Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
agmo-inc avatar

Execute

  • 1 installs
  • 4 repo stars
  • Updated July 25, 2026
  • agmo-inc/agmo-everywhere

execute is a Claude Code skill that runs an implementation plan by dispatching executor agents with model routing and an auto-verification retry loop.

About

execute processes all TODOs from an approved plan by dispatching executor sub-agents, routing each TODO to a haiku, sonnet, or opus model by complexity. It runs a verification loop with up to three retries, escalates to a debugging agent after repeated failures, and does a final architect verification. A developer uses it to autonomously carry a written plan through to a verified build. Triggers include Korean phrases and 'execute'.

  • Runs an implementation plan by dispatching executor agents with category routing (haiku/sonnet/opus)
  • Auto-triggers verification with up to 3 retries then debugging on repeated failure
  • Reports per-TODO status (DONE, DONE_WITH_CONCERNS, BLOCKED, NEEDS_CONTEXT)

Execute by the numbers

  • 1 all-time installs (skills.sh)
  • Ranked #14,102 of 16,546 AI & Agent Building skills by installs in the Skillselion catalog
  • Data as of Jul 28, 2026 (Skillselion catalog sync)
At a glance

execute capabilities & compatibility

Capabilities
agent orchestration · plan execution · verification
Use cases
orchestration · planning
Pricing
Free
From the docs

What execute says it does

Process all TODOs from a plan by dispatching executor agents with category routing. Automatically links ralph for persistent completion.
SKILL.md
quick (haiku): simple file changes, config updates
SKILL.md
If 3 failures → auto-trigger agmo:debugging (include all failure contexts)
SKILL.md
npx skills add https://github.com/agmo-inc/agmo-everywhere --skill execute

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs1
repo stars4
Last updatedJuly 25, 2026
Repositoryagmo-inc/agmo-everywhere

What it does

Autonomously run an approved implementation plan by dispatching and verifying executor sub-agents per TODO.

Who is it for?

carrying an approved multi-TODO plan through implementation and verification autonomously

Skip if: writing the plan itself or one-off single-file edits without a plan

When should I use this skill?

the user wants to run an implementation plan or after plan approval

What you get

Every TODO is implemented, verified, and reported with a status until final architect verification passes.

  • implemented TODOs
  • per-TODO status reports
  • final architect verification

By the numbers

  • 3 model tiers (haiku/sonnet/opus)
  • up to 3 verification retries
  • 4 completion statuses

Files

SKILL.mdMarkdownGitHub ↗

Execute — Plan-Based Implementation with Auto-Ralph

Overview

Process all TODOs from a plan by dispatching executor agents with category routing. Automatically links ralph for persistent completion.

Activation

When this skill is invoked, ralph is automatically activated. The execution loop continues until all TODOs are complete and final verification passes.

Process

1. Load Plan

Read the plan from the vault path provided via skill args (e.g., --plan-path {VAULT_PATH}).

Path resolution: 1. If --plan-path is provided in args → Read the plan directly using Read tool from the vault path

  • The vault path format is an absolute path, e.g.: agmo-everywhere/plans/[Plan]-제목.md
  • Resolve to full path: {AGMO_VAULT_ROOT}/{vault_path} (e.g., {AGMO_VAULT_ROOT}/agmo-everywhere/plans/[Plan]-제목.md)

2. If no path provided → invoke agmo:vault-search to find the most recent plan for the current project 3. If Read fails (file not found) → invoke agmo:vault-search as fallback

Parse all TODOs into a work queue.

Classify each TODO:

  • Category: quick / standard / complex (determines model)
  • Dependencies: which TODOs must complete first
  • Parallelizable: can this run concurrently with others?

2. Execute TODOs

For each TODO (respecting dependency order):

1. Dispatch executor agent with appropriate category
   - quick (haiku): simple file changes, config updates
   - standard (sonnet): feature implementation, most work
   - complex (opus): complex logic, architectural changes

2. Executor completes the TODO and reports results

3. Auto-trigger agmo:verification
   - If FAIL → collect failure context (approach taken + verification judgment)
   - Retry with accumulated feedback (up to 3 attempts):
     Each retry prompt MUST include ALL prior failure context:

이전 시도 #{N}:

  • 접근법: {what was changed}
  • 실패 원인: {verification judgment summary}
  • 금지: 위 접근법 재시도
     (2nd retry includes attempt #1; 3rd retry includes attempts #1 + #2)
   - If 3 failures → auto-trigger agmo:debugging (include all failure contexts)
   - If PASS → proceed to step 4

4. Tag-based post-processing
   - Check the TODO's Tags field
   - If Tags contain `frontend` or `ui`:
     → Invoke agmo:accessibility with the TODO's changed files
     → If CRITICAL issues found → dispatch executor to fix → re-verify from step 3
     → If MINOR only → log summary, mark TODO complete
   - If Tags do not contain `frontend` or `ui`:
     → Mark TODO complete, proceed to next

Completion Status

When a TODO is marked complete (after step 3 PASS and step 4 post-processing), report one of the following statuses:

StatusWhen to use
DONEAll acceptance criteria met, no known issues
DONE_WITH_CONCERNSFunctionally complete but has known limitations or warnings
BLOCKEDCannot proceed due to missing dependency or external blocker
NEEDS_CONTEXTRequires clarification or additional context from the user

Output format per TODO:

Status: DONE | DONE_WITH_CONCERNS | BLOCKED | NEEDS_CONTEXT
Evidence: [Brief summary of how completion was verified]

For DONE_WITH_CONCERNS, also include:

Concerns:
  - [Concern description]
Follow-up: [Recommended next steps]

For BLOCKED, also include:

Reason: [Why it is blocked]
Blocker: [What needs to be resolved]
Next-Steps: [What must happen to unblock]

For NEEDS_CONTEXT, also include:

Question: [What clarification is needed]
Options: [Possible approaches, if applicable]

Behavior by status:

  • DONE: Proceed to the next TODO immediately.
  • DONE_WITH_CONCERNS: Log the concerns, report to the user, then proceed to the next TODO. Do not halt execution.
  • BLOCKED: Report the blocker to the user and wait. Do not auto-proceed to the next TODO.
  • NEEDS_CONTEXT: Ask the user the clarifying question and wait for their response. Do not auto-proceed.

Connection to verification:

  • Verification PASS + no concerns → DONE
  • Verification PASS + known limitations or warnings → DONE_WITH_CONCERNS
  • Verification FAIL after 3 retries → debugging invoked; status resolved after debugging resolves
  • Dependency missing or external blocker encountered → BLOCKED
  • Acceptance criteria ambiguous before or during execution → NEEDS_CONTEXT

3. Parallel Execution

If multiple TODOs are independent (no dependency, no shared files, and no import/export dependencies between their target files):

  • Invoke agmo:parallel to dispatch multiple executors concurrently
  • Each executor gets exclusive file ownership
  • Integration phase after parallel batch completes

4. Final Verification

After all TODOs are complete:

  • Dispatch architect agent for final verification
  • Verify: build passes, tests pass, all acceptance criteria met
  • If any failure → address it and re-verify

5. Completion

When architect approves final verification: 1. Invoke agmo:save-impl to save implementation summary to Obsidian 2. If significant learnings occurred, suggest agmo:wisdom recording 3. Report completion to user

TDD Integration

If the plan specifies TDD:

  • Before each TODO, invoke agmo:tdd to enforce RED → GREEN → REFACTOR
  • Executor writes failing test first, then implementation

Error Handling

SituationAction
TODO acceptance criteria unclearAsk user for clarification
3 consecutive verification failuresInvoke agmo:debugging (3-fix limit)
Dependency deadlockReport to user, suggest plan revision
Build broken mid-executionStop, fix build, then resume
Tags missing from TODOLog warning, attempt fallback by file extension (.tsx/.jsx/.css/.html → treat as frontend). If ambiguous, skip post-processing

Related skills

FAQ

How are TODOs routed to models?

Quick TODOs go to haiku, standard feature work to sonnet, and complex logic or architectural changes to opus.

What happens when verification keeps failing?

It retries with accumulated failure context up to 3 times, then auto-triggers a debugging agent.

AI & Agent Buildingagentsautomation

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.