Tool category test rules

Pattern	Approach	Cleanup
toolSearch / loadTool / unloadTool	Test first; load one deferred tool, test, unload before next	none
ask*Question	One sandbox question	PASS-UI; continue without blocking
getWeather, searchTheWeb, extractWebPage	Known-safe query/URL	none
artifacts	One inline markdown smoke; optional markdown-file at temp path	trash temp file
create* / update* / trash* (bot, prompt, todo)	CRUD on prefixed entity; nested update variants per multi-test-cases	trash before next tool
trashConvo / restoreConvo	Oldest non-trashed DB convo (not current); trash → restore per multi-test-cases	restore before next tool
search* / find* / read*	Sandbox prefix in query or read created id	none
forge-filesystem__*	Under tempDir only; trash → list → untrash cycle (see multi-test-cases)	trash / untrash
forge-discovery__*	Read-only, small limits	none
forge-typesense__*	Short query, limit 3	none
playwright__*	One session on safe URLs; skip browser_screenshot, browser_pdf_save	browser_close
composio__*	One list/get per connected app	skip if not connected
spawnAgent	Worker group only: sync spawn minimal bot, message Reply OK	none
checkAgentJob / abortAgent	Worker group only; jobs that worker started	none
switchAgent	Orchestrator only, last, after report; hand off to continuation bot	none

List/search tools must return { count, …array… } with matching lengths.