task
/task
Create a Grid task folder or show a task's state. Use when the person starts a task ("start a task", "new task", "I want to add ..."), or asks where a task stands ("show task T-02", "what's the state of ...", "list my tasks"). Writes intent.md from their own words and task.md from the fields below.
- Author
- gridconsole ↗
- Repository
- grid-plugins ↗
- Licence
- MIT
- Last repository change
- 08.11.2026
Assessment and security
- Files in this package
- 3
- Scripts
- 2
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 08.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 2 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/alloc-lock.mjs, scripts/next-id.mjs
What it does and when to use it
Create a Grid task folder or show a task's state. Use when the person starts a task ("start a task", "new task", "I want to add ..."), or asks where a task stands ("show task T-02", "what's the state of ...", "list my tasks"). Writes intent.md from their own words and task.md from the fields below.
Triggers from the description
Original author description
Create a Grid task folder or show a task's state. Use when the person starts a task ("start a task", "new task", "I want to add ..."), or asks where a task stands ("show task T-02", "what's the state of ...", "list my tasks"). Writes intent.md from their own words and task.md from the fields below.
How it works
- Start a task
- What else lives in the folder
- Attachments
- Show a task
- Updating task.md later
SKILL.md preview
---
name: task
description: Create a Grid task folder or show a task's state. Use when the person starts a task ("start a task", "new task", "I want to add ..."), or asks where a task stands ("show task T-02", "what's the state of ...", "list my tasks"). Writes intent.md from their own words and task.md from the fields below.
---
# Task
A task is a folder: `<projects folder>/<project>/tasks/T-NN-slug/`. The projects folder is `GRID_PROJECTS`; read it with `printf %s "$GRID_PROJECTS"`. If it is empty, it is `$GRID_HOME/projects`, where `GRID_HOME` is the Grid data folder `~/.grid/<name>`, never the workspace folder (read it with `printf %s "$GRID_HOME"`). If both are empty, Grid is not set up for this folder: say so, write nothing and create no task, never one under a repo (Grid writes nothing into a project's repo folder). Never guess a default and never write outside that folder.
## Start a task
1. **Project.** The folder name in the projects folder. Use the one the person names; else the one matching the working folder's name; else the only one. If still unclear, ask.
2. **Id.** Choose the slug first: the title in 2 to 5 lowercase words joined by `-`, only `a-z`, `0-9` and `-`. Run `node "${CLAUDE_SKILL_DIR}/scripts/next-id.mjs" "<projects folder>" --reserve "<project>" "<slug>"` as a single command on its own (absolute path, no `cd`, `&&`, `$PWD` or chaining), so a person's allow rule for `node` matches it. It takes the allocation lock the app and terminal sessions share, makes the empty folder `<project>/tasks/T-NN-slug/` and prints `T-NN`; write the task's files into that folder. Ids are workspace-wide, so T-03 may sit in a different project than T-02.
3. **Owner.** The person who asked. Take the name from the conversation ("I'm Alex"), lowercase, first name, only `a-z`, `0-9` and `-`. If you do not know it, ask. It is set once; you never change it later.
4. **intent.md.** Create it with the Write tool, never a shell heredoc or redirect. The person's request word for word, no heading, no rewording, no fixes to spelling. If they said it over several messages, join them with blank lines.
5. **task.md.** Create it with the Write tool, never a shell heredoc or redirect. Frontmatter, then a body log. Get times with `date +%Y-%m-%dT%H:%M`. The block below is one example, for a code change; pick your own steps (see "Choosing the steps").
```
---
id: T-02
title: Search notes by tag
project: notes-app
owner: alex
source: idea
size: normal
workflow:
- {step: plan, kind: approve}
- {step: build, kind: run}
- {step: review, kind: approve}
step: plan
status: running
updated: 2026-03-04T10:32
summary: Let people filter the note list by tag.
next: Write the plan.
---
- 4 Mar · plan · task created
```
| Field | Rule |
|---|---|
| `id`, `title`, `project`, `owner` | as above; title is a short phrase in the person's words |
| `source` | `idea` (the person's own ask, the default), `signal`, `note` or `task` (made from another task) |
| `size` | `small` (one change in one place), `normal`, or `large` (needs a design or a preview). Your call; the plan can change it |
| `workflow` | the task's own steps, 2 to 6, one ` - {step: <name>, kind: <kind>}` line each. Names are `a-z`, `0-9` and `-`, each used once. There is no fixed phase list: see "Choosing the steps" below |
| `step` | the step now running; always one of the `workflow` names. Move it as the work moves |
| `round` | optional; a positive integer, written only when the work loops back to an earlier step: `2` the first time it goes back, then `3`, and so on |
| `status` | `running`, `waiting` (needs a person), `queued`, `stuck`, `reworking`, `pausing`, `paused` or `closed` |
| `updated` | the time of this write, the output of `date +%Y-%m-%dT%H:%M`, run just before this write, never a guess; set it on every change to task.md |
| `summary`, `next` | one line each: what the task is, and what happens next and who does it. Never leave either empty |
| `needs` | optional list of what waits on a person, such as `[Q-1]`; the questions skill keeps it |
| `outcome` | optional; one line, written by the deliver skill only: what was delivered, with the main path |
| `due` | optional; only when the person names a deadline: a local minute `YYYY-MM-DDTHH:MM`, no seconds and no zone |
| `repos` | optional; only when the task's work is in a git repo: the repos and branches its changes are on, one ` - {name: <repo>, branch: <branch>, base: <branch>}` line each. `name` is the repo's folder name (the folder project.md names, or a repo directly inside it), each used once; `branch` is the branch the changes are on and `base` the branch it started from. The execute skill writes it from the worker's Outcome; never write it at intake |
Never written: `branch`, `worktree` (the host makes one branch and worktree per task) and `phase` (no fixed phases) are gone, and a task.md with them fails the check; a task's branch goes under `repos`, never as a `branch:` field. `runner`, `gate_raised`, `paused`, `lane`, `lane_step` and `lane_round` are Grid's.
### Choosing the steps
Pick the steps from what the task delivers, not from the examples. A code change ends in a PR someone reviews; a site change ends in a deploy at a set time; a report ends in a PDF a person signs off; a newsletter ends in a send. Each needs different steps, and a small task may need only two. One line each, as a starting point:
- PR: `plan` approve, `build` run, `tests` check, `review` approve
- deploy: `fix` run, `staging` check, `sign-off` approve, `deploy` schedule
- PDF: `outline` approve, `draft` run, `sign-off` approve
- newsletter: `draft` make-together, `links` check, `approve` approve, `send` schedule
A `check` goes right before the step it guards: the tests before the review, staging before the deploy, the links before the send.
Each step has one kind, exactly one of these six:
| Kind | The step is |
|---|---|
| `check` | a quick test Claude runs before going on: an API key is there, there is disk space, staging answers |
| `ask` | a question to a person that the work waits on |
| `approve` | a person says yes to something: the idea, the plan, a customer's sign-off |
| `make-together` | a person steers in a canvas or a draft while Claude builds it |
| `run` | Claude works on its own; it may take minutes or days |
| `schedule` | something happens at a set time: a send, a deploy |
A second example: a fix to a live site, where staging failed once and the work went back to `fix`:
```
workflow:
- {step: fix, kind: run}
- {step: staging, kind: check}
- {step: sign-off, kind: approve}
- {step: deploy, kind: schedule}
step: fix
round: 2
```
6. **history.md.** Create it with the Write tool, never a shell heredoc or redirect, with the first line: `- <time> · created · by <owner>`. Every `<time>` in history.md is the output of `date +%Y-%m-%dT%H:%M`, run just before the write, never a guess. Append only. Write only these events: created, plan ready, question asked, question answered, question withdrawn, delivered (the deliver skill writes that one). Never write `plan accepted`, `preview approved`, `ship approved` or `phase set`; those are a person's approvals and Grid's.
7. **Concept note.** Run `grid-brain concept --task <id>` (the plugin's `bin/`): it writes the task's concept note into the project brain (`knowledge/concepts/task-<id>.md`, with its index and log line). Run it again whenever the task's step or status moves; it replaces the note. This is the brain's own script; Workers never run it.
8. Tell the person the folder path, the id and the workflow you chose, with each step's kind.
Do not plan, research or build in this skill. Starting a task only makes the folder.
## What else lives in the folder
Other skills add their own entries; none of them is a stray, so do not delete or tidy them: `research.md` (research), `plan.md` (plan), `spec.md` (preview, Large tasks only), `questions/<person>.md` (questions), `subagents/NN-name.md` (briefs), `artifacts/` (what the task delivers), `checks.md` and `checks/<run id>/` (the check runs and their proof: screenshots, recordings, diffs, logs; older tasks may hold proof under `evidence/`), `views/` (views, the derived pages), `ds/` (views: copies of the design stylesheets the pages link) and `memory/<step>.md` (memory: short dated lines, one file per workflow step). Starting a task makes only `intent.md`, `task.md` and `history.md`.
## Attachments
`attachments/` holds the files the app saved for a person (any kind); `inputs/` holds files the person dropped in. Every agent that works from the task reads both, and never modifies, moves, renames or deletes a file in either. How to read each kind:
- text, csv, json, md, images: the Read tool; a pdf with Read too, in page ranges when it is long.
- docx: `textutil -convert txt <file> -stdout` (macOS) or `pandoc <file> -t plain`.
- xlsx: a short script that prints every sheet, since Read cannot open it.
- zip: `unzip -l` first, then extract into a new empty temp folder outside the task folder and read from there.
- audio and video: cannot be read. Say so and raise a question (the questions skill).
Name every file that could not be read, with the reason, in your Outcome (the plan lists it as unread in its sources).
## Show a task
Read `task.md` of the named task (or list `<projects folder>/*/tasks/` with each task's id, title, step and status). Report: id, title, owner, size, step and where it sits in the workflow, status, `needs`, `next`, and how many plan steps are ticked and how many questions are open. Read only; change nothing.
## Updating task.md later
Use the Edit tool, never a shell redirect: edit the field, set `updated` (the output of `date +%Y-%m-%dT%H:%M`, run just before this edit, never a guess), and add a body line `- <D Mon> · <step> · <what happened>`. Never edit `intent.md`: it is the person's words. If they change their mind, add a dated paragraph to the end of it with their new words.
Files in this package (3)
- SKILL.md 10 KB
- scripts/alloc-lock.mjs 4 KB
- scripts/next-id.mjs 3 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · gridconsole/grid-plugins | 100 | MIT |
| SimHacker/moollm | 100 | MIT |
| mdiri-ai/SoduWork | 100 | MIT |
| codeaholicguy/ai-devkit | 100 | Apache-2.0 |
| urikanonov/ai-marketplace | 100 | MIT |
| adivmaimon/Caddis | 100 | MIT |
| editorgit/agentic-delivery | 100 | MIT |
| CipherNovaa/agent-skills | 100 | MIT |
| foxrun-ai/foxrun | 100 | MIT |
Database snapshot: 13.11.2026 · package version bdb372ac39e5 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90intake
Turn what a person typed, and any files they dropped into the task's inputs or attachments folder, into the task's intent.md and task.md. Use when a task starts from an email, a note, a screenshot or a pasted brief ("here is what the customer sent", "start a task from inputs/",… - 100new-project
Start a project with a living context. Asks what the project is about, then creates CLAUDE.md (what it is and how we work), STATE.md (where things stand) and DECISIONS.md (why) in the project folder. Use when the user starts a new project ("let's start a new project", "set up… - 95kanbaa-task
Create and move tasks on the Kanbaa to-do board in Herdr (Markdown files in the repo's to-do folder, with ids such as HER-12). Use when asked to add, create, file or log a task, to-do, ticket or follow-up for a repo; when work you are doing turns up something that should be… - 100project-start
Give a new project its own folder under ~/Projects with a CLAUDE.md, so every session on it shares pointers, decisions, memory and files. Use when the user says "new project", "/project-start", names a client, product or site that has no folder yet, or starts project work in a "No folder" session. - 100next
This skill should be used when the user asks "what's next", "what should I work on", "continue", "resume", "pick up where I left off", or names a task directly to jump to it. It lists which tasks are open in the current project and where each stands, or loads the one named, and… - 100work-folder
Sets up a git-ignored .work/ folder for each task's acceptance card, notes, scratch and outputs, and closes it out: move outputs home, archive what must be kept, delete the rest after a yes. Use when starting or finishing a multi-file task.