# Task > Unified task runner dispatcher: auto-detects mise, just, task.dev, or package.json scripts and provides commands to run, list, or detect the project task runner. Triggers on: "/task", "run task", "list tasks", "detect task runner", "what tasks", "mise run", "just run", "pnpm run", "package.json scripts". - Skill: `cloudvoyant/task` (Agent Skill, multi-file: 5 files) - Install (CLI): `npx skillmds@latest add cloudvoyant/task` - Raw SKILL.md: https://api.skillmd.com/api/skills/cloudvoyant/task/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Productivity - License: MIT - Author: cloudvoyant (https://skillmd.com/u/cloudvoyant) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/cloudvoyant/task --- # task Unified task runner skill dispatcher. Auto-detects the project's task runner and provides a consistent interface for listing, detecting, and running tasks. All detection, listing, and running is done by one shipped script (`scripts/task.sh`) invoked in a **single Bash call** — never by reading multiple markdown files. ## Critical Rules - **Do the work in one Bash call** — dispatch every verb to `scripts/task.sh`; never transcribe its logic into chat or re-read the workflow docs to run bash. - **Step 0 always runs first** — no exceptions. - **Unknown verb is a run query** — a first token that is not `detect`/`list`/`run` (or an alias) is treated as a task to run. - **Pass all remaining args through** — everything after the task name is forwarded to the runner unchanged. - **Before running raw build/test/lint/format commands** (`tsc`, `vitest`, `eslint`, `prettier`, etc.) in any other workflow, first call `/task list` to check whether a task wraps the command. If a task exists, use it. Never bypass the task runner. - **Markdown output: soft-wrap prose, never hard-wrap** — when writing markdown, write each paragraph as one continuous line; do not insert manual newlines to wrap prose at a fixed column width. Newlines still separate paragraphs, list items, headings, and code fences. ## Step 0: Parse Arguments The raw invocation args (filled by Claude Code / OpenCode slash commands): `$ARGUMENTS`. If this line is not filled in, read the verb and remaining args from the user's current message. ```bash VERB="[first non-flag token, or empty]" REMAINING_ARGS="[everything after VERB, preserving order and flags]" # Aliases case "$VERB" in r) VERB="run" ;; ls) VERB="list" ;; d) VERB="detect" ;; esac ``` `SKILL_DIR` is the directory containing this SKILL.md (resolve it from this file's own path so the script is found from any working directory). ## Step 1: Dispatch to the script (one Bash call) Run exactly one command based on `VERB`: | Invocation | Command to run | |---|---| | no verb / no args | `bash "$SKILL_DIR/scripts/task.sh" list` | | `list` (or `ls`) | `bash "$SKILL_DIR/scripts/task.sh" list` | | `detect` (or `d`) | `bash "$SKILL_DIR/scripts/task.sh" detect` | | `run [args…]` (or `r …`) | `bash "$SKILL_DIR/scripts/task.sh" run [args…]` | | any other first token ` [args…]` | `bash "$SKILL_DIR/scripts/task.sh" run [args…]` | Stream the output. The script prints a `RUNNER: ` header and the task listing for `list`; runner variables for `detect`; and for `run` it `exec`s the matched task, forwarding extra args. ## Step 2: Handle the script's markers The `run` subcommand may exit non-zero with a marker on the last line. Act on it: - **`NO_MATCH:`** — no task matched. Offer to create a task-runner-compatible entry for the detected runner, naming the exact file and showing the snippet, then create it only on user confirmation: - **mise** → add to `mise.toml`: ```toml [tasks.] description = "" run = "" ``` - **just** → add to `justfile`: ``` : ``` - **task** → add to `Taskfile.yml` under `tasks:`: ```yaml : desc: "" cmds: - ``` - **package.json** → add to the `scripts` object: `"": ""`. - **`AMBIGUOUS:`** — 2+ tasks matched the query. The candidate names are printed above the marker. Do **not** offer to create anything — present the candidates and ask which one to run (or have the user retype a more specific query). - **`ENUM_FAILED:`** — a runner is configured but its task list could not be enumerated (e.g. an untrusted mise config on a fresh clone, or a runner error). The runner's error is printed above the marker. Do **not** offer to create a task — surface the error and, for mise, suggest `mise trust`, then retry. Never treat this as "no such task." - **`NO_RUNNER`** — no runner configured. Offer to scaffold a minimal `mise.toml`: ```toml [tasks.] run = "" ``` - **`NO_QUERY`** — `run` was called with no task name; present the listing and ask which task to run. Never silently edit build config — offer the snippet and target file, and only write it after the user confirms. ## Workflow Index (reference docs — not executed on the hot path) - **detect** (`references/workflows/detect.md`) — describes the `task.sh detect` contract and variables. - **list** (`references/workflows/list.md`) — describes the `task.sh list` contract. - **run** (`references/workflows/run.md`) — describes the `task.sh run` contract and matching rules. ## Help ``` /task [args] Verbs: detect Detect which task runner this project uses list List all available tasks (default when no args given) run [args] Run a named task via the detected runner Any other first token is treated as a task to run: /task test → run the "test" task /task build -- --release Aliases: r → run ls → list d → detect Examples: /task # list all tasks /task list /task test # run the test task /task run build -- --release /task detect ``` ## Usage Note for Other Skills Other workflows should call `/task detect` at the start (or run `scripts/task.sh detect`) and use the resulting `RUNNER` / `LIST_CMD` variables instead of hard-coding shell commands.