DanielC000/loom / packages/daemon/assets/skills/codescape
codescape
VerifiedHow Loom agents should use Codescape (the code-graph MCP) when a project exposes it — the graph for structure, coordinates, and reachability (orient the shape, get exact file:line, ask "what could render / what scenarios are possible"), then a small number of targeted reads at those coordinates. NOT a nicer index to grep from; do not re-orient by reading. Shipped and kept current by Loom.
codescape
Codescape is a code-graph MCP — a per-project, ingested snapshot of the repo's structure. When your
project exposes the Codescape MCP tools (overview / list_flows / describe_symbol / what_touches /
trace_flow / boundary_map / render_tree / scenario_space), it is your PRIMARY orientation
instrument, not a fallback.
The one thing to internalize: use the graph for structure, coordinates, and reachability, then open files ONLY to read the exact bytes you must verify or edit — at the coordinates the graph gave you. The graph REPLACES orient + locate; reading is only confirm/edit. If you re-derive the map by reading, or grep to LOCATE what the graph already pinpoints, you throw the whole point away — a measured speed / fewer-reads win collapses back into verify-by-reading.
Use the graph for three things
- ORIENT — the shape. What exists and how flows run through the system end to end (frontend + backend,
across parts/services). Reach a correct mental map with FAR fewer reads than opening files. Start here
(
overview/list_flows/trace_flow) before you open anything. - LOCATE — the coordinates. Which
file:lineeach piece lives at.describe_symbol/overview/what_touches/trace_flowreturn citation-gradefile:line— take the exact coordinates FROM the graph; don't grep to find what it already pinpoints. - REACHABILITY — what could happen. "What could render" (components + conditional branches) via
render_tree, and "what scenarios are possible" for one component — its loading/error/ready state space across data and mutation states — viascenario_space, which replaces hand-tracing every conditional yourself. Grep is actively BAD at these; the graph answers questions reading-by-hand can't.
Then read — targeted, at the coordinates
Open Read/grep ONLY for the irreducible step: the actual bytes you must verify or edit, opened DIRECTLY at
the file:line Codescape gave you — never searched for from scratch. A few pinpoint reads, not a re-explore.
Honest boundaries (don't oversell it)
Codescape is complementary, not a replacement — grep/Read still win for:
- Exact-string / literal search — a specific token, error string, config value: grep it.
- Freshest state — the graph is only as fresh as the last ingest; for code that just changed, confirm against the file.
- Non-modeled content — comments, config values, prose, anything that isn't a code symbol: grep.
The two failure modes this doctrine exists to kill
- Ignoring the graph and orienting by reading — opening files to build the map the graph would hand you.
- Treating the graph as a fuzzy index, then grepping from scratch anyway — a vague sense, then re-locating by hand.
Target behavior: graph for structure + coordinates + reachability → a few targeted reads.
Not the same as graphify
If your repo also has /graphify (tree-sitter, forward-only), keep them distinct. graphify is a
forward-orientation aid that under-reports reverse dependencies (never trust its affected). Codescape is
richer — it does reverse/reachability AND returns coordinates — so its reachability answers are trustworthy
where graphify's aren't. Don't carry graphify's reverse-dependency caveat over to Codescape.
Maintainer note (not part of using this skill)
If you are EDITING this directory: prompt-block.md in this same directory is a cross-project
measurement instrument — its text must not change without following the standing commitment recorded in
project memory codescape-prompt-block-is-a-measurement-instrument.
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: 468k | 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: 321k | 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: 321k | 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: 321k | Updated | Claude CodeCodex | Verified | Install |