cat-mode-default applies on every prompt, with no regex guess

[EdbertChan]

Summary

The default context now reaches every user prompt when enabled, including acknowledgements and short execution turns.

Previously, a word list and length check guessed which prompts counted as work, so short execution turns were skipped.

The hook now uses one fixed exception: a prompt containing the typed command receives no extra context.

The flag-off path remains silent, and the documentation describes the every-prompt behavior.

Review Claim

When enabled, every prompt receives the default context unless it already contains the typed command; disabled mode remains silent.

Review Lane

behavior

Review Unit

engine-runtime

Safety Invariant

The change is limited to one hook's decision rule, its entrypoint tests, and aligned documentation. The agent-tool companion, flag resolution, and injected text remain unchanged.

Slice Rationale

This is one review claim in one hook. The decision-rule and documentation changes stay together because the documentation records the shipped behavior.

Non-goals

• No change to the agent-tool companion.
• No change to flag resolution or injected text.
• No corpus skill edits.

Architecture

Before

graph TD
    A["User prompt"] --> B["Meaning classifier"]
    B -->|"work prompt"| C["Default context"]
    B -->|"acknowledgement or short turn"| D["Silent"]

Loading

After

graph TD
    A["User prompt"] --> B["Typed /cat-mode present?"]
    B -->|"yes"| C["Silent"]
    B -->|"no, flag on"| D["Default context"]
    B -->|"no, flag off"| C

Loading

Test Plan

Test Plan

• python3 scripts/check_hook_test_coverage.py engine/hooks/cat-mode-default
• bash scripts/run_all_tests.sh
• ! git grep -n -e is_work_prompt -e WORK_VERB -e MIN_WORK_LENGTH -- engine tests
• bash scripts/scrub-handoff-artifacts.sh

Revert Plan

Revert Plan

• Safe to revert? Yes.
• Revert command: git revert <merge-sha>
• Post-revert steps: Re-run the hook coverage gate and repository test suite.
• Data migration? No.

---

Note

Medium Risk
Broadens when cat-mode context is injected on every user turn when the flag is on, which may change agent behavior on casual or meta prompts; scope is limited to one hook’s decision logic with tests updated accordingly.

Overview
When CATSTACK_CAT_MODE_DEFAULT=1, the cat-mode-default UserPromptSubmit hook now injects the cat-mode context on every user prompt instead of only prompts classified as “work” via length, acknowledgement lists, and work-verb regex.

The only skip rule left is a typed /cat-mode in the prompt (unchanged double-apply guard). Short replies like ok, bare slash commands, and brief execution turns now get the same default injection as long investigation prompts. Flag resolution, injected text, and the Agent PreToolUse companion are unchanged.

Implementation drops is_work_prompt and related heuristics from detect.py; tests and fixtures flip acknowledgement scenarios from silent to fires, add a prompt matrix, and docs (README, docs/ecosystem.md) describe the every-prompt behavior.

^{Reviewed by Cursor Bugbot for commit 9c44345. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_22dfa068-6cd0-4022-993e-4e0fcec21681)

[EdbertChan]

@Mergifyio queue

[mergify]

Merge Queue Status

• ✅ Entered queue — 2026-09-13 02:47 UTC · Rule: admin-bypass · triggered by @EdbertChan with the @mergifyio queue command
• ✅ Checks passed · on draft merge queue: checking #502 on main (814e208) #515
• ✅ Merged — 2026-09-13 02:53 UTC · at c6d2ee05ffecd866f7fbeef53604afc0cc2d77ae · squash

This pull request spent 5 minutes 33 seconds in the queue, including 4 minutes 32 seconds running CI.

Required conditions to merge

• check-success = lint
• check-success = test