seekstone
CLAUDE.md35 starsChanged yesterdayMITView on GitHub
Guidance for AI coding agents (and people) working in this repo. AGENTS.md points here, so every tool gets the same rules. seekstone is an Obsidian MCP server, published to npm as seekstone. The server (packages/server) is the product: filesystem-direct vault access with a low context tax, shipped as a single-file bundle. The measurement harness (packages/harness) is a profiler, benchmark and write-safety suite that produced the numbers the server is designed around. Both share packages/core. The harness is dev-only and not…
The file
We split this CLAUDE.md into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 34
Only the server has a build.
- Distinctive: line 49
Use ordinary words. Skip "leverage", "robust", "seamless", "comprehensive", "delve", "crucial" and the like.
- Distinctive: line 50
No sales tone or inflated claims.
- Distinctive: line 51
Don't stage a reveal ("Here's the thing:"), don't build "not X, but Y" contrasts, and don't end on a one-line summary that repeats the …
- Distinctive: line 52
Avoid lists of three by reflex, piles of bold labels, and dashes used as punctuation everywhere.
- Distinctive: line 53
Commit messages are imperative ("add tag index tool").
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
Guidance for AI coding agents (and people) working in this repo. AGENTS.md points here, so every tool gets the same rules.
What this repo is
seekstone is an Obsidian MCP server, published to npm as seekstone. The server (packages/server) is the product: filesystem-direct vault access with a low context tax, shipped as a single-file bundle. The measurement harness (packages/harness) is a profiler, benchmark and write-safety suite that produced the numbers the server is designed around. Both share packages/core. The harness is dev-only and not published.
The claim the project is judged on: filesystem-direct beats REST-proxy, and the win is mostly payload size ("context tax"), not raw CPU. Speed matters as much. Don't trade a benchmark regression for a feature.
Where the rules live
| Working on | Read |
| Server tools, writes, dispatch, the admission bar for new tools | packages/server/CLAUDE.md |
| Shared primitives (walk, frontmatter, extract, embed) | packages/core/CLAUDE.md |
| Harness commands, the benchmark vault, env vars | packages/harness/CLAUDE.md |
| Package graph, server layers, request flow | docs/ARCHITECTURE.md. Keep it in sync when you add a layer, tool or adapter. |
| Setup, CI gates, PR checks, changesets | CONTRIBUTING.md |
| Write safety guarantees | docs/WRITE-SAFETY.md |
Commands
npm test # all workspaces (vitest)
npm run lint # biome check
npm run format # biome write; stage every file it changes
npm run build -w seekstone # bundle the server to dist/ (tsup)
npx tsc -p packages/server/tsconfig.json --noEmit
npx vitest run packages/server/src/tools/search.test.ts
npm run bench -w seekstone # hot-path benchmarksOnly the server has a build. The harness runs from source through tsx; tsc is typecheck only.
Conventions
- Relative imports use
.jsextensions, even for TypeScript sources. NodeNext, tsx and tsc all agree on that;.tsextensions are avoided. - Tests ship with the change, co-located as
*.test.ts. Don't delete, skip or loosen tests to get green; CI flags it. - Never mutate a real vault. Writes go through the safety harness or an adapter pointed at a scratch copy. Tests use temp copies or the committed fixtures.
- No telemetry and no network calls on the serving path.
no-network.test.tsguards it. - Changesets name only
seekstone; the other packages are private.
Writing
Docs, PR descriptions, commit messages, changesets and code comments are read by people. Write them plainly:
- Say what changed and why, in short sentences. Lead with the point.
- Use ordinary words. Skip "leverage", "robust", "seamless", "comprehensive", "delve", "crucial" and the like.
- No sales tone or inflated claims. Give the number instead of an adjective.
- Don't stage a reveal ("Here's the thing:"), don't build "not X, but Y" contrasts, and don't end on a one-line summary that repeats the paragraph.
- Avoid lists of three by reflex, piles of bold labels, and dashes used as punctuation everywhere.
- Commit messages are imperative ("add tag index tool"); conventional prefixes are welcome.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

