takemo101/taskp

Dernière activité source enregistrée: 25 mars 2026 à 23:16

Catalogue SkillsMP mis à jour: 29 octobre 2026 à 18:30

skills collectés: 15

Étoiles GitHub: 1
Forks GitHub: 0

Installer avec un assistant IA

Copiez ce prompt dans l’assistant IA que vous utilisez.

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/takemo101/taskp.

README du dépôt

Fichier source du README : README.md

Lire le README source

Afficher le README complet

# taskp A CLI tool that runs skills (task procedures) defined in Markdown — collecting arguments via interactive prompts, then executing them with an LLM or template engine. [日本語版 README はこちら](README.ja.md) ## Features - **Skills defined in Markdown** — Human-readable, easy to write, and Git-friendly - **Two execution modes** — Template rendering (no LLM required) and AI agent execution - **Multi-action skills** — Group related operations (add/delete/list) into a single skill - **Multi-provider support** — Anthropic / OpenAI / Google / Ollama / Kimi / z.ai - **Skill hooks** — Run setup/teardown commands before and after skill execution - **Session ID & output forwarding** — Track executions and pipe results to hooks - **MCP server** — Usable from AI tools like Claude Code and pi - **MCP client** — Use external MCP server tools (GitHub, Slack, etc.) in agent mode ## Installation ```bash bun install -g github:takemo101/taskp ``` To install a specific version: ```bash bun install -g github:takemo101/taskp#v0.1.7 ``` > **Requirements:** Bun >= 1.2.0 If `taskp` is not found after installation, add the Bun global bin directory to your PATH: ```bash echo 'export PATH="$HOME/.bun/bin:$PATH"' >> ~/.zshrc source ~/.zshrc ``` ### Update ```bash bun install -g github:takemo101/taskp ``` ### Uninstall ```bash bun remove -g taskp ``` ## Quick Start ### 1. Initialize the project ```bash taskp setup ``` This creates the `.taskp/` directory with a commented-out `config.toml`. Uncomment the lines you need: ```toml [ai] default_provider = "anthropic" default_model = "claude-sonnet-4-20250514" [ai.providers.anthropic] api_key_env = "ANTHROPIC_API_KEY" ``` A JSON Schema and `.taplo.toml` are also generated, enabling autocompletion in editors that support [Taplo](https://taplo.tamasfe.dev/). For global configuration (shared across projects): ```bash taskp setup --global ``` ### 2. Create a skill ```bash taskp init deploy ``` This generates `.taskp/skills/deploy/SKILL.md`. ### 3. Edit the skill ```markdown --- name: deploy description: Deploy the application mode: template inputs: - name: environment type: select message: "Select deployment target" choices: [staging, production] - name: branch type: text message: "Branch name?" default: main --- # Deploy Deploying branch {{branch}} to the {{environment}} environment. ## Steps \`\`\`bash git checkout {{branch}} git pull origin {{branch}} npm run build npm run deploy:{{environment}} \`\`\` ``` ### 4. Run the skill ```bash taskp run deploy # → "Select deployment target" [staging / production] # → "Branch name?" [main] # → Commands are executed in order ``` ### 5. Run in AI agent mode Change to `mode: agent` and the LLM will interpret and execute the skill's content. ```bash taskp run code-review --model anthropic/claude-sonnet-4-20250514 ``` ### 6. Create a multi-action skill Group related operations into a single skill with `actions`: ```markdown --- name: task description: Manage tasks mode: template actions: add: description: Add a task inputs: - name: title type: text message: "Task title?" list: description: List tasks delete: description: Delete a task mode: agent tools: [bash] inputs: - name: id type: text message: "Task ID?" --- # Task Management ## action:add ｀｀｀bash echo "Adding task: {{title}}" ｀｀｀ ## action:list ｀｀｀bash echo "Listing tasks..." ｀｀｀ ## action:delete Find and delete task {{id}}, confirming with the user first. ``` ```bash taskp run task:add # Run a specific action taskp run task:add --set title="Buy milk" taskp tui # Select actions interactively ``` Each action can have its own `mode`, `model`, `inputs`, `tools`, `context`, and `hooks` — mixing template and agent mode within a single skill. ## Commands ### `taskp run <skill>` Run a skill. ```bash taskp run deploy taskp run deploy --model ollama/qwen2.5-coder:32b taskp run deploy --dry-run taskp run deploy --set environment=production --set branch=main taskp run deploy --no-input taskp run task:add # Run a specific action ``` | Option | Short | Description | |--------|-------|-------------| | `--model` | `-m` | LLM model (`provider/model` format supported) | | `--dry-run` | | Show execution plan without running | | `--force` | `-f` | Continue on error (template mode) | | `--verbose` | `-v` | Show detailed logs | | `--no-input` | | Disable interactive prompts (use defaults) | | `--set` | `-s` | Set variables directly (`--set key=value`) | ### `taskp setup` Initialize project configuration. ```bash taskp setup taskp setup --global taskp setup --force ``` | Option | Short | Description | |--------|-------|-------------| | `--global` | `-g` | Initialize global configuration (`~/.taskp/`) | | `--force` | `-f` | Overwrite existing files | ### `taskp list` List available skills. ```bash taskp list taskp list --global taskp list --local ``` - `taskp list` discovers `.taskp/skills` from the current working directory. When the working directory is under the home/global boundary, it also walks ancestor directories up to that boundary, then falls back to `~/.taskp/skills`. - `taskp list --local` shows only project scopes discovered from the current working directory. - `taskp list --global` shows only `~/.taskp/skills`. ### `taskp init <name>` Generate a skill scaffold. ```bash taskp init my-task taskp init my-task --global taskp init my-task --mode agent taskp init my-task --actions add,delete,list ``` ### `taskp show <skill>` Show skill details. ```bash taskp show deploy taskp show task:add # Show action details ``` ### `taskp tui` Launch the interactive TUI. ```bash taskp tui ``` Select skills with fzf-style fuzzy search, fill in parameters, and execute. Agent mode results are streamed as rendered Markdown. #### Key Bindings | Screen | Key | Action | |--------|-----|--------| | Skill selector | ↑↓ | Navigate | | Skill selector | Enter | Select | | Skill selector | Esc | Quit | | Input form | Tab / Shift+Tab | Next / previous input | | Input form | Enter | Confirm value | | Input form | Esc | Back | | Execution done | Enter | Return to skill selector | | Execution done | Esc | Quit |  ### `taskp serve` Start as an MCP server (see [Using as an MCP Server](#using-as-an-mcp-server) for details). ## Creating Skills Skills are defined as Markdown files named `SKILL.md`. ### File Structure ``` <skill-name>/ └── SKILL.md ``` ### Locations | Location | Scope | Use Case | |----------|-------|----------| | `<cwd>/.taskp/skills/<name>/SKILL.md` | Local | Tasks specific to the current working directory | | `<ancestor>/.taskp/skills/<name>/SKILL.md` | Parent | Tasks shared by a parent project or workspace | | `~/.taskp/skills/<name>/SKILL.md` | Global | Personal tasks shared across projects | Discovery prefers nearer scopes first: `local` → `parent` → `global`. Ancestor scopes are included only when the current working directory is inside the same home/global boundary. ### Frontmatter ```yaml --- name: my-skill # Skill name description: About this # Shown in list command mode: template # template | agent inputs: # Input definitions - name: target type: text message: "Target?" model: anthropic/claude-sonnet-4-20250514 # For agent mode (provider/model format) tools: # Tools for agent mode (optional) - bash - read - write - edit - grep - fetch context: # Sources auto-included in context (optional) - type: file path: "src/{{target}}" - type: command run: "git diff --cached" - type: image path: "docs/diagram.png" hooks: # Lifecycle hooks (optional) before: - "git stash --include-untracked" after: - "git stash pop || true" on_failure: - "echo 'Failed: $TASKP_ERROR'" actions: # Multi-action definitions (optional) build: description: Build the project inputs: - name: target type: text message: "Build target?" test: description: Run tests mode: agent tools: [bash, read] --- ``` ### Actions A skill can define multiple actions via the `actions` field. Each action can override `mode`, `model`, `inputs`, `tools`, `context`, and `timeout` — unspecified fields inherit from the skill level. If an action omits `inputs`, it inherits the skill-level `inputs`. Action bodies are defined in the Markdown body using `## action:<name>` sections: ```markdown ## action:build ｀｀｀bash npm run build --target={{target}} ｀｀｀ ## action:test Analyze test coverage and suggest improvements. ``` Run actions with colon syntax: ```bash taskp run my-skill:build taskp run my-skill:test --model anthropic/claude-sonnet-4-20250514 ``` ### Input Types | Type | UI | Return Type | |------|-----|------------| | `text` | Free text input | `string` | | `textarea` | Multi-line input (confirm with Meta+Enter) | `string` | | `select` | Choose from options | `string` | | `confirm` | Yes/No | `boolean` | | `number` | Numeric input | `number` | | `password` | Masked input | `string` | ### Variable Expansion Use `{{variable_name}}` in the body to expand input values. ### Agent Tools In agent mode, the following built-in tools are available: | Tool | Description | |------|-------------| | `bash` | Execute shell commands | | `read` | Read file contents | | `write` | Write to files | | `edit` | Replace a specific string in a file (exact single match required) | | `grep` | Search file contents for a pattern (regex supported) | | `fetch` | Fetch text content from a URL (http/https only) | | `glob` | Find files by glob pattern | | `ask_user` | Prompt the user for input during execution | | `taskp_run` | Invoke another template-mode skill | | `mcp:<server>` | All tools from an MCP server | | `mcp:<server>/<tool>` | A specific tool from an MCP server | Specify the tools you need in the `tools` field: ```yaml tools: - bash - read - grep - fetch ``` #### MCP Tools In agent mode, you can also use tools provided by external MCP servers. Define MCP server connections in `config.toml`, then reference them with the `mcp:` prefix: ```yaml tools: - bash - read - mcp:github # All tools from the "github" MCP server - mcp:slack/post_message # Only the "post_message" tool from "slack" ``` ```toml # .taskp/config.toml [mcp.servers.github] transport = "stdio" command = "npx" args = ["-y", "@modelcontextprotocol/server-github"] env = { GITHUB_TOKEN = "${GITHUB_TOKEN}" } ``` #### `taskp_run` The LLM can invoke other template-mode skills using the `taskp_run` tool: ```yaml --- name: diagnose mode: agent tools: - bash - read - taskp_run # Enable cross-skill invocation --- ``` Constraints: - Only template-mode skills can be called (no agent nesting) - Recursive calls are detected and blocked - Maximum nesting depth: 3 ### Image Context Skills can include images as context for multimodal LLM processing: ```yaml context: - type: image path: "docs/architecture.png" - type: image path: "screenshots/{{target}}.png" # Variable expansion supported ``` Supported formats: PNG, JPEG, GIF, WebP. The image is sent as binary data directly to the LLM. ## Skill Hooks Define `before`, `after`, and `on_failure` commands in the `hooks` field to run setup/teardown logic around skill execution. Each field accepts a single command string or an array. ```yaml hooks: before: "git stash --include-untracked" # single string OK after: - "git stash pop || true" # array also OK on_failure: - "curl -X POST https://slack.example.com/webhook -d '{\"text\": \"Failed: $TASKP_ERROR\"}'" ``` ### Execution Order ``` skill hooks.before → skill body → skill hooks.after → skill hooks.on_failure → global hooks ``` - **`before`** — Blocking. If any command fails, the skill body is skipped (but `after` still runs for cleanup). - **`after`** — Always runs after the skill body (success or failure). Failures are warnings only. - **`on_failure`** — Runs only when the skill body or `before` failed. Runs after `after`. Global hooks defined in `config.toml` run independently after skill hooks. See [Config Spec](docs/CONFIG-SPEC.md#hooks--ライフサイクルフック設定) for details. Actions can also define their own `hooks`, which completely override the skill-level hooks (no merging). Hooks are skipped when `--dry-run` is specified. ### Hook Environment Variables In addition to the [global hook environment variables](docs/CONFIG-SPEC.md#フックに渡される環境変数), skill hooks receive: | Variable | Description | |----------|-------------| | `TASKP_HOOK_PHASE` | Current phase (`before` / `after` / `on_failure`) | | `TASKP_OUTPUT_FILE` | Absolute path to the output file (see [Output Forwarding](#output-forwarding)) | All hooks (including global hooks) also receive these reserved variables, matching the template variables: | Variable | Template Equivalent | Description | |----------|-------------------|-------------| | `TASKP_SKILL_DIR` | `{{__skill_dir__}}` | Absolute path to the SKILL.md directory | | `TASKP_CWD` | `{{__cwd__}}` | Current working directory at skill start | | `TASKP_DATE` | `{{__date__}}` | Date at skill start (`YYYY-MM-DD`) | | `TASKP_TIMESTAMP` | `{{__timestamp__}}` | Timestamp at skill start (ISO 8601) | ## Session ID Each skill execution is assigned a unique session ID in the format `tskp_<random>` (e.g., `tskp_a1b2c3d4e5f6`). - Available as the reserved variable `{{__session_id__}}` in skill templates - Passed to hooks via the `TASKP_SESSION_ID` environment variable - Included in the agent mode system prompt - Nested `taskp_run` calls share the parent's session ID ```yaml --- name: deploy mode: template --- ``` ```markdown # Deploy (Session: {{__session_id__}}) ｀｀｀bash curl -X POST https://api.example.com/deploy \ --header "X-Session-Id: {{__session_id__}}" ｀｀｀ ``` ## Output Forwarding Skill execution output is written to a temporary file, accessible from hook commands via `$TASKP_OUTPUT_FILE`. - **Template mode**: stdout from all commands, concatenated with newlines - **Agent mode**: final LLM text output ```bash # after hook: save output to a log file cp "$TASKP_OUTPUT_FILE" "logs/${TASKP_SKILL_REF}_$(date +%Y%m%d).txt" # after hook: feed output into another skill taskp run summarize --set content="$(cat $TASKP_OUTPUT_FILE)" ``` The output file is created before execution and cleaned up (directory deleted) after all hooks complete. For full details, see [Skill Spec — Output Forwarding](docs/SKILL-SPEC.md#出力フォワーディング). ## Custom System Prompt You can customize the system prompt used in agent mode by placing a `SYSTEM.md` file. | Location | Scope | |----------|-------| | `.taskp/SYSTEM.md` | Project-local (takes priority) | | `~/.taskp/SYSTEM.md` | Global | When `SYSTEM.md` exists, its content replaces the default system prompt. The available tools list and environment information are automatically appended. When no `SYSTEM.md` is found (or the file is empty), the built-in default system prompt is used. For details, see [Skill Spec — Custom System Prompt](docs/SKILL-SPEC.md#カスタムシステムプロンプトsystemmd). ## Configuration Configuration is written in TOML format. Project settings take priority over global settings. ### Global config `~/.taskp/config.toml` ```toml [ai] default_provider = "anthropic" default_model = "claude-sonnet-4-20250514" [ai.providers.anthropic] api_key_env = "ANTHROPIC_API_KEY" [ai.providers.openai] api_key_env = "OPENAI_API_KEY" [ai.providers.ollama] base_url = "http://localhost:11434/v1" default_model = "qwen2.5-coder:32b" ``` ### Project config `.taskp/config.toml` ```toml [ai] default_provider = "ollama" default_model = "qwen2.5-coder:14b" ``` ## Using MCP Tools (Client) taskp can connect to external MCP servers as a client, allowing LLMs to use tools like GitHub, Slack, and other APIs during agent execution. See [MCP Client Spec](docs/MCP-SPEC.md) for details. ## Using as an MCP Server taskp can run as an MCP (Model Context Protocol) server, making it accessible from AI tools like Claude Code and pi. ```bash taskp serve ``` Exposed tools are generated from registered skills and actions. Examples: - `deploy` — Run the `deploy` skill - `review__fix` — Run the `fix` action of the `review` skill - `task__add` — Run the `add` action of the `task` skill ## Documentation See the `docs/` directory for detailed specifications. - [Concept](docs/CONCEPT.md) - [Architecture](docs/ARCHITECTURE.md) - [Skill Spec](docs/SKILL-SPEC.md) - [CLI Spec](docs/CLI-SPEC.md) - [AI Integration Spec](docs/AI-SPEC.md) - [Config Spec](docs/CONFIG-SPEC.md) - [MCP Client Spec](docs/MCP-SPEC.md) ## License MIT

Chargement du README du dépôt…