Systematic codebase exploration that maps architecture, components, and dependencies. Use when you need to understand how a feature works across a codebase, explore an unfamiliar project's architecture, trace data flow through multiple layers, or plan changes that span several components. Maps the high-level structure first, then dispatches parallel agents to explore each area in depth. Produces a synthesis with file:line references, execution flows, and actionable recommendations.
87
81%
Does it follow best practices?
Impact
90%
1.63xAverage score across 6 eval scenarios
Passed
No known issues
Two-phase exploration process and synthesis output format
Mandatory announcement
0%
37%
Red flags checklist
0%
0%
Phase 1 before Phase 2
0%
100%
Haiku for discovery
0%
0%
Sonnet for deep dives
0%
0%
State tracking block
0%
100%
State tracking updated
0%
100%
Parallel deep-dive dispatch
0%
100%
Synthesis: Architecture Overview
100%
100%
Synthesis: Component Map
100%
100%
Synthesis: per-perspective analysis
100%
100%
Synthesis: Integration Points
100%
100%
Synthesis: Recommendations
100%
100%
Pressure resistance and mandatory announcement with red flags
Mandatory announcement
0%
40%
Production pressure flag checked
0%
33%
Does not just grep
100%
100%
Explains systematic over ad-hoc
100%
100%
Phase 1 performed first
20%
100%
State tracking present
0%
37%
Explains trade-offs to user
90%
100%
Synthesis sections present
87%
100%
File:line references
100%
100%
No premature completion
100%
100%
Adaptive parallel deep dives and per-component synthesis
Discovery before deep dives
0%
100%
One deep-dive per service
0%
100%
Parallel deep-dive dispatch
0%
100%
PERSPECTIVES_FOUND tracked
0%
25%
DEEP_DIVES_COMPLETE tracked
0%
62%
Per-service analysis sections
100%
100%
Cross-cutting patterns
50%
100%
Integration points section
100%
100%
Actionable guidance not raw dumps
100%
100%
File:line evidence
100%
100%
Recommendations section
100%
100%
Skip-when criteria compliance
Function signature present
100%
100%
All parameters documented
100%
100%
Parameter types correct
100%
100%
Return value documented
100%
100%
Errors/exceptions documented
100%
100%
No discovery report produced
100%
100%
No skill state tracking block
100%
100%
No multi-phase process documented
100%
100%
Approach is direct lookup
100%
100%
Approach explains why
100%
100%
Discovery output format and deep-dive context passing
## DISCOVERY SUMMARY section
0%
100%
## ARCHITECTURE PATTERN section
0%
100%
## COMPONENTS/MODULES section
0%
100%
## LAYERS/BOUNDARIES section
0%
100%
## RECOMMENDED DEEP DIVES section
0%
100%
Architecture confidence level
0%
100%
Components table has paths
25%
100%
File:line evidence in discovery
0%
100%
No speculation in discovery
100%
100%
Dispatch plan includes target
100%
100%
Dispatch plan includes area path
100%
100%
Dispatch plan includes focus points
100%
100%
Sonnet model in dispatch plan
0%
100%
Deep-dive output format and synthesis conflict resolution
Deep-dive SUMMARY sections
0%
100%
Deep-dive ENTRY POINTS sections
25%
100%
Deep-dive EXECUTION FLOW sections
50%
100%
Deep-dive DATA TRANSFORMATIONS sections
50%
100%
Deep-dive INTEGRATION POINTS sections
25%
100%
Pricing representation documented
100%
100%
Conflict identified in synthesis
100%
100%
Conflict resolution in synthesis
100%
100%
Cross-cutting patterns section
100%
100%
Integration points in synthesis
75%
100%
File:line evidence throughout
0%
100%
No unsupported speculation
100%
100%
cf5a710
Table of Contents
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.