pm-graph

npx skills add majiayu000/claude-skill-registry --skill pm-graph --agent claude-code

Same command for any agent — swap --agent for codex, cursor, copilot.

## Runtime Configuration (Step 0) Read `ops/derivation-manifest.md` for vocabulary. --- ## EXECUTE NOW **Target: $ARGUMENTS** Parse immediately: - If empty: generate overview graph showing all decisions and their connection counts - If target is a decision name: generate full lineage trace for that decision - If "clusters": identify topically clustered decisions - If "central": find the most-connected decisions (highest graph centrality) **START NOW.** --- ## Philosophy **The graph reveals what individual notes cannot: the shape of institutional memory.** A PM system with 50 decisions and 200 connections between them is fundamentally different from one with 50 decisions and 12 connections. The first has compound institutional memory — you can reason about relationships. The second has an indexed list. /pm-graph makes the connection density visible. High-connection nodes are architectural load-bearers — changing them has cascade effects. Low-connection nodes are either orphans (structural failure) or genuinely isolated facts (acceptable). Clusters reveal coherent reasoning zones. Bridges between clusters reveal architectural principles that span domains. This analysis runs without s

For each decision file, count outgoing links
for f in decisions/*.md; do
echo "$links $basename"
done | sort -rn | head -30
Find the decision file
Find all decisions that LINK TO this decision
rg "\[\[$target_file\]\]" decisions/ --include="*.md" -l
Find all decisions this decision LINKS TO
rg "\[\[([^\]]+)\]\]" "decisions/$target_file" -o --no-filename
For each register, get the decisions it contains