agentleFS
skixmixskixmix/AntanI/CLAUDE.md

AntanI

CLAUDE.md2 starsChanged 10 days agoMITView on GitHub

App name: AntanI. Use this name in code, commits, and docs. AntanI is a personal macOS desktop app that orchestrates CLI coding agents (OpenCode, Claude Code, Codex), terminals, and an embedded VS Code across local project folders — one window, a sidebar of projects, a tabbed workspace per project. It exists because the tool it replaces (Superset) is heavy: low RAM is the whole point. That single constraint drives the biggest technical choice: Tauri v2 on the system WebView. Never…

The file

We split this CLAUDE.md into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are 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.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Update *installation* stays out of scope too (no self-updating binary — the app is unsigned, so a silent self-replace would be a bad idea …

    line 29
  2. Distinctive:

    Minimalism / YAGNI: Prefer deleting features over abstracting.

    line 47
  3. Distinctive:

    No comments by default: Write zero comments unless the *why* is non-obvious to a future reader: a hidden constraint, a subtle invariant, a …

    line 73
  4. Distinctive:

    Caveats belong in CLAUDE.md, not just inline: When you hit a non-obvious constraint that would help future work in an area (a platform …

    line 77
  5. Distinctive:

    Test observable behavior and real logic, never implementation details or markup:

    line 88

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

AntanI — Agent Guide (root)

App name: AntanI. Use this name in code, commits, and docs.

What this is, and why

AntanI is a personal macOS desktop app that orchestrates CLI coding agents (OpenCode, Claude Code, Codex), terminals, and an embedded VS Code across local project folders — one window, a sidebar of projects, a tabbed workspace per project. It exists because the tool it replaces (Superset) is heavy: low RAM is the whole point.

That single constraint drives the biggest technical choice: Tauri v2 on the system WebView. Never introduce Electron or bundle Chromium — that would defeat the reason this project exists.

It serves one person on one or two Macs and ships unsigned via a personal Homebrew tap. There is no cloud, no account, no telemetry. Weigh every addition against that scope.

What we deliberately do NOT build

Saying no is a feature here. Out of scope: git worktree management, PR/task boards, session history/restore, cloud sync, Windows/Linux support, a plugin API, telemetry. Terminal tabs are session-only — never persisted, so every launch starts clean; only projects and settings persist. That asymmetry is intentional, not a missing feature.

Update installation stays out of scope too (no self-updating binary — the app is unsigned, so a silent self-replace would be a bad idea anyway). The one exception is a passive update check: src/lib/updateCheck*.ts hits the GitHub releases API once per launch and shows a badge in the status bar if a newer tag exists; clicking it opens the releases page. No download, no install, no background polling — installing the update is still brew upgrade by hand.

Code intelligence

This repo has a .codegraph/ index. Before grepping or reading files to understand or locate code, use codegraph first — the MCP tool codegraph_explore (or codegraph explore "<question>" in the shell). One query returns the relevant symbols' verbatim source plus the call paths between them, including dynamic-dispatch hops (renders, callbacks) that grep can't follow.

Rules

  • Minimalism / YAGNI. Prefer deleting features over abstracting. Add a dependency only when it clearly pays for itself. Duplication beats a premature abstraction. Fix bugs minimally — don't refactor while fixing.
  • Zero personal data in the repo. It is meant to go public. Never commit names, emails, credentials, API keys, tokens, absolute home paths, or machine-specific values. Use generic placeholders (e.g. /Users/foo/...) in tests and examples. projects.json lives in the OS app-data dir, never in the repo. The one intentional exception is the LICENSE copyright line, which names the author for attribution — don't "scrub" it.
  • Pin every dependency to an exact version (package.json with no ^/~, and Cargo.toml where practical) so the tool builds identically every time. Upgrades are deliberate, reviewed changes.
  • Verify current APIs before coding. Tauri v2, its plugins, and agent hook formats drift. Read current docs; don't code IPC/plugin/hook details from memory.
  • Never bypass the quality gate. The husky pre-push hook runs the fast checks — Biome (lint + format), tsc, knip, cargo fmt. The heavier checks (Vitest, clippy -D warnings, cargo test, coverage) run from their scripts and are the bar before shipping. No --no-verify; fix the cause, not the gate.
  • Tooling is fixed. bun is the package manager + runtime; Biome does both lint and format (Bun ships neither); knip guards against dead code. Don't add ESLint, Prettier, or a second formatter.
  • Clean code over clever code. Favor readable, self-explanatory code: descriptive names, small focused functions, early returns over nested conditionals, no unnecessary indirection. Code should read as what it does without needing a comment to explain it — see the comments rule below for the corollary.
  • No comments by default. Write zero comments unless the why is non-obvious to a future reader: a hidden constraint, a subtle invariant, a platform-specific workaround. Never narrate what the code does, never add section headers, never write docstrings that restate the function name.
  • Caveats belong in CLAUDE.md, not just inline. When you hit a non-obvious constraint that would help future work in an area (a platform quirk, a footgun, a "why this and not the obvious alternative"), write it up in the relevant CLAUDE.md — root or the nearest subdirectory one — rather than leaning on an inline comment alone. src/CLAUDE.md's "Tauri / WKWebView platform notes" is the model to follow. If a new area of the codebase (a new subsystem, a new top-level directory) accumulates its own conventions or caveats, create a CLAUDE.md for it instead of piling everything into the root file.

Testing philosophy: behavior, not brittle

Test observable behavior and real logic, never implementation details or markup:

  • Rust — unit-test the state and merge logic, above all anything that writes to the user's own files (the agent-hook config merge most of all: it must never drop or overwrite a user's existing hooks). Assert outcomes — which project is active, the resulting order, a save→load round-trip — not internal layout.
  • Frontend — Vitest covers pure logic only (src/lib). We deliberately write no React component / e2e tests and no Tauri IPC mocks: they are brittle, break on harmless markup changes, and are a maintenance burden a one-person tool shouldn't carry. The UI is verified by running the app.

Per-area rules live in src/CLAUDE.md and src-tauri/src/CLAUDE.md.

Commits

Conventional messages (feat:, fix:, chore:, …). Only commit when explicitly asked.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.