Unique-Divine/jiyuu / ai-skills/mori-graph
mori-graph
VerifiedCode graph for cross-repo context. Use Mori, a local code graph and code memory served as MCP server mori-graph (Tree-sitter parsing, LSP-style call resolution), for structural and cross-repo questions; install it from Mori releases with mise, index repos, and link GraphQL clients to gqlgen resolvers across repos. Triggers on who calls this, what does X call, trace a call chain, impact analysis, which client uses this API field or resolver, unused or deprecated API fields, cross-repo links, dead code, graph query syntax, Cypher examples, index a repo, .cbmignore, reindex, cross-repo intelligence, installing or upgrading Mori, and working on the Mori source in jiyuu/mori.
Mori Graph
Mori indexes repositories into a local graph (SQLite) and answers structural questions through MCP tools: who calls a function, what a change affects, which document section explains a symbol, and how services in different repos call each other. It parses with Tree-sitter, resolves calls with typed, LSP-style resolution where the language allows, and keeps the graph between sessions as code memory. A graph query costs far fewer tokens than reading files, but it's a lead, not proof: confirm material claims in source.
Mori lives in mori/ of https://github.com/Unique-Divine/jiyuu and ships as
mori/v* releases. It carries GraphQL cross-repo linking (a GraphQL client
operation links to the gqlgen resolver that answers it). The MCP server key
is mori-graph, so the tools are mcp__mori-graph__search_graph and so on.
The binary is still named codebase-memory-mcp until the Rust mori binary
ships; commands below use that name. Read setup.md to
install, register, and index; graphql.md for API
questions across repos; tool-source.md to work
on Mori itself.
Start of a Task
list_projects: pick the project whoseroot_pathcontains the working directory and pass itsnameasproject, never a folder name.index_statusfor that project. A stale or missing index answers wrong; reindex first (see setup.md).- Choose an evidence tier (below), then query.
Which Tool
| Question | Call |
|---|---|
| Who calls X? | trace_path(direction="inbound") |
| What does X call? | trace_path(direction="outbound") |
| Find a symbol | search_graph(name_pattern=".*Name.*"), then get_code_snippet |
| Exact text | search_code |
| Calls between services or repos | query_graph with Cypher on HTTP_CALLS, GRAPHQL_CALLS, CROSS_* |
| Effect of local changes | detect_changes() |
| Dead code | search_graph(max_degree=0, exclude_entry_points=true) |
| Repo overview | get_architecture |
trace_path needs exact names, so find them with search_graph first. Use
direction="both" when callers may live in another repo.
Evidence Tiers
- Scout: a quick positive lookup. Results are provisional; never claim absence or completeness from it.
- Verify (default): targeted searches, the relevant trace directions, snippets for every material claim, all result pages.
- Audit: a bounded scope checked completely, both call directions, with remaining gaps stated.
- Every tier: call
check_index_coveragewith the evidence paths before relying on the graph. For a "nothing uses this" claim, also include the scope. Partial, skipped, or stale coverage means read the source.
Results Are Paged
search_graph returns 50 per page and query_graph stops at an output
budget; both report has_more. Page until done before counting or claiming
absence. A short first page once looked like missing GraphQL links.
Cypher Examples
MATCH (a)-[r:HTTP_CALLS]->(b) RETURN a.name, b.name, r.url_path LIMIT 20
MATCH (a)-[r:GRAPHQL_CALLS]->(b) RETURN a.name, a.file_path, b.name LIMIT 50
MATCH (h)-[:HANDLES]->(r:Route) RETURN r.name, h.name LIMIT 20
search_graph(relationship="HTTP_CALLS") filters nodes by degree; to see the
edges themselves, use query_graph.
Delegating
Query the graph in the parent first, then pass the child the project name, index freshness, exact symbols and paths, findings, and open questions. A child without the MCP tools works from that evidence and reads source; it must not claim graph results it didn't get.
Upgrades
Install and upgrade through mise and the mori/v* releases, as in
setup.md. Never run codebase-memory-mcp install or
update: they rewrite this skill, register the server under the old key,
and fetch a build without Mori's GraphQL linking.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |