Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/sdsrss/code-graph-mcp/windows-compatnpx skills add sdsrss/code-graph-mcp --skill windows-compatgit clone --depth 1 https://github.com/sdsrss/code-graph-mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00165 | $0.02205 |
| Opus 5 | $0.00082 | $0.01103 |
| Sonnet 5 | $0.00033 | $0.00441 |
| Haiku 4.5 | $0.00016 | $0.00220 |
Grade A, and why
windows-compat scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 196 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Windows compatibility (code-graph-mcp)
Three defects reached a Windows user from one root cause: two spellings of
the same path compared unequal. CI already ran windows-latest and caught none
of them, because nothing asserted on path spellings at all. Closing that gap is
what this skill is for.
The invariant everything hangs off
indexer::merkle::normalize_rel_path stores every indexed path as a
/-separated relative path, rewriting separators only under
#[cfg(windows)]. So any path that becomes a lookup key, a dedup key, or
stdout must land in exactly that spelling.
Almost every defect below is one producer handing over a different spelling of the same file. The producer table and the full catalogue are in references/failure-modes.md — read it before changing path-handling code. It is short and every entry is a real defect with its shipped fix.
The doctrine: platform is a parameter, not a cfg!
This is the single rule that makes everything else testable.
// NO — the Windows branch only runs on Windows, so the Linux and macOS legs
// prove nothing, and a Windows-only CI failure is expensive to debug.
fn normalize(path: &str) -> String {
if cfg!(windows) { path.replace('\\', "/") } else { path.into() }
}
// YES — one thin cfg! entry point, all logic in a function taking the platform.
fn normalize(path: &str) -> String {
normalize_on(path, cfg!(windows))
}
fn normalize_on(path: &str, backslash_is_sep: bool) -> String { ... }
Then assert both values from any platform:
assert_eq!(normalize_on(r"src\a.rs", true), "src/a.rs"); // Windows behaviour
assert_eq!(normalize_on(r"src\a.rs", false), r"src\a.rs"); // `\` is a legal Unix filename char
Live examples: merkle.rs::normalize_rel_str_on (the crate's single
separator-rewriting implementation — everything else delegates to it),
cli.rs::normalize_path_display_on, cli.rs::relativize_path_on.
Skip dunce / path-slash here. They do the right thing but decide internally
on cfg(windows), forfeiting exactly the property that makes the Linux leg
useful.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 196 lines · 165 tokens per session scan A 65b3b3afdfd2
windows-compat is a skill published in the GitHub repository sdsrss/code-graph-mcp (70 stars, last pushed 6d ago), licensed MIT. It adds 165 tokens to every session and 2,205 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
gograph
Go repository intelligence for Claude Code. Use when reading, navigating, editing, reviewing, or refactoring a Go codebase. Exposes 64 query, analysis, and workflow capabilities through the local gograph MCP server, including bounded first-call exploration, AST-aware call graphs, blast-radius analysis, impact, and…
gossiphs
Use when you need to find related files, understand code dependencies, or analyze impact scope. Triggers when the user asks about: "Which files use this function/class?" "What files will be affected if I change this?" "Find code related to X" "What does this file depend on?" "Where is this symbol defined/referenced?"…
frigg-first-code-search
Prefer Frigg MCP for indexed source. resulthandle + readmatch for proof.
test-harness
Test codeweave-mcp tools against any project using TestHarness. Use this whenever verifying tool behavior, testing a code change, or running custom assertions — e.g. "test the harness", "run tests against X project", "does semanticsearch work correctly".
risk-assessment
Assess project/domain health — scan risk zones across dimensions (bugs, hotspots, tech debt). Use when asked evaluate risks, find problematic areas, assess code health, identify zones needing attention — NOT for specific bug symptoms (use bug-hunt instead).
executing-plans
Execute written implementation plan whose Tasks edit code, per-Task SAFE/CAUTION/UNSAFE git-signal verdict before edit AND code-style cascade for code-generation Tasks (style from silo authors, strategy+template from proven neighbors). Triggers on "execute the plan", "start Task N", "выполни план", "начни задачу"…