ix-understand

ix-claude-plugin shared model

Check command -v ix first. If unavailable, stop and say so.

Flag parsing

Parse $ARGUMENTS before doing anything else:

• The first non-flag token is TARGET (may be empty — means whole repo)
• --shallow: orient only, no agents (default if no flag given)
• --medium: single agent regardless of system count
• --deep: full parallel agent strategy
• --save [path]: if present, set SAVE_PATH; path is optional — if absent, auto-generate ix-understand-<target-slug>.md in cwd (target slug = TARGET with spaces and slashes replaced by -, or repo if TARGET is empty); if --save is not given at all, SAVE_PATH is empty.

MANDATORY for --medium and --deep: This skill MUST use the Agent tool (subagent_type: "ix-memory:ix-system-explorer") for all exploration work. Do NOT run ix commands yourself except for the Phase 1 orient commands below. All subsystem exploration MUST be delegated to agents.

Phase 1 — Orient

Run these commands in parallel to discover the architecture:

ix subsystems --format llm
ix subsystems --list --format llm
ix rank --by dependents --kind class --top 15 --exclude-path test --format llm
ix rank --by callers --kind function --top 15 --exclude-path test --format llm
ix stats --format llm

From the results, identify:

• All top-level systems (names, file counts, cohesion, coupling scores)
• The top 10-15 structurally important classes and functions
• Total codebase scale (files, nodes, edges)

If TARGET is set, scope the orient to that target's subsystems.

Confidence check: Scan confidence scores in the ix subsystems results:

• Any system with confidence < 0.5: add this caveat to the final output header: ⚠ Graph boundary confidence is low for [system] (${confidence}). Structural claims for this region may not reflect actual file relationships.
• Any system with confidence < 0.3: report fuzzy boundary as an explicit finding. Label all structural claims for that region as [uncertain].