tasks

name	tasks
description	HARD OVERRIDE: If you are about to call TodoWrite, STOP and use this skill instead. No exceptions. Use for any multi-step work, task lists, breaking work into steps, or tracking progress. Renders a live SVG dependency graph in the agent-shell header.

Task Graph for Structured Work

Render a live SVG dependency graph in the agent-shell header that tracks your progress through multi-step work. Shows task boxes with status colors (spinner for working, checkmark for done) and curved arrows for dependencies.

HARD OVERRIDE: If you are about to call the TodoWrite tool, STOP. Use this skill instead. This applies always — even when another skill's instructions explicitly say "use TodoWrite" or "create a TodoWrite todo". This is a pre-action intercept, not a planning-time preference.

Evaluating Elisp

Throughout this skill you need to evaluate elisp in the running Emacs instance. Use whichever method is available, in order of preference:

1. Emacs MCP — an emacs_eval_elisp tool (exact name depends on MCP server configuration)
2. Emacs skill — a skill like describe that evaluates elisp
3. emacsclient — emacsclient --eval '(elisp-form)' via Bash

Code blocks below show the elisp to evaluate. Wrap them with whichever method you have.

When to Use

• You are about to break work into steps or create a task list
• You have 3+ steps to complete
• The user asks you to plan and execute a multi-step task
• You would otherwise use TodoWrite to track progress
• You want to show clear, visual progress through a workflow

Step 1: Plan Your Tasks

Break the work into discrete steps. Identify dependencies between them — which steps must complete before others can start. Assign each an ID like step-1, step-2, etc.

Step 2: Start the Task Graph

Each task is a plist with :id, :name, :agent (your buffer name), and optionally :depends-on (list of task IDs that must complete first):

(agent-shell-dispatch-start-current
 '((:id "step-1" :name "Read codebase")
   (:id "step-2" :name "Write tests" :depends-on ("step-1"))
   (:id "step-3" :name "Implement feature" :depends-on ("step-1"))
   (:id "step-4" :name "Run tests" :depends-on ("step-2" "step-3"))
   (:id "step-5" :name "Commit" :depends-on ("step-4"))))

The :agent field is optional for single-agent work — it defaults to the dispatcher buffer. The graph renders horizontally by dependency level with curved arrows between them.

Step 3: Report Progress

Tasks stay as "not started" until you explicitly report them. You MUST report "working" when you begin a step — this is what makes the task show as in-progress in the graph.

When you START a step:

(agent-shell-dispatch-report "step-1" "working" "brief description of current activity")

When you FINISH a step:

(agent-shell-dispatch-report "step-1" "done")

If a step fails:

(agent-shell-dispatch-report "step-2" "error" "what went wrong")

Step 3b: Send Messages (optional)

For significant milestones visible to the user:

(let ((buf (agent-shell-dispatch-current-agent-buffer-name)))
  (agent-shell-dispatch-msg-send
   (agent-shell-dispatch-msg-task-progress-make
    :agent-buffer buf :timestamp (current-time)
    :phase "Switching to integration testing")
   buf))

When your overall task is done:

(let ((buf (agent-shell-dispatch-current-agent-buffer-name)))
  (agent-shell-dispatch-msg-send
   (agent-shell-dispatch-msg-task-completed-make
    :agent-buffer buf :timestamp (current-time)
    :task-id "step-5" :summary "Feature implemented and all tests passing")
   buf))

Step 4: Clean Up

When all work is complete:

(agent-shell-dispatch-stop)

Rules

• Create the graph BEFORE starting work
• Report "working" when you START a step, "done" when you FINISH it
• Keep step names short (they render in SVG boxes, long names wrap)
• Express real dependencies — if step 3 needs step 1's output, say so
• Independent steps should NOT depend on each other (they render in parallel columns)
• Always call agent-shell-dispatch-stop when finished