From ck
Queries symbol-level dependencies (IMPORTS, CALLS, EXTENDS, IMPLEMENTS, DEPENDS_ON) from graphify-out/graph.json for blast radius, task partitioning, and code research. Graceful grep fallback if missing.
npx claudepluginhub juliusbrussee/cavekitThis skill uses the workspace's default tool permissions.
A knowledge graph turns "what will break if I change this file" from a guess
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Guides agent creation for Claude Code plugins with file templates, frontmatter specs (name, description, model), triggering examples, system prompts, and best practices.
A knowledge graph turns "what will break if I change this file" from a guess into a query. Cavekit uses it only when it is available — nothing in the pipeline depends on it.
pip install graphifyy
graphify build . # writes graphify-out/graph.json
If the file is missing, this skill returns no-ops and every caller falls back to grep + ripgrep search.
NetworkX node-link JSON. Nodes are symbols (functions, classes, modules). Edges carry:
type: DEPENDS_ON | IMPORTS | CALLS | EXTENDS | IMPLEMENTSconfidence: EXTRACTED (high) | INFERRED (medium) | AMBIGUOUS (low)community: cluster ID (for partitioning big graphs into readable slices)cavekit-tools.cjs graph-status # is the graph present?
cavekit-tools.cjs graph-query --term auth # search by name substring
cavekit-tools.cjs graph-dependents --file X # who imports/calls this?
cavekit-tools.cjs graph-summary # top-level community list
(These subcommands are optional extensions; the base cavekit-tools.cjs
ships without them, and they activate only if graphify-out/graph.json is
present.)
/ck:sketch) — query existing symbols for a proposed kit name
to avoid collision with existing code./ck:map) — use community IDs to partition tasks into coherent
tiers. Two tasks whose affected symbols share no edges can run in parallel.ck:researcher) — query existing before fetching external.
If the graph already answers the question, skip the web./ck:make) — load the subgraph of the current task's files only.
Smaller context → faster, cheaper agent.Do not treat low-confidence edges (AMBIGUOUS) as true without verification.
When blast radius includes an AMBIGUOUS edge, fall back to grep and confirm.
When the graph is missing, every graph-* query returns:
{ "available": false, "fallback": "grep" }
Callers check available and fall back to Grep + Read. No error is
raised — this is a degradation, not a failure.