# Agentic Engineering

This site distributes skill instructions and their accompanying command-line
tools without requiring Claude Code or a plugin marketplace client.

## Discover

Fetch https://tylerpayne.github.io/agentic-engineering/index.json for the complete machine-readable index.
Its `plugins` array lists names, descriptions, skill paths, binary paths,
bundle URLs, SHA-256 checksums, and every downloadable file with its URL.
Paths are relative to the named plugin's directory. An empty skills array
means the package is a CLI utility rather than a SKILL.md-based skill.

Index schema version 2 uses these custom fields (this site's convention):
- `agent_invoked`: boolean. The agent applies the skill or invokes its actions.
- `user_invoked`: boolean. The user directly operates its commands or actions.
- `invocation_notes`: explanation of the intended workflow.

The flags are independent: backlog is both; later is user-invoked only;
python and plain-english are agent-invoked only. A user requesting or activating
an agent skill does not count as invoking its actions. These flags describe the
intended workflow, not permissions or technical restrictions on who can run code.

For your own use as an agent, select entries with `agent_invoked: true`. Later
remains in the index for users; agents receive the submitted notes and do not
need to download or operate the plugin itself. All bundles remain available.

## Download a particular skill and its dependencies

1. Select an entry with `agent_invoked: true` whose `skills` array contains the SKILL.md you need.
2. Download its `archive` URL and verify the archive against `sha256` in the index.
3. Extract the tar.gz into a local tools directory. Each archive has a top-level
   directory named after the plugin. Keep its layout intact: bin/ tools import
   modules from the sibling lib/ directory. The archive includes its license.
4. Read the included README.md and the selected SKILL.md. Register the skill
   with your agent's skill mechanism, or load its instructions directly.
5. Run tools by their extracted absolute path, or add that plugin's bin/
   directory to PATH. Backlog requires Python 3 with SQLite support.
   `python3 /absolute/path/backlog/bin/backlog --help` describes its CLI.
   The python plugin requires explicit setup: run
   `uv sync --locked --project /absolute/path/python` after extraction or updates.
   Keep its pyproject.toml, uv.lock, .python-version, bin/, and lib/ together.
   Its launcher runs the prepared .venv Python directly without syncing.

Example (from a directory where you want to store tools):

    curl -fL https://tylerpayne.github.io/agentic-engineering/index.json -o index.json
    curl -fL https://tylerpayne.github.io/agentic-engineering/downloads/backlog.tar.gz -o backlog.tar.gz
    # Compare `shasum -a 256 backlog.tar.gz` to the index entry before extracting.
    tar -xzf backlog.tar.gz
    python3 backlog/bin/backlog --help

To fetch individual files instead, use the `files` URLs and preserve their
relative paths. Download the bin/ and lib/ files as well as the skill; a binary
alone is insufficient. Plain-english needs only its SKILL.md to be used as
instructions. Direct HTTP downloads do not preserve executable permissions;
invoke Python scripts with python3 or set their executable bit yourself.

## Adapting the instructions

These packages originate as Claude Code plugins. Claude slash commands, hook
registration, allowed-tools metadata, and automatic PATH setup are not portable.
Do not register Claude hooks in another agent. Invoke the CLI directly instead.
When a skill references ${CLAUDE_SKILL_DIR}/backlog, substitute the extracted
backlog/bin/backlog absolute path. Keep the skill inside its complete package
and register a reference to it if your agent supports that.

Run commands from the intended project directory. The tools store project data
under .claude/ even when used by another agent. Claude session discovery and
automatic prompt interception require Claude-specific integration; downloading
these files does not provide those features in another agent. Read README.md
for behavior and storage details before using commands that change project data.

The index and bundles describe one deployment, but these stable URLs update on
future deployments. If checksum verification fails, fetch the index and bundle
again together. The optional marketplace.json is for Claude Code clients only.

## Available packages

## backlog

Read and update a project's SQLite kanban board.
agent_invoked: true
user_invoked: true
Users capture notes through commands or the CLI; agents use the backlog-board skill and CLI to read, claim, and update work items.
Bundle: https://tylerpayne.github.io/agentic-engineering/downloads/backlog.tar.gz
Skills: skills/backlog-board/SKILL.md
Binaries: bin/backlog
- https://tylerpayne.github.io/agentic-engineering/plugins/backlog/README.md
- https://tylerpayne.github.io/agentic-engineering/plugins/backlog/skills/backlog-board/SKILL.md

## later

Claude Code commands for users to park notes and submit them later.
agent_invoked: false
user_invoked: true
User-facing Claude Code integration. The agent receives submitted notes; it does not use the plugin itself.

## plain-english

Writing instructions for plain, declarative English.
agent_invoked: true
user_invoked: false
The agent reads and applies the plain-english skill. No binaries are required.
Bundle: https://tylerpayne.github.io/agentic-engineering/downloads/plain-english.tar.gz
Skills: skills/plain-english/SKILL.md
Binaries: None.
- https://tylerpayne.github.io/agentic-engineering/plugins/plain-english/README.md
- https://tylerpayne.github.io/agentic-engineering/plugins/plain-english/skills/plain-english/SKILL.md

## python

Python conventions and a CLI for project bootstrap and verification.
agent_invoked: true
user_invoked: false
The agent applies the Python skill and uses python-style to bootstrap and verify projects. A user activating the skill does not make its actions user-invoked.
Bundle: https://tylerpayne.github.io/agentic-engineering/downloads/python.tar.gz
Skills: skills/python/SKILL.md
Binaries: bin/python-style
- https://tylerpayne.github.io/agentic-engineering/plugins/python/README.md
- https://tylerpayne.github.io/agentic-engineering/plugins/python/skills/python/SKILL.md