From karmaiq-architect
Surface architectural risk across the karmaIQ service mesh. Loads when the user mentions architecture review, structural risk, technical debt audit, pre-launch readiness, single points of failure (SPOF), circular dependencies, cycles, orphan services, dead code, hot CPU methods, or fan-in/fan-out analysis. Delegates full-domain audits to the karmaiq-architect subagent.
How this skill is triggered — by the user, by Claude, or both
Slash command
/karmaiq-architect:reviewing-system-architectureThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
The user wants to surface structural risk in the service mesh. This is deliberate work, not incident response — favor thoroughness over speed.
The user wants to surface structural risk in the service mesh. This is deliberate work, not incident response — favor thoroughness over speed.
Delegate to the karmaiq-architect subagent when:
The subagent runs all five analyses in one isolated pass and returns a consolidated risk table.
Stay inline when:
| User asked for… | Tool |
|---|---|
| Circular dependencies | analyze_architecture(analysis_type="cycles", max_length=10, max_results=20) |
| Single points of failure | analyze_architecture(analysis_type="spof", min_dependents=3) |
| Orphan services | analyze_architecture(analysis_type="orphans", traffic_threshold=0) |
| Most-depended-on interfaces | analyze_architecture(analysis_type="fan_in", top_k=20) or rank_interfaces(metric="fan_in") |
| Highest-fanout services | analyze_architecture(analysis_type="fan_out") |
| Dead code in a service | analyze_codebase_methods(service_name=..., strategy="dead_code") |
| Hot CPU methods in a service | analyze_codebase_methods(service_name=..., strategy="hot", cpu_threshold=1.0) |
Read active domain from ${CLAUDE_PLUGIN_DATA}/../karmaiq-core/domain.txt. Pass domain=... on every call.
## Finding
<1–2 lines summarizing the risk found — or "no risk surfaced">
## Evidence
<table or 3–6 bullets with concrete numbers — cycle lengths, dependent counts, dead-method counts, CPU %>
## Next step
<one concrete action: fix the top SPOF, refactor the largest cycle, remove top N dead methods, etc.>
For zero-finding cases, still produce the Finding ("No cycles ≤ length 10 in active domain") with Evidence (the query parameters).
When multiple findings exist, rank by:
get_system_overview first if not already called this session.node_ids byte-for-byte. Architecture analyses return canonical IDs; echo them as-is.${CLAUDE_PLUGIN_DATA}/../karmaiq-core/domain.txt. If unset, stop and tell the user to run /karmaiq-core:setup.analyze_codebase_methods returns empty for services without method-level instrumentation — say so, don't claim the service has no dead code.regression_diff (canary gate, not architecture)/api/foo/? and /api/foo/([^/]+)/? are distinct)analyze_architecture output without ranking / interpretationnpx claudepluginhub codekarma-tech-public/codekarma-mcp-plugin --plugin karmaiq-architectCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.