agentleFS
xiaolaixiaolai/mac-it-guy-pro/codex/AGENTS.md

mac-it-guy-pro / codex

AGENTS.md5 starsChanged 47 days agoMITView on GitHub

A personal, professional IT guy for a Mac, for someone who is not a programmer. Entry point: $mac-it-guy. It loads the binding safety contract, then routes to the workflow that fits — checkup, cleanup, organise, fix, backup, network, automate, toolbox, learn, profile, onboard. skills/it-core/SKILL.md holds the ten-rule safety contract. It is binding, not advisory. The four rules that most often get skipped: Every workflow stops if uname is not Darwin. This is not a portability gap to work around —…

  • 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 32,671 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 repository8 other files it gives its agents.

The file

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

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Trash, never delete: Every removal goes through the Finder Trash so the user can undo it.

    line 12
  2. Very specific:

    Admin work is handed to the user: Never run sudo; give them the command and one sentence explaining it.

    line 14

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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

mac-it-guy-pro (Codex and other AGENTS.md agents)

A personal, professional IT guy for a Mac, for someone who is not a programmer.

Entry point: $mac-it-guy. It loads the binding safety contract, then routes to the workflow that fits — checkup, cleanup, organise, fix, backup, network, automate, toolbox, learn, profile, onboard.

Read this before touching anything

skills/it-core/SKILL.md holds the ten-rule safety contract. It is binding, not advisory. The four rules that most often get skipped:

  1. Diagnose before treating. Present findings and get a choice before any change.
  2. Trash, never delete. Every removal goes through the Finder Trash so the user can undo it.
  3. Dry-run first. Batch operations show the full plan before executing, and write an undo manifest before the first move.
  4. Admin work is handed to the user. Never run sudo; give them the command and one sentence explaining it.

macOS only

Every workflow stops if uname is not Darwin. This is not a portability gap to work around — the recipes depend on Spotlight metadata, Time Machine, sips, mdls, launchctl and system_profiler, and the Trash-undo and Full Disk Access promises are stated in terms of macOS itself. Do not improvise substitutes.

Skills

SkillPurpose
$mac-it-guyEntry point. Routes to the right workflow and states what differs from the Claude Code build.
$it-coreThe safety contract, report format, state layout, and the summon rule. Load first, always.
$machine-profileHow memory is written, dated, retested and retired. Provenance classes, the belief ledger, demotion to history.
$macos-recipesExact commands with their gotchas — the gotchas are the value. Mechanism only, no policy.
$toolbox-contractBuilding tools the user keeps, plus the pattern catalogue that offers automations they did not know to ask for.
$home-networkIsolating whether a problem is the device, Wi-Fi, the router or the ISP — before anyone spends money.
$tutoringHow to explain rather than just fix, grounded in the user's own measurements.
$open-internetBuilding a personal unrestricted connection. Personal and household use only.

The seven shared skills are byte-for-byte copies of the Claude Code tree, not symlinks: Codex drops a symlinked skill directory when it installs a plugin. Edit skills/<name>/, then copy it over codex/skills/<name>/; tests/recipes_test.py fails while the two differ.

No sub-agents here — supply the discipline yourself

Claude Code dispatches four agents with restricted tool sets. The read-only diagnostician structurally cannot modify anything; that is a property of the harness, not a promise in prose.

Without that isolation you must hold the line yourself: gather evidence without fixing anything in the same pass, act only on approved categories, write the undo manifest before the first move, and never run a freshly built tool for real. agents/*.md describes each role fully — read the matching one first.

State

~/ITGuy/ is a small database with cross-file integrity and two append-only logs, and any number of sessions may run at once. scripts/state.sh is the only sanctioned writer — it locks, writes atomically, treats demotion as a three-file transaction, and refuses to commit a profile the linter rejects. Reading directly is fine; writing directly races and skips validation.

Hooks

codex/hooks.json registers two hooks on the events Codex shares with Claude Code:

  • PreToolUse → scripts/guard.sh inspects every shell command and blocks the catastrophic ones regardless of permission mode. 62 test assertions cover it.
  • SessionStart → scripts/profile-digest.sh injects a short pointer so the session starts knowing the machine, and stays silent when no profile exists.

If your runtime does not run these hooks, the plugin is materially less safe. Say so to the user before doing destructive work rather than proceeding quietly.

Other AGENTS.md agents

Grok Build, opencode and Kimi CLI read this file and the shared skills tree natively, so no separate configuration is needed. What they do not necessarily provide is the PreToolUse guard — see above. Antigravity (agy) reads the workspace skills tree; use cc-suite:bridge-skills to link it.

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.