fix(tui): keep task shell tools eagerly loaded (#2253)

[Hmbown]

Summary

• keep task_shell_start and task_shell_wait in the default eager native-tool catalog
• prevent first-use schema hydration from replacing execution for the long-running shell flow
• add regression coverage to the native deferral policy

Closes #2253.

Thanks @bevis-wong for the clear Windows/deepseek-v4-flash report and logs.

Verification

• cargo fmt --all -- --check
• git diff --check
• cargo test -p codewhale-tui non_yolo_mode_retains_default_defer_policy -- --nocapture
• cargo check -p codewhale-tui --all-features --locked

---

Greptile Summary

This PR fixes a regression where task_shell_start and task_shell_wait were omitted from the eagerly-loaded native tool catalog, causing first-use calls to be intercepted by the deferred schema-hydration path instead of executing — breaking the long-running shell flow, notably on Windows with deepseek-v4-flash.

• tool_catalog.rs: Inserts task_shell_start and task_shell_wait into DEFAULT_ACTIVE_NATIVE_TOOLS in the correct alphabetical position, so should_default_defer_tool returns false for both and maybe_hydrate_requested_deferred_tool no longer intercepts their first call.
• tests.rs: Adds two assertions to the existing non_yolo_mode_retains_default_defer_policy test, providing targeted regression coverage for the fix.

Confidence Score: 5/5

Safe to merge — the change is a two-line catalog addition with a matching regression test; no existing behavior is altered for any other tool.

The fix is minimal and surgical: two tool names added to an alphabetically-ordered constant, a targeted should_default_defer_tool check confirmed by two new test assertions, and the logic path through maybe_hydrate_requested_deferred_tool is already well-exercised by the surrounding test suite. There is no risk of side-effects on other tools in the catalog.

No files require special attention.

Important Files Changed

Filename	Overview
crates/tui/src/core/engine/tool_catalog.rs	Adds task_shell_start and task_shell_wait to DEFAULT_ACTIVE_NATIVE_TOOLS, keeping them eagerly loaded and preventing first-use schema hydration from intercepting the long-running shell flow.
crates/tui/src/core/engine/tests.rs	Extends non_yolo_mode_retains_default_defer_policy with two new assertions confirming task_shell_start and task_shell_wait are not deferred; regression coverage is correct and well-placed.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Model requests task_shell_start / task_shell_wait"] --> B{"Is tool in DEFAULT_ACTIVE_NATIVE_TOOLS?"}
    B -- "Before fix: NO (deferred)" --> C["maybe_hydrate_requested_deferred_tool intercepts call"]
    C --> D["Returns schema-hydration response\n(tool NOT executed, retry required)"]
    D --> E["Long-running shell flow broken\n(regression on Windows / deepseek-v4-flash)"]
    B -- "After fix: YES (eager)" --> F["should_default_defer_tool returns false\ndefer_loading = Some(false)"]
    F --> G["maybe_hydrate_requested_deferred_tool returns None\n(tool proceeds to execution)"]
    G --> H["task_shell_start / task_shell_wait execute normally"]

Loading

_{Reviews (1): Last reviewed commit: "fix(tui): keep task shell tools eagerly ..." | Re-trigger Greptile}

[gemini-code-assist]

Code Review

This pull request adds "task_shell_start" and "task_shell_wait" to the list of default active native tools in the TUI engine's tool catalog, and includes corresponding unit tests to verify that these tools do not default to a deferred policy. I have no feedback to provide as there are no review comments to evaluate.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Hmbown]

Independent review:

Confirmed minimal & correct. Two-line insertion into DEFAULT_ACTIVE_NATIVE_TOOLS preserves alphabetical order, and should_default_defer_tool (lines 56-72 in tool_catalog.rs) returns false via the existing DEFAULT_ACTIVE_NATIVE_TOOLS lookup. Regression assertions in tests.rs are well-placed inside non_yolo_mode_retains_default_defer_policy and exercise the exact code path that was failing.

Merges cleanly against main (no conflicts). The test only covers AppMode::Agent — AppMode::Yolo first-invocation isn't asserted, but the catalog lookup is mode-agnostic so coverage is sufficient.

v0.8.48 (PR #2256): Supersedes this fix. That PR rewrites should_default_defer_tool to _name/_always_load and unconditionally returns false for every tool, making the DEFAULT_ACTIVE_NATIVE_TOOLS membership check moot. No intent conflict — both land on "task_shell_* must not defer." Recommend landing #2271 first as the targeted hotfix for #2253; the catalog entries become harmless once #2256 merges, but the test assertions remain valid regression coverage and should be kept.

[Hmbown]

@Hmbown — quick harvest note (cross-author since you opened both). #2256's should_default_defer_tool change to always-return-false is the broader supersession of this targeted fix; intent-aligned but #2271 ships the narrower, safer-to-cherry-pick version.

Recommendation: land #2271 as the targeted hotfix for #2253 now (it's clean against main, low risk), and the assertions in non_yolo_mode_retains_default_defer_policy remain valid regression coverage even after #2256's broader change lands. So no work lost — this gets credited as "first fix" + the broader policy change supersedes the implementation but not the intent. Both contribute.