agentleFS
UtsavBalar1231UtsavBalar1231/oh-my-claudeagent/scripts/AGENTS.md

oh-my-claudeagent / scripts

AGENTS.md3 starsChanged 32 days agoMITView on GitHub

Repository tooling, the libraries it shares, the docs capture pipeline and a manual QA harness. No hook handler lives here: hooks are the mod under hooks/ and the server handlers under servers/hooks/. Every file is TypeScript on bun. - validate.ts and validate/: the plugin validator. bun scripts/validate.ts [--check <group>]... runs the groups claims, formats, hooks, mod, tree, engine and mcp (all of them by default), prints one line per check and a final Summary: line, and exits 1 on any…

  • Commits and pushes

The file

Files
AGENTS.md

Scripts

Repository tooling, the libraries it shares, the docs capture pipeline and a manual QA harness. No hook handler lives here: hooks are the mod under hooks/ and the server handlers under servers/hooks/. Every file is TypeScript on bun.

Layout

  • validate.ts and validate/: the plugin validator. bun scripts/validate.ts [--check <group>]... runs the groups claims, formats, hooks, mod, tree, engine and mcp (all of them by default), prints one line per check and a final Summary: line, and exits 1 on any failure. Each check family is one module under validate/ with its *.spec.ts beside it. validate/core.ts holds the shared pieces: the check types, Run for a finished child process, envWithout for a child environment, and the tracked-file listing. The depersonalization check scans every shipped file, and validate/allowlist.txt holds its exceptions; an entry that matches nothing fails. engine packages a copy of the tree and runs claude plugin validate --strict on it, and skips without the claude CLI. mcp runs the omca server through the handshake fixtures, and qa/install-verify.ts reuses it on a packaged copy. The tree group fails a tracked python, bash or sh script, by its shebang or by a .py, .sh or .bash name. The formats group reads every committed state fixture (registries, ledgers, plans, notepads) through the shared parsers; EXPECTED_INVALID in validate/formats.ts lists the fixtures that must keep failing. workflow.spec.ts parses .github/workflows/ and checks both directions: CI runs every command of every just ci recipe, and every command CI runs belongs to one of those recipes or carries a reason to run in CI only.
  • package.ts: bun scripts/package.ts <dest> copies the shipped tree, and --dry-run prints its file list. Only files git tracks ship. EXCLUDES keeps out the repository tooling: every *.spec.ts, tests/, scripts/ except scripts/setup-statusline.ts (KEEP), the root package.json, bun.lock, bunfig.toml, the tsconfig files, .oxlintrc.json, justfile, .pre-commit-config.yaml, .editorconfig, CONTRIBUTING.md, opencode/, .opencode/ and video/ and design-lab/. Claude Code installs npm packages whenever the plugin root holds package.json and a lockfile. A leading / in EXCLUDES anchors a pattern at the root. Packaging deletes whatever in <dest> is not on the shipped list, so it refuses a <dest> that is the repository or a directory holding it, and a non-empty <dest> that is not an earlier package (one whose .claude-plugin/plugin.json names oh-my-claudeagent). A spec checks that every relative import and every path a shipped file names resolves inside the shipped tree.
  • release.ts: bun scripts/release.ts <version>, behind just release <version>. It refuses a version that is not semver, a tracked change in the working tree, a checkout that is not main or not even with origin/main, a local plugin branch that does not contain origin/plugin, a version with no ## [<version>] heading in CHANGELOG.md, and an existing v<version> or plugin-v<version> tag. It writes the version into .claude-plugin/plugin.json, both version fields of .claude-plugin/marketplace.json and package.json, commits the bump and tags it v<version>. It checks that tag out into a temporary worktree, runs packageTree, and commits the result to the orphan plugin branch tagged plugin-v<version>, on top of the local branch or, in a fresh clone, of origin/plugin. A child commit of the bump rewrites the marketplace plugin source to a url source on ref: plugin with that commit's sha. It never pushes and prints one git push --atomic of both branches and both tags. A failure part way restores the manifests, resets to the starting HEAD with git reset --keep, deletes both tags and puts plugin back on its previous tip, then says so.
  • bench.ts: just bench. Runs the working tree or --candidate-ref against --baseline-ref through the mock model. It builds each ref in a git worktree. A baseline ref that has no statusline/main.ts is installed with uv sync, so uv must be on PATH for it.
  • benchmarks/compare/ (outside this directory, packaging-excluded): the harness behind just compare, which measures OMCA against similar plugins in Docker with a mock model. Its README has the stages and the files.
  • setup-statusline.ts: the status line setup behind /oh-my-claudeagent:omca-setup, and the one file under scripts/ that ships. It copies statusline/launcher.ts to the config directory, prints a diff of the settings file and, after confirmation, sets statusLine and subagentStatusLine. --uninstall removes them.
  • postedit-check.ts: runs just typecheck after an edit to a .ts file and reports the first lines of a failure as hook context. A contributor wires it as a PostToolUse entry in their own .claude/settings.json; CONTRIBUTING.md has the entry.
  • build-opencode.ts: compiles the terminal companion with OpenTUI's Solid plugin, leaving host packages external. just build-opencode writes opencode/tui.generated.js; --check refuses an artifact that differs from its sources. Git/package installs load this artifact because source transformation excludes node_modules.
  • docs/: the capture pipeline. screenshots.ts runs real Claude Code sessions in tmux against the mock model and attaches kitty to each one on a private Xvfb display. With no name it writes the README shots to .github/assets/: a PNG still per scene (ImageMagick import, optimized with magick) and a GIF for pane-tour and guard-dialog (ffmpeg x11grab, then palettegen and paletteuse). clips writes the video footage to video/public/footage/: per clip a BT.709 MP4 at 30 fps, a <clip>.json manifest of the frames at which its states first appeared (clip-manifest.ts, which video/src/footage.ts mirrors) and a <clip>.sheet.png contact sheet of those frames. Each session gets a scratch HOME and config directory with the built-in dark theme, except clip-board-light, which uses light, and fixtures/acme-app is the project it opens. Pixels cannot be masked afterwards, so privacy.ts checks the pane's plain text before and after each still and throughout each recording, and refuses the capture when it shows the username, the home path, the hostname or the scratch directory's name. It needs Xvfb, kitty, tmux, xdotool, ImageMagick, ffmpeg and ffprobe, and a JetBrains Mono font that fontconfig (fc-list) finds. An interrupted run tears its sessions and scratch directory down. just screenshots runs it.
  • qa/: the QA harness and the checks CI shares with it. just qa is the manual pre-release run. ci-smoke.ts is the CI smoke job and just smoke: one claude -p session with the packaged plugin loaded, in which the guard must deny a recursive removal (through PowerShell as well on Windows). junit-complete.ts reads the JUnit report of the CI bun spec run and fails unless the run has no failures and the report lists every spec file under the roots, so a crashed run that exits 0 still fails. session-smoke.ts, hook-live-probe.ts and worktree-bash.ts run claude -p, and mcp-live.ts runs an interactive session in tmux, each with the packaged plugin against the mock model (mock-model.ts), in a scratch project with its own CLAUDE_CONFIG_DIR. mcp-live.ts runs once per MCP handshake mode, against the real omca server with a fake ast-grep that sleeps. It checks the handshake in the client's debug log, the progress line on a running tool call, that Escape cancels the call and ends the fake process, and the title and badge of each tool in /mcp. install-verify.ts and statusline-probe.ts check the packaged tree itself. worktree-bash.ts and agent-effort.ts are steps of just qa. visual.ts backs just visual <view>, masks the scratch directory's random suffix in each capture, and exports the tmux helpers mcp-live.ts and docs/screenshots.ts share. lib.ts holds the shared helpers: REPO, checks, scratch directories, the claude -p launcher, the debug-log matchers and turn builders, signal-time cleanup and the real-config drift watch. eval-tasks.ts lists tests/evals/tasks/*.json.
  • The two bunfig.toml preloads live in tests/fixtures/: canonical-tmp.ts gives every spec the long form of TEMP on Windows, and plain-output.ts drops FORCE_COLOR. spec-preload.spec.ts checks the first.

Conventions

  • Scripts use erasable syntax and .ts import extensions.
  • A spec sits beside the script as *.spec.ts and runs under just test.
  • A spec or script that spawns a process passes env explicitly.
  • A script that runs on a contributor's machine uses node:fs and Bun.spawn argv, so it runs on Linux, macOS and Windows. The justfile recipes run in bash (Git Bash on Windows).

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.