Manage Agent Skills from the command line or Python. Creates specification-compliant skills, installs from GitHub or dependencies, and keeps them up to date.
uvx skilly --help # Python (uvx/pip)
npx @xelandernt/skilly --help # Node (npx)
brew install xelandernt/skilly/skilly # Homebrewuvx skilly --helpShips CLI + Python import surface. Pre-built wheels for Linux x64, macOS arm64/x64, Windows x64.
npx @xelandernt/skilly --helpShips native Rust CLI (macOS arm64/x64, Linux x64 glibc, Windows x64). No Python import surface.
brew tap xelandernt/skilly https://github.com/xelandernt/skilly
brew install xelandernt/skilly/skillyShips native Rust CLI (macOS arm64/x64, Linux x64).
See the installation guide for a full capability comparison.
skilly create deployment-checks \
--description "Validate deployment readiness." \
--instructions "# Instructions\n\nRun the deployment checklist." \
--yes
skilly list| Command | Purpose |
|---|---|
scan |
Find skills provided by Python and Node project dependencies |
download <github-url> |
Install one or more skills from GitHub |
list |
Browse, update, or remove installed skills |
update |
Preview available updates; --yes applies all |
remove <name> |
Remove an installed skill by directory name |
skillsmp search <query> |
Search SkillsMP and install a selected result |
create |
Create a valid skill through a terminal wizard or explicit options |
configure |
Set which directories skilly manages via TUI or CLI flags |
Run skilly <command> --help for all options.
Run skilly --version to print the installed package version.
See creating skills.
See dependency scanning.
All management commands accept the same destination options:
uvx skilly list --local # .agents/skills
uvx skilly list --global # ~/.agents/skills
uvx skilly list --claude # .claude/skills
uvx skilly list --codex # .codex/skills
uvx skilly list --copilot # .github/skills (local), ~/.copilot/skills (global)
uvx skilly list --directory ~/custom # Explicit directory| Flags | Resolved destination |
|---|---|
| none | SKILLY_DEFAULT_DIRECTORY if set, otherwise .agents/skills |
--local |
.agents/skills |
--global |
~/.agents/skills |
--claude |
.claude/skills |
--claude --global |
~/.claude/skills |
--codex |
.codex/skills |
--codex --global |
~/.codex/skills |
--copilot |
.github/skills |
--copilot --global |
~/.copilot/skills |
--directory <path> |
That directory (after ~ expansion) |
Set a default destination:
export SKILLY_DEFAULT_DIRECTORY="$HOME/.config/skilly/skills"--directory overrides all other destination options and SKILLY_DEFAULT_DIRECTORY.
skilly configure lets you choose which directories skilly should manage and which one opens by default. Interactive terminals open a two-tab TUI (Global / Local) showing all known agent directories as toggleable checkboxes (agents, claude, codex, copilot). Non-interactive runs accept flags.
uvx skilly configure # Open the TUI
uvx skilly configure --show # Print current config as TOML
uvx skilly configure --reset # Restore defaultsIn the TUI:
- Space toggles a known directory on or off, or removes a custom one.
- Enter sets the highlighted directory as the default (marked with a star).
- Ctrl+S saves; you must have a default directory selected before saving.
Skill-selection menus support / to filter by skill name. Press /, type a substring, and the list narrows to matching items. Backspace edits the filter, Esc clears it.
Add or remove custom directories via CLI:
uvx skilly configure --add-global /opt/skills
uvx skilly configure --add-local .project/skills
uvx skilly configure --remove-global /opt/skills
uvx skilly configure --remove-local .project/skillsConfiguration is stored in ~/.skilly.toml:
default_directory = ".agents/skills"
[global]
directories = ["~/.agents/skills", "/opt/skills"]
[local]
directories = [".agents/skills", ".project/skills"]The default directory opens first in interactive menus (list, scan, etc.).
Set a token for higher API rate limits (first available wins:
SKILLY_GITHUB_TOKEN, GITHUB_TOKEN, GH_TOKEN):
export SKILLY_GITHUB_TOKEN=ghp_your_tokenAll GitHub-fetching commands also accept --github-token.
Full Python API reference with
SkillRepository, discovery functions, source types, SkillsMP client, and
custom filesystem protocol.
from pathlib import Path
from skilly import (
ProjectSettings,
PythonSource,
Skill,
SkillRepository,
)
repository = SkillRepository(
directory=Path(".agents/skills"),
project=ProjectSettings(
sources=(
PythonSource(
dependency_groups=("dev",),
optional_dependencies=("docs",),
),
),
),
)
repository.install(
Skill(
name="code-review",
description="Review code for correctness and maintainability.",
content="# Instructions\n\nReview the proposed change.",
)
)
for match in repository.scan_project():
print(match.available.name, match.status)Stateless discovery functions for one-shot reads:
from skilly import (
NodeSource,
PythonSource,
discover_installed_skills,
discover_package_source_skills,
)
installed = discover_installed_skills()
python_skills = discover_package_source_skills(PythonSource())
node_skills = discover_package_source_skills(NodeSource())ProjectSettings accepts PythonSource, NodeSource, and MavenSource:
from skilly import (
MavenSource,
NodeSource,
ProjectSettings,
PythonSource,
SkillRepository,
)
repository = SkillRepository(
directory=Path(".agents/skills"),
project=ProjectSettings(
sources=(
NodeSource(
include_dependencies=True,
include_dev_dependencies=False,
),
),
),
)Use ProjectSettings(sources=()) to disable scanning, or pass
individual PackageSource entries to scan only specific ecosystems.
SkillRepository() defaults to Python, Node, and Maven sources.
Maven skills are discovered from JAR artifacts in the local Maven
repository (~/.m2/repository by default). The scanner:
- Reads only direct
<dependencies>frompom.xml— profiles, plugins, and<dependencyManagement>are ignored. - Resolves
${property}references defined in the same file's<properties>block. - Loads skills from recognized archive layouts:
.agents/skills/<name>/SKILL.mdandskills/<name>/SKILL.md. - Preserves binary resources inside JARs.
- Rejects coordinates with path traversal components.
Known limitations:
- Only the local repository is used; no remote artifact resolution.
- No POM inheritance or effective-model merging.
- No Gradle build file support.
- Build execution and dependency graph traversal are not performed.
- Scopes are controlled via
include_*_scopeflags (default: compile, runtime, and test; provided and system are excluded).
SkillsMP client with typed results:
from skilly.skillsmp import ClientSettings, SkillsMp, SkillsMpSearchQuery
client = SkillsMp(settings=ClientSettings(base_url="https://skillsmp.com/api/v1"))
result = client.search(SkillsMpSearchQuery(text="python", limit=5))
print(result.data.skills[0].github_url)just install
just lint
just test
just typecheck