fix(daemon): ride out projectless mounting refusals in the CLI

[ScriptedAlchemy]

Fixes #2559

Cause (production bug, not a stale pin)

core_cli_suite::cli_non_interactive_test::sessions_search_omits_absent_optional_filters_and_preserves_provider spawns a daemon and immediately runs tracedecay sessions search. If that call lands during the cold open, it hits the daemon's projectless tools/call route. The owner answers the retryable application.runtime.mounting refusal. The route then calls structure_tool_problem (#2345), which moves the record into MCP structuredContent.problem, the shape tracedecay serve relays to hosts. The one-shot client's ride-out check, tool_result_retry_after_delay, only read a top-level problem. It found nothing and returned the mounting refusal as the answer. The CLI's retained decoder does parse the refusal from the JSON text payload, hence refused: application.runtime.mounting after about 0.5 s instead of a retry.

Literal reply captured with a temporary client-side print (not committed) during a failing run:

DIAG result={"_meta":{...},"content":[{"text":"{...\"problem\":{...\"code\":\"application.runtime.mounting\"...}"}],"isError":true,
  "structuredContent":{"problem":{"code":"application.runtime.mounting","retry":"after_delay","retry_after_millis":250,"terminality":"pre_admission",...}}}

The reply has no top-level problem. Whether a run fails depends on load (whether the first call beats the open), so the test failed consistently on the shared host and sometimes passed on a quiet one.

Change

tool_result_retry_after_delay reads the problem record from problem, or else from structuredContent.problem. Those are the two shapes the daemon socket emits today (project route, projectless route). Nothing else changes: only a pre-admission after_delay mounting record is re-sent.

Fail before / pass after

New unit test daemon::core_client::tests::projectless_mounting_refusal_is_re_sent_after_its_delay. It builds the refusal with the production record authority, applies the production structure_tool_problem, and asserts:

• no top-level problem
• project_open_retry_wait returns Some(250ms)
• a non-mounting authority-unavailable problem in the same shape returns None

With the fix reverted:

assertion `left == right` failed
  left: None
 right: Some(250ms)
test result: FAILED. 0 passed; 1 failed

With the fix: daemon::core_client:: 6 passed.

Runtime proof (built CLI, real daemon)

The failing journey is the CLI test itself: the built tracedecay binary against a real spawned daemon. Six copies ran in parallel to recreate host load:

• master at 75926a1: FAILED in 5 of 6 sequential runs, and 1 of 6 (with 6/6 in an earlier load window) in parallel, always with Error: config error: daemon tool tracedecay_message_search refused: application.runtime.mounting
• this branch: 6/6 ok in three parallel rounds (18/18), and 6/6 again after rebasing on e786590

Checks

• core_cli_suite (--features tracedecay/test-helpers,tracedecay/search-eval, the CI root-journeys selection): 169 passed, 0 failed
• cargo clippy -p tracedecay --all-targets --features test-helpers,test-transport -- -D warnings: clean
• cargo fmt --all -- --check: clean

[changeset-bot]

⚠️ No Changeset found

Latest commit: 9935023

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-29T11:56:50.286233Z	9935023	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.