ask-investigateworkflow23
Minokun/xdev
What it does

Parallelize xdev health-check across 6 code dimensions

Best for

Parallelizing /xdev:ask health-check to break single-thread sampling budget; each dimension can operate independently with its own rg pattern + optional graphify read.

Inputs
  • · graphState: fresh|stale|none (passed from ask.md orchestrator after graph freshness check)
  • · scans all 6 dimensions by default (or focused single-dimension if ask.md routes directly)
Outputs
  • · per-dimension findings (severity: high|medium|low, what, evidence file:line, suggestion)
  • · degraded boolean per dimension (true if graph-dependent scan downgraded to rg/dir estimate)
  • · aggregated top 10 findings (ranked by severity × dimension priority)
  • · dropped dimension names (if agent failed)
  • · note field explaining scan method + limitations (Chinese)
Requires
  • · agent fan-out (1 per health-check dimension, parallel)
  • · rg (pattern search per dimension: security, testing, errors, architecture, deadcode, observability)
  • · graphify query (only if graphState='fresh', read-only, for architecture/deadcode cycles + cross-cluster coupling)
  • · file read (core modules, entry points)
Preconditions
  • · graphState must be provided (fresh|stale|none) by ask.md orchestrator
  • · 6 dimensions defined with rg patterns + hints (security, testing, errors, architecture, deadcode, observability)
  • · graph freshness already judged by ask.md (only reads, never writes graph)
  • · only called from /xdev:ask体检 mode (not direct skill call)
Failure modes
  • · graphState=none or stale but architecture dimension tries graphify (fails gracefully, marked degraded)
  • · rg pattern too broad → noise in findings
  • · dimension agent fails entirely → dropped, logged in droppedDimensions array
  • · top 10 deduplicate logic misses cross-dimension duplicates
  • · evidence file:line missing (violates 源码确认 honesty level)
  • · cyclic dependencies missed if graph is stale (degraded=true annotation required)
Trust signals
  • · 6 health-check dimensions named + defined with rg patterns (security, testing, errors, architecture, deadcode, observability)
  • · graphState passed in (fresh|stale|none) to control graphify usage per dimension
  • · FINDINGS_SCHEMA per dimension: dimension, findings[], degraded, note (all Chinese)
  • · evidence must include file:line (源码确认 requirement)
  • · suggestion field points to next action (e.g., /xdev:bugfix route)
  • · aggregation step post-processes to top 10 + rank by severity
  • · degraded flag + note explain why (e.g., 'Graphify needed for accurate coupling judgment, using rg estimate')