name: work-on-task description: Work on a specific task assigned to you in the agent swarm
Working on a Task
If no taskId is provided, you should call the poll-task tool to get a new task assigned to you.
Workflow
Once you get a task assigned, you need to immediately start working on it. To do so, the first thing you need to do is call the MCP tool get-task-details to get all the details about the task you need to work on.
Once you have the task details, you should:
- Figure out if you need to perform any research or planning before starting (see below)
- Use the
/skill:todosto add a new todo item indicating you are starting to work on the task (e.g. "Work on task XXX:"). This will help on restarts, as it will be easier to remember what you were doing. - Call
store-progresstool to mark the task as "in-progress" with a progress set to something like "Starting work on the task XXX, blah blah". Additionally use/skill:swarm-chatto notify the swarm, human and lead when applicable. Do not be too verbose, nor spammy. - Start working on the task, providing updates as needed by calling
store-progresstool, use theprogressfield to indicate what you are doing. - Once you either done or in a dead-end, see the "Completion" section below.
Research and Planning
As you start working on a task, consider whether you need to:
- Research: For research tasks, gather information from the web, codebase, or documentation before starting implementation.
-
Create a plan: For development tasks, create a detailed plan before implementing. Write it to
/workspace/personal/plans/or/workspace/shared/thoughts/shared/plans/. - Implement a plan: If you already have a plan, follow it step by step.
Communication
- Use
/skill:swarm-chatto communicate with other agents in the swarm if you need help or want to provide updates.
Decision guidelines
When the task is a research task, you should ALWAYS perform thorough research before proceeding.
When the task is a development task, you should ALWAYS create a plan first, then implement it.
If the task is straightforward with clear instructions, proceed normally without extensive planning.
Interruptions
If you get interrupted by the user, that is fine, it might happen. Just make sure to call store-progress tool to update the task progress once you get back to it. If the user provides new instructions, make sure to adapt your work on the task accordingly.
Once you get back to it, make sure to call /skill:work-on-task again with the same taskId to resume working on it.
Completion
Once you are done, or in a real dead-end, you should call store-progress tool to mark the task as "complete" or "failed" as needed. You should always use the output and failureReason fields to provide context about the task completion or failure.
If you used the /skill:todos to add a todo item when starting the task, make sure to mark it as completed or remove it as needed.
Once you are done (either ok or not), perform the Post-Task Reflection below, then finish the session by just replying "DONE".
Post-Task Reflection (REQUIRED)
After calling store-progress to complete or fail a task, do the following before finishing:
-
Transferable learning? If you learned something reusable (a pattern, a gotcha, a fix), write it to
/workspace/personal/memory/<descriptive-name>.md -
Swarm-relevant? If the learning applies to all agents (not just you), write it to
/workspace/shared/memory/<descriptive-name>.mdinstead - Identity update? If you discovered a new area of expertise or working style preference, update your IDENTITY.md
- Tools update? If you found a new service, API, or tool, update your TOOLS.md
Skip this section ONLY if the task was trivially simple (single file edit, no debugging, no new knowledge gained).
chat Comments (0)
Sign in to join the discussion and leave a comment.
Skill Details
Related Skills
Build your own?
Join 12,000+ developers contributing to the Claude ecosystem.
No comments yet. Be the first to share your thoughts!