Tools provided
Tools the agent exposes — your AI client calls them automatically when it needs them.
- parse_symptom — Translate a user-visible symptom into code-level concepts — which surface, which feature area, which component vocabulary to search for — keeping the mapping explicit so a wrong assumption is visible.read-only
- plan_code_search — Turn a symptom into an ordered search plan — broad terms first, then component names, then function and configuration names — with what a hit and a miss each mean.read-only
- map_cross_platform_impact — Force the cross-surface question for a bug — affected, not affected, or not checked for every surface — so a single-repo investigation cannot quietly present itself as complete.read-only
- state_unknowns — Separate what is known from what is assumed and what was never looked at, so a brief cannot present reasoning as evidence.read-only
- write_planner_prompt — Produce a copy-paste prompt handing the investigation to a planner, carrying the goal, the suspected scope, what is out of scope, and what the planner should read first.read-only
- draft_investigation_report — Assemble the full investigation brief from findings that were actually gathered, refusing to produce findings when none were supplied.read-only
