[BunsDev]

Summary

Adds the real-terminal approval journey #411 asks for. Until now no journey reached the engine's own permission path, because the fake claude runs its own tools. A small local fake of the Messages API now asks the engine for real Bash calls.

Context

• Related: [TUI v2][P0] T4: ship first-class tools, diffs, and authoritative approval workflows #411. Test only, with no engine changes, so it branches from main and does not depend on the other open PRs.
• Evidence scope: engine-local only. Protected, daemon-backed approvals are a different path and are not covered.
• Not covered yet from [TUI v2][P0] T4: ship first-class tools, diffs, and authoritative approval workflows #411's list: expiry or revocation while a prompt is open, switching session mid-approval, double submit, a changed file after preview, timeout with a committed outcome, and huge or binary diffs.

Changes

• fake_anthropic.py: standard library only, bound to 127.0.0.1, with a fake key. It streams a tool_use for the prompt, then reports what the engine sent back, so the result is visible on screen. It logs each request's last message for checks.
• 14_next_approval_journey.sh checks each decision three ways: on screen, by whether the command ran (a marker file), and by what the model was told.

Step	Checked
New request	Panel opens with deny highlighted, names the real enforcer, and nothing has run
Enter	Refuses; command does not run; the model receives an error; row says denied
Rushed Enters and a burst of y as the panel arrives	Command never runs
y	Runs once; the model receives success
Next call	Asks again
Escape	Refuses; command does not run; the model receives an error
Y	Runs; the same command runs again with no prompt; a different command still asks; settings unchanged

Validation

• scripts/tui-tests/run.sh approval: 25 passed, 0 failed, 3 runs in a row under bash and once under zsh.
• An earlier run failed on my own check of a success result, because the engine omits is_error on success. The check now requires that it is not true.
• N/A: cargo fmt/check/clippy/test; no Rust changed.
• Not run: Linux locally, Windows.

Refs #411

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
docs	Ready	Preview	Nov 4, 2026 2:13pm UTC