Skip to content

By-ear pass: quit feel + entry-authorization setup flow (spec 03-03 §5.3, spec 01 §3.6) #138

Description

@wine-fall

What it is

The sensory acceptance owed by the quit-correctness + guide auto-mode ship (#136, #137). Both changed how leaving and consenting FEEL, and the deterministic evidence stops short of that: #136's exit latency was measured in plain-mode pty runs only (0.03–0.12 s /quit-to-exit across 10 runs; SIGINT 0.05 s), not in the TUI where Ctrl-C rides the wire as a typed /quit; #137 retired the per-action y/N wall for entry authorization + conversational checkpoints, so the 2026-07-06 / 2026-08-01 interactive passes recorded in spec 03-03 §5.3 ran under a flow that no longer exists. The real-SDK smokes proved the mechanics (interrupt cuts the session in 0 ms; a .env read is denied by the PreToolUse hook; node --version runs with zero questions), not the conversation's feel.

Spec

specs/spec03/03-03-guide-harness.md §5.3 (the human-run repair bar, revision note in the criterion) and §7.3; spec 01 §3.6 (leaving, extended to the Q&A flows).

Done when

Each box is the user's verdict from a real run (make dev on a machine with at least one gap, TUI front-end); a "no" spawns its own follow-up issue rather than reopening this contract:

  • TUI Ctrl-C / typed /quit mid-setup-conversation: off the air in about a second, no swallowed second /quit, no orphaned front-end (dev.log command received → exit).
  • Entry authorization feels right: one y at the offer card, then the guide investigates and fixes while narrating — no permission wall, and it still stops to ASK (in prose) at anything destructive, forked, or costing money.
  • The secret guard never intrudes on a routine diagnosis (no spurious "credential" denial mid-repair), and the voice walkthrough's open .../app/api-keys step goes through.
  • The live policy check lands in at most two fetch rounds or degrades honestly with the link — a pass here also closes Voice guide's live policy check burns ~6 consent rounds — cap the attempts, then degrade honestly #102 by its own Done-when.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    by-earUser-run sensory acceptance; not assertable in a test

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions