Track follow-ups from the 0.24.1 release

Toolchain pinning, the CallErr simplification, and two more parallel-run
test flakes alongside dirge-c5i5.

Author: Yogthos

.beads/issues.jsonl

@@ -1052,3 +1052,7 @@
 {"_type":"issue","id":"dirge-dr9","title":"Semantic cache key should combine mtime + size (C5+M7)","description":"index.rs:36-43 compares only mtime. On 1-sec-granularity FS (HFS+, FAT, NFS), rapid same-second writes after our cache fill yield identical mtime. Add metadata.len() to the cache-validity key.","status":"closed","priority":4,"issue_type":"bug","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-22T00:43:33Z","created_by":"Yogthos","updated_at":"2026-05-22T00:49:45Z","started_at":"2026-05-22T00:43:47Z","closed_at":"2026-05-22T00:49:45Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-r57","title":"Doom-loop detection bypassed in Accept mode too (M1 partial)","description":"checker.rs Accept mode flows through track_doom_loop, but the bypass concern from M1 was specifically about Accept being permissive. Verify: doom-loop should fire even in Accept; if so this is closed already. Otherwise extend.","status":"closed","priority":4,"issue_type":"bug","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-22T00:43:33Z","created_by":"Yogthos","updated_at":"2026-05-22T00:49:47Z","started_at":"2026-05-22T00:43:47Z","closed_at":"2026-05-22T00:49:47Z","close_reason":"Verified: Accept mode IS subject to doom-loop detection (track_doom_loop fires inside the action!=Deny branch). Only Yolo bypasses, and Yolo's whole semantic is 'all rules off'. Coupled with the R4 H11 warning at startup, this is by-design behavior + visible to the user — not a code bug.","dependency_count":0,"dependent_count":0,"comment_count":0}
 {"_type":"issue","id":"dirge-a8l","title":"Clojure defmethod parent_class is the multifn, not the dispatch value","description":"adapters/clojure.rs:155-169 — (defmethod shape :circle …) gets parent_class=Some('shape'), self-referential. Dispatch value :circle is what makes the method unique. Either use dispatch value as parent_class or set None.","status":"closed","priority":4,"issue_type":"bug","assignee":"Yogthos","owner":"yogthos@gmail.com","created_at":"2026-05-22T00:32:14Z","created_by":"Yogthos","updated_at":"2026-05-22T00:37:38Z","started_at":"2026-05-22T00:33:25Z","closed_at":"2026-05-22T00:37:38Z","close_reason":"Closed","dependency_count":0,"dependent_count":0,"comment_count":0}
+{"_type":"issue","id":"dirge-jpma","title":"Pin Rust toolchain via rust-toolchain.toml","description":"CI uses 'toolchain: stable' in ci.yml, so clippy's verdict changes without the commit changing. rustc 1.98 tightened result_large_err and turned main red (fixed in PR #805).\n\nrelease.yml explicitly skips clippy/tests at tag time, justified in a comment as 'fmt / clippy / tests are deterministic per commit, so a SHA that was green stays green'. That reasoning only holds with a pinned toolchain. As-is a release can be cut from a SHA whose clippy would now fail.\n\nAdd rust-toolchain.toml pinning a specific stable, and bump it deliberately.","status":"open","priority":2,"issue_type":"task","owner":"yogthos@gmail.com","created_at":"2026-08-22T05:16:33Z","created_by":"Yogthos","updated_at":"2026-08-22T05:16:33Z","dependency_count":0,"dependent_count":0,"comment_count":0}
+{"_type":"issue","id":"dirge-qd33","title":"agent_loop::trace tests flake on macOS under parallel runs","description":"trace.rs tests fail intermittently on the macOS CI runner and pass on re-run. Same class as dirge-c5i5.\n\nSeen across PR #805 and #806, neither of which touches trace.rs:\n- oversized_payloads_are_bounded_and_marked (trace.rs:675)\n- ui_records_name_the_handle_the_tui_would_render_under (trace.rs:605)\n- the_models_words_reach_the_trace_from_the_finalized_turn (trace.rs:743)\n\nThe :605 failure showed a list missing its first entry, which points at shared trace state across tests rather than a real ordering bug. All pass in isolation.","status":"open","priority":3,"issue_type":"bug","owner":"yogthos@gmail.com","created_at":"2026-08-22T15:48:26Z","created_by":"Yogthos","updated_at":"2026-08-22T15:48:26Z","dependency_count":0,"dependent_count":0,"comment_count":0}
+{"_type":"issue","id":"dirge-lvsu","title":"CallErr: drop ServiceError instead of boxing it","description":"PR #805 fixed clippy::result_large_err by boxing CallErr.service_error. @chriselrod independently wrote a better fix on the #803 branch (569cfd51): store only a 'transport: bool' classification, since the sole consumer is is_transport_failure() and the full ServiceError is never surfaced (message already carries its Display).\n\n#805 landed first so main has the boxed version. Swap to the bool when convenient — avoids an allocation on every MCP tool error and removes a field nothing reads.","status":"open","priority":3,"issue_type":"task","owner":"yogthos@gmail.com","created_at":"2026-08-22T15:45:38Z","created_by":"Yogthos","updated_at":"2026-08-22T15:45:38Z","dependency_count":0,"dependent_count":0,"comment_count":0}
+{"_type":"issue","id":"dirge-0ysu","title":"agent::tools and input_reader tests flake under parallel runs","description":"Two more tests join dirge-c5i5's pattern: they fail under a full parallel 'cargo test --bin dirge' but pass in isolation, and which ones fail varies run to run.\n\n- agent::tools::tests::a_pending_permission_prompt_marks_the_call_as_waiting_on_a_person\n- ui::input_reader::tests::tty_is_dead_true_after_peer_close\n\nSeen on macOS while validating PR #803. Same shared-global/pty contention shape as dirge-c5i5.","status":"open","priority":3,"issue_type":"bug","owner":"yogthos@gmail.com","created_at":"2026-08-22T15:15:51Z","created_by":"Yogthos","updated_at":"2026-08-22T15:15:51Z","dependency_count":0,"dependent_count":0,"comment_count":0}