agentleFS
PaulRBergPaulRBerg/agent-skills/toolkit/AGENTS.md

agent-skills / toolkit

AGENTS.md104 starsChanged 19 days agoMITView on GitHub

This workspace lives at toolkit/ inside the agent-skills catalog repository. The catalog skill rules in the root AGENTS.md govern skills/, not the toolkit. The toolkit follows this file and its package-local AGENTS.md files. Toolkit commits use the catalog's natural-language commit format. - The nightly Rust workspace contains the ai-commit, ai-coord, ai-handoff, ai-notify, and ai-skillet crates. Keep shared Rust configuration at the workspace root (toolkit/) and crate behavior within its crate. - apps/coord-dashboard is an independent Bun package with its own…

The file

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

10 rules. About 0 are common, 4 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Keep shared Rust configuration at the workspace root (toolkit/) and crate behavior within its crate.

    line 10
  2. Very specific:

    Use toolkit/justfile for local validation (just <recipe> inside toolkit/, or just toolkit::<recipe> from the catalog root).

    line 20
  3. Distinctive:

    Keep the dashboard local-only and bound to loopback, with no external requests.

    line 52
  4. Very specific:

    Never edit a version by hand.

    line 66
  5. Distinctive:

    Before a task ends, release each CLI that the task changed in a user-facing way.

    line 90
  6. Distinctive:

    Commit the change normally.

    line 99
  7. Distinctive:

    Write the entry at the top with today's date.

    line 105
  8. Very specific:

    Commit the three files with the message Release <crate> <version>, then push.

    line 107
  9. Very specific:

    Run just release-publish <crate>.

    line 108
  10. Very specific:

    Follow the install-cli rules in Compatibility and safety.

    line 109

≈ “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

Toolkit

This workspace lives at toolkit/ inside the agent-skills catalog repository. The catalog skill rules in the root AGENTS.md govern skills/, not the toolkit. The toolkit follows this file and its package-local AGENTS.md files. Toolkit commits use the catalog's natural-language commit format.

Workspace boundaries

  • The nightly Rust workspace contains the ai-commit, ai-coord, ai-handoff, ai-notify, and ai-skillet crates. Keep shared Rust configuration at the workspace root (toolkit/) and crate behavior within its crate.
  • apps/coord-dashboard is an independent Bun package with its own lock and package-local validation. Do not combine its dependencies, scripts, or build outputs with the Rust workspace.
  • xtask is unpublished maintainer tooling for changelog validation and releases. It is not a CLI, and the installation recipe never installs it.
  • Context is source-owned. This file describes workspace-wide behavior. Package-local AGENTS.md files own product and workflow guidance. Update the owning package rather than duplicating its guidance here.

Validation

Use toolkit/justfile for local validation (just <recipe> inside toolkit/, or just toolkit::<recipe> from the catalog root). Run focused tests while iterating, then the complete affected gate before committing:

Change scopeFinal local gate
Rust code, Cargo dependencies, Rust toolchainjust rust-check
Coordination dashboardjust coord-dashboard-check
Shared workspace wiring or Rust/app contractsjust check

just rust-check checks formatting, Clippy with warnings denied, tests, builds, changelog consistency, and platform-specific installation tests for the locked Rust workspace. Installation tests use mock commands and do not install binaries. The dashboard gate checks lint, formatting, types, tests, and the production build. UI changes also need the owning package's rendered verification. just check runs both gates. None of these checks installs CLI binaries. Documentation-only edits follow the root formatting and factual-verification rules.

In Linux containers, preserve Git repository metadata and use an init process, such as Docker's --init option. The ignore-rule fixtures require repository metadata. Process-group tests require orphaned child processes to be reaped.

Compatibility and safety

  • Maintain macOS and Linux support for ai-commit, ai-coord, ai-handoff, and ai-skillet. Keep ai-notify macOS-only. The installation recipe selects supported packages and restarts an existing launchd service only on macOS.
  • Preserve each tool's documented command, output, and persisted-data contracts. Do not add compatibility paths, migrations, aliases, or dual formats unless explicitly requested. Reject incompatible persisted versions with actionable errors.
  • Changes that can invalidate live ai-coord agents, state, hooks, or CLI installation require explicit authorization and an isolated AI_COORD_STATE_DIR for development and validation. Never reset coordination state or replace global hooks from this checkout implicitly.
  • Updating the installed CLIs (just install-cli, or just toolkit::install-cli from the catalog root) is pre-authorized when the committed changes preserve the documented command, output, and persisted-data contracts. It is also pre-authorized when ai-coord shows no other live agent sessions on this machine. Otherwise ask the user to close the other agents before installing.
  • Keep the dashboard local-only and bound to loopback, with no external requests. Its handoffs view must remain read-only.

Versioning and releases

Model

  • Each CLI has an independent SemVer version. The [package] version in its Cargo.toml is the source of truth. The top entry of its CHANGELOG.md mirrors that version.
  • CHANGELOG.md follows Common Changelog. Write each entry in imperative mood (Add, Fix, Bump). Write one line per change. Give at most two references and keep the best. Do not add an Unreleased section or authors.
  • Each release has the tag <crate>-v<version>, for example ai-coord-v1.0.0, and one GitHub release for that tag. Do not publish prereleases.
  • Never edit a version by hand. Only just release-prepare changes versions.
  • just changelog-check enforces these rules, and just rust-check runs it:
    • The file starts with # Changelog.
    • Release headings are ## [X.Y.Z] - YYYY-MM-DD. Use stable versions only, newest first.
    • Each version is the exact next major, minor, or patch step. A **Breaking:** item requires a major step. An ### Added group requires at least a minor step.
    • A release body has an optional single-line italic notice first. Groups follow in this order: ### Changed, ### Added, ### Removed, ### Fixed.
    • A group holds only one-line - items. An item may start with **Breaking:**. Breaking items come first in the group.
    • Each item starts with an uppercase letter and ends with a references parenthetical. The parenthetical holds at least one commit link, for example ([`abc1234`](https://github.com/PaulRBerg/agent-skills/commit/abc1234)).
    • The file ends with one definition per release: [X.Y.Z]: https://github.com/PaulRBerg/agent-skills/releases/tag/<crate>-vX.Y.Z.

SemVer for CLIs

  • Major: a breaking change to a documented command, flag, output, exit code, hook, or persisted-data contract. This includes rejecting previously accepted persisted data.
  • Minor: a backward-compatible new command, flag, output field, or capability.
  • Patch: a fix, a performance change, or a runtime dependency bump with no contract change.

When to release

Before a task ends, release each CLI that the task changed in a user-facing way. A change is user-facing when it changes installed-binary behavior or a documented contract. One release may cover several commits from the task.

  • Tests, internal refactors, docs, and routine dependency bumps do not trigger a release. Notable ones ride along in the next entry. Omit dev-only changes.
  • A shared workspace change that alters runtime behavior releases every affected CLI.

Steps

  1. Commit the change normally.
  2. Acquire ai-coord scopes for toolkit/<dir>/CHANGELOG.md, toolkit/<dir>/Cargo.toml, and toolkit/Cargo.lock. After READY, re-read the changelog.
  3. Draft from the history since the tag of the current version. Run this command from toolkit/: git log --format='%h %s' <crate>-v<current>..HEAD -- <dir> Cargo.toml Cargo.lock rust-toolchain.toml. Include only commits that no entry references yet.
  4. Write the entry at the top with today's date. Add its link definition at the top of the definitions.
  5. Run just release-prepare <crate>.
  6. Commit the three files with the message Release <crate> <version>, then push.
  7. Run just release-publish <crate>. A rerun completes a partial publish.
  8. Follow the install-cli rules in Compatibility and safety.

Pushing release tags and creating GitHub releases with just release-publish is pre-authorized, as is installation.

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.