execute

jero2rome's avatarfrom jero2rome

Use when the user wants to execute tasks autonomously, run compounder on a tasks.md file, start autonomous development, compound through a task list, or implement tasks iteratively. Guides autonomous task execution using compound loops.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 7, 2026
[Developer Tools]

When & Why to Use This Skill

The Execute skill is a powerful orchestration framework designed for autonomous task management and iterative software development. It enables Claude to process complex task lists (such as tasks.md) through 'compound loops,' ensuring each step is implemented, verified via testing, and documented before moving to the next. This skill effectively bridges the gap between high-level planning and technical execution by maintaining a continuous feedback loop until the project goals are fully realized.

Use Cases

  • Autonomous Feature Development: Automatically implement a series of technical requirements by iterating through a tasks.md file, marking progress, and handling debugging cycles independently.
  • Test-Driven Development (TDD) Automation: Run a compound loop that writes code and executes test suites iteratively, ensuring that the 'completion promise' is only met when all tests pass.
  • Complex Refactoring Workflows: Manage large-scale code changes by breaking them into smaller, verifiable steps that the agent executes and validates sequentially to prevent regressions.
  • Project State Synchronization: Use the skill to detect project phases (Specify, Plan, Execute) and automatically transition from planning documents to active implementation with integrated monitoring.
nameexecute
descriptionUse when the user wants to execute tasks autonomously, run compounder on a tasks.md file, start autonomous development, compound through a task list, or implement tasks iteratively. Guides autonomous task execution using compound loops.

Execute Skill

Orchestrates autonomous task execution using compounder's iterative loop mechanism.

Quick Start

When user has tasks ready for autonomous execution:

/compounder:compound-loop "Execute tasks from tasks.md. Mark completed with [X]. Run tests after each." --max-iterations 20 --completion-promise "ALL_TASKS_COMPLETE"

Phase Detection

Before starting, detect project state:

  1. Check for tasks.md: Look for tasks.md, .specify/specs/*/tasks.md, or similar

    • If found with - [ ] uncompleted tasks: Ready for execution
    • Count tasks to estimate iterations (2-3 per task)

  2. Check for plan.md/spec.md: If no tasks.md exists

    • Guide user to create tasks first (manually or via speckit)

Launching Compounder

Calculate Parameters

iterations = (uncompleted_tasks * 2) + 5  # buffer for debugging
completion_promise = "ALL_TASKS_COMPLETE" or custom from user

Prompt Template

Build a prompt that:

  1. References the tasks file explicitly
  2. Instructs to follow task order (phases if present)
  3. Mark tasks [X] as completed
  4. Run verification after each task
  5. Output <promise>COMPLETION</promise> only when truly done

Example Invocations

For a tasks.md with 10 tasks:

/compounder:compound-loop "Execute all tasks in tasks.md following TDD. After each task: mark [X], run tests. When ALL tasks complete and tests pass, output <promise>ALL_TASKS_COMPLETE</promise>" --max-iterations 25 --completion-promise "ALL_TASKS_COMPLETE"

For a single complex task:

/compounder:run-task "Implement user authentication" --iterations 10 --done-when "AUTH_WORKING"

Speckit Integration

If speckit commands are available in the project:

Phase Command Output
Specify /speckit.specify spec.md
Plan /speckit.plan plan.md
Tasks /speckit.tasks tasks.md
Execute /compounder:compound-loop Implementation

Completion Signals

The compound loop ends when:

  • <promise>COMPLETION_PROMISE</promise> is output (and TRUE)
  • Max iterations reached
  • /compounder:cancel-compound is run

Monitoring Progress

# Check current iteration
head -10 .claude/compounder-*.local.md

# View task completion
grep -E "^\- \[.\]" tasks.md