fix: avoid magic keyword activation on informational queries

[Yeachan-Heo]

Summary

• skip magic keyword activation when prompts are asking what a keyword means or how to use it
• cover informational intent patterns for English, Korean, Japanese, and Chinese in both source and packaged keyword-detector artifacts
• make session-scoped state_clear also remove legacy non-session state files (including older root-level legacy files) and add regression tests

Testing

• npx vitest run src/hooks/keyword-detector/tests/index.test.ts src/installer/tests/hook-templates.test.ts src/tools/tests/state-tools.test.ts src/lib/tests/mode-state-io.test.ts src/hooks/tests/bridge-routing.test.ts src/hooks/tests/bridge.test.ts src/hooks/tests/bridge-openclaw.test.ts src/tests/hooks.test.ts src/tools/tests/cancel-integration.test.ts src/tests/smoke-slack-and-state.test.ts
• npm run build

Notes

• a broader unrelated test, src/tests/cleanup-validation.test.ts, timed out in builtin skill resolution during an exploratory run and was not needed for validating this fix

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2617876b95

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Stop treating imperative "explain" prompts as informational

The English informational regex now includes bare verbs like explain and describe, so actionable prompts such as use ralph to explain this bug or autopilot explain this code are classified as informational and no mode is activated. This is a functional regression from the previous keyword behavior and breaks common command phrasing where users intentionally invoke a mode to provide an explanation.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Scope informational checks to the matched clause

The fixed ±80-character context window can suppress a valid command mention if an informational phrase appears nearby in the same prompt; for example, What is ralph? ralph fix this bug fails to trigger because the second ralph still falls inside the what is window. This causes mixed "ask then execute" prompts to be dropped even when they contain an explicit actionable keyword.

Useful? React with 👍 / 👎.