agentleFS
kannandreamskannandreams/agent-top/.claude/skills/tuff-cli-guide/SKILL.md

tuff-cli-guide

Skill22 starsChanged 2 months agoMITView on GitHub

Reference for using Tuff CLI to manage agent capabilities — create, add (including MCP servers), list, diff, check, update, and scope operations.

  • Installs packages

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are 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.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run tuff --version. If the command is missing, ask the user before installing anything.

    line 12
  2. Very specific:

    If uv is not available, use brew tap kannandreams/homebrew-tuff && brew install tuff on macOS, or cargo install tuffcli where Rust is …

    line 18
  3. Very specific:

    If the project has no tuff.lock, run tuff init before anything else.

    line 22
  4. Distinctive:

    No separate source directory — agent files are the source

    line 100

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
SKILL.md

Tuff CLI Guide

Tuff is a capability lifecycle manager. It installs, versions, diffs, and validates the skills, tools, hooks, workflows, and MCP servers a repository gives its coding agents. Use these commands when the user asks about managing any of those, or when they mention "drift", "baseline", or "tuff".

Before You Start

Run tuff --version. If the command is missing, ask the user before installing anything. When they agree:

uv tool install tuffcli

If uv is not available, use brew tap kannandreams/homebrew-tuff && brew install tuff on macOS, or cargo install tuffcli where Rust is present. Avoid bare pip install: most systems refuse it as an externally-managed environment, and a virtual environment install is not on PATH afterwards.

Confirm with tuff --version before continuing. If the command is still not found, the install directory is not on PATH; ~/.local/bin is the usual one.

If the project has no tuff.lock, run tuff init before anything else. Never hand-edit tuff.lock or the files Tuff emits under a harness directory; edit the source under .agents/ and let Tuff re-emit.

Available Commands

Install

  • tuff add <path> -a <agent> — install a local capability (auto-detect type)
  • tuff add skill <path> [name] -a <agent> — install a skill
  • tuff add tool <path> [name] -a <agent> — install a tool
  • tuff add hook <path> [name] -a <agent> — install a hook
  • tuff add workflow <path> [name] -a <agent> — install a workflow
  • tuff add .agents/skills/<id> -a open-agents — track existing agent files in place
  • tuff add <git-url> skill <name> -a <agent> — install skill from git
  • tuff add <git-url> tool <name> -a <agent> — install tool from git
  • tuff add <git-url> hook <name> -a <agent> — install hook from git
  • tuff add skill <git-url> <name>@<version|range> -a <agent> — install a tagged release (1.2.0, ^1.2, >=1, <2); the lockfile records the tag, the requirement, and the commit
  • without @, a git install records the version the source declares (version in tuff.toml, or version:/metadata.version: in SKILL.md frontmatter), shown as 1.2.0 (declared) since it is weaker than a release; a source declaring nothing records the commit SHA
  • tuff create <type> <id> -a <agent> — create and track a capability
  • tuff add pack <artifact.tuffpack> -a <agent> — atomically install a verified capability pack
  • tuff add mcp <catalog-id|path|git-url>... -a <agent> — wire external MCP servers (catalog: filesystem, memory, github, fetch, git, time, sequentialthinking, everything, brave-search, notion, playwright, sentry, linear, context7) into each harness config; secrets stay as ${VAR} references. Prompts once per required variable at a real terminal (add --yes to skip, or it's automatic in a non-interactive shell)

Packs

  • tuff pack build --name <name> — package accepted project capabilities into tuff-dist/<name>-0.1.0.tuffpack
  • tuff pack build --name <name> --capability <id> --version <version> -a <agent> — build a selected project pack
  • tuff pack init <name> --from-project --capability <id> — save a reusable ID-based definition under tuff-packs/
  • tuff pack check [path] — validate a local source pack
  • tuff pack build [path] -o <artifact.tuffpack> — build a reusable project definition or standalone source pack
  • tuff pack inspect <artifact.tuffpack> — inspect verified metadata
  • tuff pack verify <artifact.tuffpack> — verify metadata and every stored file
  • tuff pack push <artifact.tuffpack> <registry/repository:tag> — publish exact pack bytes as an OCI artifact
  • tuff pack pull <registry/repository:tag-or-digest> -o <artifact.tuffpack> — download and verify an OCI-distributed pack without overwriting an existing file
  • tuff pack extract <artifact.tuffpack> -a <agent> -o <dir> — extract a runtime-native target

Inspect

  • tuff list — show all installed capabilities with drift status
  • tuff list --type skill — filter by capability type
  • tuff list --scope global — show global scope
  • tuff status — detailed status with override warnings
  • tuff diff <id> — show local changes against baseline
  • tuff diff <id> --upstream — show upstream changes (git sources only); for a release-pinned entry, against the newest release its requirement allows
  • tuff diff <id>@<range> --upstream — preview what a different requirement would change before tuff update <id>@<range>
  • tuff outdated — check all capabilities for available updates; repointed or tag missing on a release-pinned entry means the tag moved or vanished upstream, and tuff update <id> --force replaces the install with what the tag names now
  • tuff mcp doctor — spawn each installed MCP server for real and verify the handshake + tool list, not just that the config entry exists

Update & Merge

  • tuff update <id> --check — preview local baseline promotion or upstream changes
  • tuff update <id> — accept local edits or reconcile with upstream
  • tuff update <id> -a <agent> — update one recorded agent
  • tuff update <id> --force — overwrite local changes with source output
  • tuff update <id>@<range> — change a tag-pinned capability's requirement and move to the newest release it allows; a bare update never leaves the recorded range

CI & Validation

  • tuff check — validate all capabilities (exit 1 on any failure)
  • tuff check --json — machine-readable output for CI
  • tuff list --json, tuff outdated --json, tuff diff <id> --json — the same rows as JSON, with the same type/target/status keys as check --json; prefer these over parsing tables

Manage

  • tuff remove <id> — remove a capability
  • tuff agent list — show available agent harnesses
  • tuff agent add <id> — register an agent and initialize its project directory
  • tuff init --global — initialize global scope

Agents

  • open-agents — Codex, Cursor, OpenCode, Copilot, Gemini CLI, Roo, Cline
  • claude — Claude Code

Scope

  • Project (default): tuff.lock in repo root — committed with project
  • Global state: XDG config/state/cache Tuff directories — available across all projects
  • Project always wins when both exist

Directory Model

  • .agents/skills/ — create and edit skills here (single source of truth)
  • .agents/tools/ — create and edit tools here
  • .agents/hooks/ — create and edit hooks here
  • .agents/workflows/ — create and edit workflows here
  • .agents/mcp-servers/ — generated server.toml records for external MCP servers (edit the manifest, not these)
  • tuff.lock — committed capability identity and lifecycle metadata
  • tuff.config.json — project preferences; tuff.lock remains the project source of truth
  • No separate source directory — agent files are the source

Status Values

  • clean — installed content matches baseline
  • modified — local changes detected (run tuff diff <id> to see)
  • missing — installed file no longer exists

Quick Cheat Sheet

tuff init                              # initialize repo and register open-agents
tuff add <path> -a open-agents         # install capability (auto-detect type)
tuff add skill <path> [name]           # install a skill explicitly
tuff add mcp github -a claude         # wire a catalog MCP server
tuff list                          # check drift
tuff diff <id>                     # see what changed
tuff check                         # CI validation
tuff outdated                      # check for updates
tuff update <id>                   # accept local edits or update from source

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.