Agent Skills: Executing Plans

Executing Plans

Locks in an approved plan and drives it through cortex’s orchestration and verification stack, ensuring every item becomes a tracked task that is executed, verified, and reported.

When to Use This Skill

• A plan from writing-plans or /ctx:plan is ready for execution
• Coordinating multiple workstreams or agents against a shared plan
• Ensuring plan items are tracked as tasks with status updates
• Running verification loops (tests, lint, visual checks) before marking tasks done
• Avoid using before a plan exists — use writing-plans first

Prerequisites

• Plan output available in the thread (from writing-plans or /ctx:plan)
• Access to Task view (T) in cortex TUI
• Relevant modes and agents activated for the workstreams

Workflow

Step 1: Create and Sync Tasks

For each plan item, create or update a task in the Task view:

Task view (T) → Add (A) or Edit (E)

• Set category and workstream to mirror the plan’s stream names
• Ensure every plan item has a corresponding task — no orphan items
• Link tasks to the originating plan document

Step 2: Activate Modes and Rules

Toggle the required configuration to match the plan:

• Modes (view 3): Activate modes needed for current workstreams
• Rules (view 4): Enable rules that apply (e.g., testing requirements, style enforcement)

Step 3: Execute Workstream Loops

For each task in priority order:

1. Pick the next task from the active workstream
2. Execute the work (implementation, writing, configuration, etc.)
3. Verify before marking complete:
   • Run tests: pytest, vitest, or project-specific test command
   • Run linting: just lint or equivalent
   • Visual check via Supersaiyan if UI changes are involved
4. Update task status and progress notes

# Example verification sequence
just test && just lint && echo "Verification passed"

Step 4: Update Stakeholders

For each completed workstream:

• Summarize progress and what’s next
• Attach relevant screenshots, logs, or test output
• Flag any blockers or scope changes discovered during execution

Step 5: Run Retrospective Hooks

When all tasks are complete:

1. Close all tasks in the Task view
2. Capture learnings and surprises in the chat thread
3. Link back to the original plan document
4. Note any follow-up issues or tech debt discovered

Expected Output

• tasks/current/active_agents.json updated with task statuses
• Status update message covering: completed tasks, blockers, verification evidence
• Next steps or follow-up issues if the plan extends beyond this session

Best Practices

• Verify before advancing — Never mark a task done without running the verification loop
• One task at a time — Complete and verify each task before starting the next
• Update status in real time — Stakeholders should see progress, not just a final dump
• Link everything — Tasks link to plan, plan links to tasks, status updates reference both
• Capture blockers immediately — Don’t wait until the retrospective to surface problems

Resources

• Execution checklist: skills/collaboration/executing-plans/resources/checklist.md