agentleFS
RhovianRhovian/niles/AGENTS.md

niles

AGENTS.md4 starsChanged 43 days agoApache-2.0View on GitHub

Rust CLI that coordinates coding agents from different model families in tmux. Single crate, Unix only. Discover modules and commands from the source and niles --help rather than a list here. Machine-specific or private notes go in AGENTS.local.md (gitignored); read it if present. CI runs the checks on push; don't rerun them locally. Before reporting a change done, run only what CI doesn't cover, such as the focused tests for the code you changed, and say what printed. - Simple,…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository13 other files it gives its agents.

The file

We split this AGENTS.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
  1. Distinctive:

    Commit and push only when explicitly asked.

    line 12
  2. Very specific:

    No silent defaults on failure: propagate with ?

    line 33
  3. Distinctive:

    No catch-and-continue; invalid state stops.

    line 37
  4. Distinctive:

    No branches for impossible states.

    line 38
  5. Distinctive:

    Never copy-paste-tweak; extract a function or generic.

    line 44
  6. Distinctive:

    No pass-through functions unless they change visibility, satisfy a trait, or cross a module boundary under a name the caller's domain needs.

    line 45

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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
AGENTS.md

Niles

Rust CLI that coordinates coding agents from different model families in tmux. Single crate, Unix only. Discover modules and commands from the source and niles --help rather than a list here.

Machine-specific or private notes go in AGENTS.local.md (gitignored); read it if present.

Git

  • Branch for new work, named <issue>-<slug>. Don't commit to main unless told to.
  • Commit and push only when explicitly asked. Finishing a change is not a cue to commit it.

Checks

CI runs the checks on push; don't rerun them locally. Before reporting a change done, run only what CI doesn't cover, such as the focused tests for the code you changed, and say what printed.

Standards

  • Simple, idiomatic, first-principles. Find root causes; no shortcuts, temporary fixes, or "just in case" code.
  • Minimal diffs: touch only what the change needs.
  • APIs, internal ones included, are self-documenting and hard to misuse.
  • Names and types carry meaning. Comment only what the code can't say: why, invariants, non-obvious tradeoffs. No comments that restate the code.
  • For a non-trivial change, ask whether there is a more elegant way before presenting it. Don't over-engineer simple fixes.

Fail loudly

  • No silent defaults on failure: propagate with ? or fail fast.
  • A default is a decision. Make it at the parse boundary or as a named const, never inline at a use site. A site that truly owns the decision takes #[expect(clippy::disallowed_methods, reason = "…")].
  • No catch-and-continue; invalid state stops.
  • No branches for impossible states; encode impossibility in types.

Modularity

  • Source files stay under ~500 lines; split by responsibility, one module per job.
  • DRY by the rule of three; don't abstract at two occurrences.
  • Never copy-paste-tweak; extract a function or generic.
  • No pass-through functions unless they change visibility, satisfy a trait, or cross a module boundary under a name the caller's domain needs.

Correctness

  • Make invalid states unrepresentable with enums and newtypes. A newtype must carry weight: an invariant enforced at construction, a distinction between same-typed values, or impls the inner type can't have.
  • Parse, don't validate: type raw input (manifests, status lines, tmux output) once at the boundary.
  • debug_assert! invariants the types can't express.

Lints

Mechanically checkable rules live in [lints.clippy] in Cargo.toml and in clippy.toml, not in prose. The gate lints all targets. Suppress a lint with #[expect(…, reason = "…")], not #[allow]. The one exception is test code, which allows unwrap_used and expect_used at the crate root; tests/common may use #[allow(dead_code)], because each test binary uses a different subset of it.

Invariants

  • This repository's skills live in skills/ as ordinary disk skills. They are shareable instructions, so edit them with the same care as code.
  • Agent families and their launch mechanics are defined in src/agents/families.rs; models and effort levels in src/agents/roster.yaml.
  • niles wait reserves stdout for wake lines; diagnostics go to stderr. Its workspace exit codes (0 wake, including generated gone:, 11 superseded, 22 timeout) are a contract the lead relies on.
  • Changes that move a trust boundary (agent execution, workspace state, pane text) update the threat model.

To check new agent CLI releases or what Niles should change for them, follow CONTRIBUTING.md#agent-cli-compatibility.

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.