agent-plugins 0.0.0__tar.gz → 0.1.0__tar.gz

agent_plugins-0.1.0/.agent-plugin/plugin.json

@@ -0,0 +1,11 @@
+{
+  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
+  "name": "agent-plugins",
+  "description": "Ship Agent Plugins with Python packages and inspect their installed files.",
+  "author": {
+    "name": "Péter Ferenc Gyarmati",
+    "email": "dev.petergy@gmail.com"
+  },
+  "repository": "https://github.com/peter-gy/agent-plugins",
+  "keywords": ["python", "agent-plugins", "agent-skills"]
+}

agent_plugins-0.1.0/.agent-plugin/skills/agent-plugins/SKILL.md

@@ -0,0 +1,195 @@
+---
+name: agent-plugins
+description: Ship Agent Skills, MCP server configuration, and extension files with a Python package, choosing build-time or runtime access as needed. Use when adding an Agent Plugin to a Python project, configuring uv_build or Hatchling, locating installed plugin and skill paths, traversing skill files, reading manifest, skill source, or MCP values, or verifying wheel, source distribution, and editable installs.
+---
+
+# Agent Plugins
+
+Use `agent-plugins` when a Python package should carry the instructions and MCP
+configuration that match its installed code version.
+
+An [Agent Plugin](https://agent-plugins.org/) is an open, vendor-neutral
+portable directory format for reusable agent components. Its fixed locations
+let compatible clients find [Agent Skills](https://agentskills.io/specification)
+and [Model Context Protocol (MCP)](https://modelcontextprotocol.io/specification)
+server configuration in the same package. Distribution, permissions, and user
+experience remain with each client.
+
+## Build the plugin tree
+
+Keep one Agent Plugin tree in the codebase:
+
+```text
+my-plugin/
+├── plugin.json
+├── skills/
+│   └── use-my-package/
+│       ├── SKILL.md
+│       ├── scripts/
+│       └── references/
+├── mcp.json
+└── com.example.client/
+    └── hooks/
+```
+
+- `plugin.json` identifies the plugin and its Agent Plugins schema.
+- `skills/` contains Agent Skills and their nested files.
+- `mcp.json` describes stdio, Streamable HTTP, or legacy HTTP+SSE servers.
+- Reverse-domain directories contain client-specific extension files.
+
+Create a minimal `plugin.json` at the plugin root:
+
+```json
+{
+  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
+  "name": "my-project"
+}
+```
+
+Find the Python package's `pyproject.toml` and set
+`[tool.agent-plugins].root` relative to that file. For uv_build:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "uv_build==0.12.2"]
+build-backend = "agent_plugins.build.uv_build"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+For Hatchling:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "hatchling==1.31.0"]
+build-backend = "agent_plugins.build.hatchling"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+The build selects `plugin.json`, the complete `skills/` tree, and `mcp.json`
+when present. Select other root-relative files explicitly:
+
+```toml
+[tool.agent-plugins]
+root = "../.."
+include = ["bin/**", "com.example.client/**"]
+```
+
+Every include pattern must stay within the plugin root and match at least one
+file.
+
+Use the build plan directly when another build system owns artifact writing:
+
+```python
+import agent_plugins as ap
+
+plan = ap.build_plan("packages/python")
+for file in plan.files:
+    print(file.source, file.target)
+```
+
+`agent_plugins.build.BuildBackend` provides the wheel, source distribution, and
+editable hooks used by the bundled adapters.
+
+## Choose build-time or runtime access
+
+Keeping `agent-plugins` in `[build-system].requires` makes it available during
+the build. Add the package to `[project].dependencies` when installed Python
+code needs to locate or inspect Agent Plugins:
+
+```toml
+[project]
+dependencies = ["agent-plugins==0.1.0"]
+```
+
+Locate a plugin with the Python package name used by pip:
+
+```python
+import agent_plugins as ap
+
+plugin = ap.locate("my-package")
+
+print(plugin.path)
+print(plugin.manifest.path)
+print(plugin.manifest.name)
+
+for skill in plugin.skills:
+    print(skill.path)
+    print(skill / "SKILL.md")
+
+if mcp := plugin.mcp:
+    for name, server in mcp.servers.items():
+        print(name, server)
+```
+
+`plugin.path` is the absolute installed plugin root. Each item in
+`plugin.skills` is an `ap.Skill` rooted at one immediate directory under
+`skills/`. Use `Path(skill)` or `skill.path` for that directory. Use `/` to
+build native paths to its instructions, references, scripts, or assets:
+
+```python
+skill = plugin.skills[0]
+
+print(skill / "SKILL.md")
+print(skill / "references" / "api.md")
+print(skill.tree(max_depth=2))
+```
+
+Use `skill.frontmatter` for the source text between the `---` delimiters. Use
+`skill.body` for the Markdown source after the frontmatter. The first access to
+either property reads and splits `SKILL.md`, then caches both strings. Path and
+tree access leave the document unread so an agent can choose which files and
+content to load.
+
+`plugin.manifest` is an `ap.Manifest`. `plugin.mcp` is an `ap.MCPConfig` when
+`mcp.json` exists. Each object exposes `.path` immediately. Accessing a parsed
+field such as `manifest.name` or `mcp.servers` reads, validates, and caches its
+document. MCP access validates the manifest first.
+
+MCP servers are frozen `ap.StdioServer`, `ap.StreamableHTTPServer`, or
+`ap.SSEServer` values in a read-only mapping. `manifest.issues` records ignored
+manifest fields. `mcp.issues` records invalid server entries skipped during
+loading. Document-level failures raise `ap.ValidationError` on parsed value
+access.
+
+Display the plugin to inspect its packaged tree:
+
+```python
+print(plugin)
+print(plugin.tree(max_depth=2))
+print(plugin.tree(max_depth=None, max_files=None))
+```
+
+`Path(plugin)`, `Path(skill)`, and `Path(plugin.manifest)` use the native path
+protocol. When `plugin.mcp` is present, `Path(plugin.mcp)` does too.
+`ap.installed()` returns each discovered plugin keyed by installed Python
+package name.
+
+## Verify the package
+
+Inspect the selected paths before building:
+
+```console
+agent-plugins plan path/to/python-project
+```
+
+Then verify the package through its installation boundaries:
+
+1. Build a wheel and source distribution.
+2. Build a wheel from the source distribution.
+3. Install the wheel in a clean environment.
+4. Install the Python package as editable.
+5. Call `ap.locate()` in both environments.
+6. Access `plugin.manifest.name` and `plugin.mcp.servers` when MCP exists to run
+   schema validation and the Agent Plugins rules.
+7. Confirm each `skill.path`, `skill / "SKILL.md"`, and `skill.files` points to
+   the packaged skill tree.
+8. Access `skill.frontmatter` and `skill.body` to verify the packaged
+   `SKILL.md` structure.
+
+Handle `ap.AgentPluginError` when a requested Python package or usable plugin
+root is absent. Handle `ap.ValidationError` when an installed plugin document
+or `SKILL.md` structure is invalid.

agent_plugins-0.1.0/.agent-plugin/skills/agent-plugins/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Agent Plugins"
+  short_description: "Ship agent components with Python packages"
+  default_prompt: "Use $agent-plugins to ship agent skills and MCP configuration with this Python package."

agent_plugins-0.1.0/PKG-INFO

@@ -0,0 +1,222 @@
+Metadata-Version: 2.3
+Name: agent-plugins
+Version: 0.1.0
+Summary: Ship Agent Plugins with Python packages and inspect their installed files
+Author: Péter Ferenc Gyarmati
+Author-email: Péter Ferenc Gyarmati <dev.petergy@gmail.com>
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Typing :: Typed
+Requires-Dist: tomli==2.3.1 ; python_full_version < '3.11'
+Requires-Python: >=3.10, <3.15
+Project-URL: Source, https://github.com/peter-gy/agent-plugins
+Description-Content-Type: text/markdown
+
+# agent-plugins
+
+`agent-plugins` packages Agent Skills, Model Context Protocol configuration, and
+client extension files with a Python distribution. Each installed wheel carries
+the agent files that match its code version.
+
+The package implements the portable [Agent Plugins](https://agent-plugins.org/)
+directory format and supports Python 3.10 through 3.14.
+
+## Ship an Agent Plugin
+
+Keep one Agent Plugin tree beside the code it documents:
+
+```text
+my-project/
+├── plugin.json
+├── skills/
+│   └── use-my-package/
+│       ├── SKILL.md
+│       ├── agents/
+│       ├── references/
+│       └── scripts/
+├── mcp.json
+└── packages/
+    └── python/
+        └── pyproject.toml
+```
+
+`skills/` contains [Agent Skills](https://agentskills.io/specification).
+`mcp.json` contains [Model Context Protocol](https://modelcontextprotocol.io/specification)
+server configuration when the plugin provides MCP servers.
+
+Create `plugin.json` at the plugin root:
+
+```json
+{
+  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
+  "name": "my-project"
+}
+```
+
+Configure the Python package to wrap `uv_build`:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "uv_build==0.12.2"]
+build-backend = "agent_plugins.build.uv_build"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+`root` is relative to `pyproject.toml`. The build selects `plugin.json`, the
+complete `skills/` tree, and `mcp.json` when present.
+
+Build, install, and locate the packaged plugin:
+
+```console
+uv build packages/python --out-dir dist
+python -m pip install "agent-plugins==0.1.0" dist/my_package-*.whl
+agent-plugins locate my-package
+```
+
+```text
+/.../site-packages/my_package-1.2.3.agent-plugin
+```
+
+The wheel contains the plugin directory and an `agent_plugins.json` marker in
+the distribution metadata. A source distribution stages the same files for a
+reproducible wheel rebuild. An editable install points the marker at the authored
+plugin tree.
+
+### Use Hatchling
+
+Keep the plugin settings and select the Hatchling adapter:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "hatchling==1.31.0"]
+build-backend = "agent_plugins.build.hatchling"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+### Include other plugin files
+
+Add root-relative patterns for executables or client extensions:
+
+```toml
+[tool.agent-plugins]
+root = "../.."
+include = ["bin/**", "com.example.client/**"]
+```
+
+Each pattern must stay within the plugin root and match at least one file.
+
+## Inspect an installed plugin
+
+Add `agent-plugins` to the runtime dependencies of Python code that calls the
+inspection API:
+
+```toml
+[project]
+dependencies = ["agent-plugins==0.1.0"]
+```
+
+```python
+import agent_plugins as ap
+
+plugin = ap.locate("my-package")
+
+print(plugin.manifest.name)
+print(plugin.manifest.path)
+
+for skill in plugin.skills:
+    print(skill.path)
+    print(skill / "SKILL.md")
+
+if mcp := plugin.mcp:
+    for name, server in mcp.servers.items():
+        print(name, server)
+```
+
+`ap.locate()` accepts the distribution name used by `pip`. `ap.installed()`
+returns every discovered Agent Plugin keyed by distribution name.
+
+| Object | Access |
+| --- | --- |
+| `plugin.path` | Absolute plugin root |
+| `plugin.manifest` | Lazy `plugin.json` document |
+| `plugin.skills` | `ap.Skill` objects rooted under `skills/` |
+| `plugin.mcp` | Lazy `mcp.json` document, when present |
+| `plugin.files` | Files selected by the package build |
+| `plugin.tree()` | Bounded ASCII tree of the installed plugin |
+| `skill.frontmatter` | Source text between the `---` delimiters |
+| `skill.body` | Markdown after the frontmatter |
+| `skill.files` | Files selected below the skill root |
+
+Manifest, MCP, and skill documents load on first parsed-field access and cache
+their result. Call `ap.locate()` again to read a fresh snapshot. Invalid
+documents raise `ap.ValidationError`.
+
+MCP servers are frozen `ap.StdioServer`, `ap.StreamableHTTPServer`, or
+`ap.SSEServer` values. Their fields preserve placeholders such as
+`${PLUGIN_ROOT}` for the agent client to resolve.
+
+## Inspect a build plan
+
+Preview the files selected by `[tool.agent-plugins]` before building:
+
+```console
+agent-plugins plan packages/python
+```
+
+Use `--json` for machine-readable output. Python build integrations can consume
+the same plan:
+
+```python
+import agent_plugins as ap
+
+plan = ap.build_plan("packages/python")
+for file in plan.files:
+    print(file.source, "->", file.target)
+```
+
+List every plugin visible in the current environment:
+
+```console
+agent-plugins list --json
+```
+
+## Develop and release
+
+From a repository checkout, install the locked development environment and run
+the local checks:
+
+```console
+uv sync --locked
+uv run ruff format --check src tests
+uv run ruff check src tests
+uv run ty check
+uv run pyrefly check
+uv run pytest -q
+./scripts/build-dist.sh
+```
+
+Prepare the version and lockfile in a release pull request:
+
+```console
+uv version --bump patch
+```
+
+After the release commit reaches `main` and its push CI passes, update the local
+branch and start the tag-driven release:
+
+```console
+git pull --ff-only origin main
+./scripts/release.sh --dry-run
+./scripts/release.sh
+```
+
+Before the first tag, configure PyPI to trust `.github/workflows/publish.yml`
+through the repository's `pypi` environment. PyPI documents the setup in
+[Adding a Trusted Publisher](https://docs.pypi.org/trusted-publishers/adding-a-publisher/).

agent_plugins-0.1.0/README.md

@@ -0,0 +1,205 @@
+# agent-plugins
+
+`agent-plugins` packages Agent Skills, Model Context Protocol configuration, and
+client extension files with a Python distribution. Each installed wheel carries
+the agent files that match its code version.
+
+The package implements the portable [Agent Plugins](https://agent-plugins.org/)
+directory format and supports Python 3.10 through 3.14.
+
+## Ship an Agent Plugin
+
+Keep one Agent Plugin tree beside the code it documents:
+
+```text
+my-project/
+├── plugin.json
+├── skills/
+│   └── use-my-package/
+│       ├── SKILL.md
+│       ├── agents/
+│       ├── references/
+│       └── scripts/
+├── mcp.json
+└── packages/
+    └── python/
+        └── pyproject.toml
+```
+
+`skills/` contains [Agent Skills](https://agentskills.io/specification).
+`mcp.json` contains [Model Context Protocol](https://modelcontextprotocol.io/specification)
+server configuration when the plugin provides MCP servers.
+
+Create `plugin.json` at the plugin root:
+
+```json
+{
+  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
+  "name": "my-project"
+}
+```
+
+Configure the Python package to wrap `uv_build`:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "uv_build==0.12.2"]
+build-backend = "agent_plugins.build.uv_build"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+`root` is relative to `pyproject.toml`. The build selects `plugin.json`, the
+complete `skills/` tree, and `mcp.json` when present.
+
+Build, install, and locate the packaged plugin:
+
+```console
+uv build packages/python --out-dir dist
+python -m pip install "agent-plugins==0.1.0" dist/my_package-*.whl
+agent-plugins locate my-package
+```
+
+```text
+/.../site-packages/my_package-1.2.3.agent-plugin
+```
+
+The wheel contains the plugin directory and an `agent_plugins.json` marker in
+the distribution metadata. A source distribution stages the same files for a
+reproducible wheel rebuild. An editable install points the marker at the authored
+plugin tree.
+
+### Use Hatchling
+
+Keep the plugin settings and select the Hatchling adapter:
+
+```toml
+[build-system]
+requires = ["agent-plugins==0.1.0", "hatchling==1.31.0"]
+build-backend = "agent_plugins.build.hatchling"
+
+[tool.agent-plugins]
+root = "../.."
+```
+
+### Include other plugin files
+
+Add root-relative patterns for executables or client extensions:
+
+```toml
+[tool.agent-plugins]
+root = "../.."
+include = ["bin/**", "com.example.client/**"]
+```
+
+Each pattern must stay within the plugin root and match at least one file.
+
+## Inspect an installed plugin
+
+Add `agent-plugins` to the runtime dependencies of Python code that calls the
+inspection API:
+
+```toml
+[project]
+dependencies = ["agent-plugins==0.1.0"]
+```
+
+```python
+import agent_plugins as ap
+
+plugin = ap.locate("my-package")
+
+print(plugin.manifest.name)
+print(plugin.manifest.path)
+
+for skill in plugin.skills:
+    print(skill.path)
+    print(skill / "SKILL.md")
+
+if mcp := plugin.mcp:
+    for name, server in mcp.servers.items():
+        print(name, server)
+```
+
+`ap.locate()` accepts the distribution name used by `pip`. `ap.installed()`
+returns every discovered Agent Plugin keyed by distribution name.
+
+| Object | Access |
+| --- | --- |
+| `plugin.path` | Absolute plugin root |
+| `plugin.manifest` | Lazy `plugin.json` document |
+| `plugin.skills` | `ap.Skill` objects rooted under `skills/` |
+| `plugin.mcp` | Lazy `mcp.json` document, when present |
+| `plugin.files` | Files selected by the package build |
+| `plugin.tree()` | Bounded ASCII tree of the installed plugin |
+| `skill.frontmatter` | Source text between the `---` delimiters |
+| `skill.body` | Markdown after the frontmatter |
+| `skill.files` | Files selected below the skill root |
+
+Manifest, MCP, and skill documents load on first parsed-field access and cache
+their result. Call `ap.locate()` again to read a fresh snapshot. Invalid
+documents raise `ap.ValidationError`.
+
+MCP servers are frozen `ap.StdioServer`, `ap.StreamableHTTPServer`, or
+`ap.SSEServer` values. Their fields preserve placeholders such as
+`${PLUGIN_ROOT}` for the agent client to resolve.
+
+## Inspect a build plan
+
+Preview the files selected by `[tool.agent-plugins]` before building:
+
+```console
+agent-plugins plan packages/python
+```
+
+Use `--json` for machine-readable output. Python build integrations can consume
+the same plan:
+
+```python
+import agent_plugins as ap
+
+plan = ap.build_plan("packages/python")
+for file in plan.files:
+    print(file.source, "->", file.target)
+```
+
+List every plugin visible in the current environment:
+
+```console
+agent-plugins list --json
+```
+
+## Develop and release
+
+From a repository checkout, install the locked development environment and run
+the local checks:
+
+```console
+uv sync --locked
+uv run ruff format --check src tests
+uv run ruff check src tests
+uv run ty check
+uv run pyrefly check
+uv run pytest -q
+./scripts/build-dist.sh
+```
+
+Prepare the version and lockfile in a release pull request:
+
+```console
+uv version --bump patch
+```
+
+After the release commit reaches `main` and its push CI passes, update the local
+branch and start the tag-driven release:
+
+```console
+git pull --ff-only origin main
+./scripts/release.sh --dry-run
+./scripts/release.sh
+```
+
+Before the first tag, configure PyPI to trust `.github/workflows/publish.yml`
+through the repository's `pypi` environment. PyPI documents the setup in
+[Adding a Trusted Publisher](https://docs.pypi.org/trusted-publishers/adding-a-publisher/).