| name | traverse-template |
| description | Navigation protocol template for skill graphs. Copy and adapt per domain to teach agents how to efficiently traverse MOC hierarchies. Agent-only — not user-invocable. Use when building a new skill graph that needs a traversal protocol. |
| metadata | {"author":"shihwesley","version":"1.0.0"} |
Skill Graph — Traversal Protocol
You are navigating a skill graph for development. The graph is at .claude/docs/<slug>-graph/.
How to Navigate
Step 1 — Read the index. Read <slug>-graph/index.md. The TLDR hook serves a summary with all domain MOCs and their descriptions. Pick 1-2 MOCs relevant to your current task.
Step 2 — Read the relevant MOC(s). Each MOC lists skills, docs, and agents with file paths and one-line descriptions. Pick the 2-3 items you actually need.
Step 3 — Read the target files. Now read the specific skill or doc files the MOC pointed you to. These have the operational knowledge.
Step 4 — API details (if needed). For exact API signatures or reference lookups, search the domain knowledge store:
cd ~/Source/neo-research && .venv/bin/python3 scripts/knowledge-cli.py search "<query>" --project <store-name> --top-k 5
Rules
- Never load all MOCs. Pick only what matches the task.
- Never read files the MOC didn't point you to. The graph already filtered for you.
- Don't re-read files you just wrote or edited. You know the contents.
- Follow
[[cross-references]] only when they're relevant. MOCs link to other MOCs — follow those links only if your task spans domains.
- TLDR summaries are your first pass. If the hook serves a summary, scan it before requesting full content. Most navigation decisions happen at the summary level.
Cost Budget
A well-navigated traversal costs ~3-4k tokens:
- Index summary: ~400 tokens
- 1-2 MOC summaries: ~600-1200 tokens
- 2-3 target file reads: ~1500-2500 tokens
Compare to loading all docs linearly: ~25k+ tokens. The graph pays for itself on the first task.
When to Skip the Graph
- If you already know the exact file path — read it directly
- If the task is trivial and targeted — no navigation needed
- If the user tells you which skill to load — load it, don't traverse
The graph is for discovery, not ceremony.
