prop claims surfaced three gaps where the docs over-promised relative to the implementation. This release closes all three.

Fixed

• Query DSL didn't search intent text. The doc's headline use case ("where is that prompt I wrote last month for the query parser?") returned field is not queryable in v1: title. intent.title and intent.description are now in the field registry and accept all standard operators including the substring ~ operator.

  ait query --on intent 'title~"query parser"'
  ait query --on attempt 'title~"auth"'
  ait query --on intent 'description~"staging session"'
  

• raw_prompt_ref was always null for shell-adapter runs. The schema column existed but nothing wrote to it. Every wrapped attempt now records the launched command at .ait/prompts/<attempt-id>.txt and updates evidence_summaries.raw_prompt_ref. Native-hook adapters (Claude Code, Codex, Gemini) still capture richer per-turn prompts via their hook bridges; this fallback ensures every attempt has some prompt reference.
• Docs claimed "no daemon" but ait runs src/ait/daemon.py on every wrapped run. The intent ("nothing leaves the machine") was right; the wording was wrong. New phrasing: "harness daemon is local-only — Unix socket, no network. No telemetry, no SaaS, no cross-machine sync." Updated in 6 places across README, docs site, and why-ait.md (en + 繁中).

After this release, all 10 why-ait claims hold up to behavioural verification end-to-end.

Install: pipx install ait-vcs or npm install -g ait-vcs.