xuiltul

machine-tool

Delegates work to external agent CLIs (machine tools) for large code changes, investigation, or analysis. Use when: offloading implementation via the machine command, heavy refactors, or batched agent runs.

xuiltul 254 45 Updated 5mo ago
GitHub

Install

npx skillscat add xuiltul/animaworks/machine-tool

Install via the SkillsCat registry.

About this skill

The machine-tool skill delegates tasks to external CLI agents for large code modifications, investigations, or analysis. It solves the problem of offloading heavy implementation work that would otherwise require manual effort. It is appropriate for extensive refactors, batch processing, or when running specialized command-line tools.

SKILL.md

Machine Tool

Delegate tasks to external agent CLIs.
Offload heavy work like code changes, investigation, and analysis to external agents.

Design Philosophy

You are the craftsperson. The machine is a machine tool (CNC, laser cutter, etc.).
A machine tool can cut with incredible precision, but it doesn't decide what to build.
It has no memory, no communication, no identity.
Your job is to provide precise blueprints (instructions).

CLI Usage

animaworks-tool machine run [options] "instruction" -d /path/to/workdir

CLI Options

Option Description
-e ENGINE Engine selection (omit for auto-selected default; use -h to list available engines)
-d PATH Working directory (default: current directory)
-t SECONDS Timeout in seconds (default: 600s sync, 1800s background)
-m MODEL Model override (default: engine's default)
--background Background execution (1800s timeout; output streams to state/cmd_output/)
-j / --json Output result as JSON

Basic Examples

# Minimal (default engine, current directory)
animaworks-tool machine run "detailed instruction"

# Specify engine and directory
animaworks-tool machine run -e cursor-agent "instruction" -d /path/to/workdir

# Background execution
animaworks-tool machine run --background "instruction" -d /path/to/workdir

# Custom timeout
animaworks-tool machine run -t 300 "instruction" -d /path/to/workdir

Writing Good Instructions (Important)

Vague instructions lead to poor results. Always include:

  1. Goal — What to accomplish
  2. Target files/modules — What to modify
  3. Constraints — Coding conventions, API compatibility, etc.
  4. Expected output — Code, report, diff, etc.

Pass Long Instructions via File

Instructions containing Bash special characters (|, `, $) will cause shell errors
if passed directly. Write to a file first:

# Write instruction to file
cat > /tmp/instruction.txt << 'INSTRUCTION'
## Task: PR #2087 Code Review

| Aspect | Check |
|--------|-------|
| Correctness | Meets issue requirements |
| Maintainability | Readability, tests, SRP |

Target file: `app/Services/Movacal/MovacalApiClient.php`
INSTRUCTION

# Read from file and execute
animaworks-tool machine run "$(cat /tmp/instruction.txt)" -d /path/to/workdir

Parallel Execution (--background)

Use --background to run multiple machines simultaneously.
Output streams in real-time to state/cmd_output/.

3-Parallel Review Example

# Launch 3 review perspectives in parallel
animaworks-tool machine run --background "Correctness review..." -d /path &
animaworks-tool machine run --background "Maintainability review..." -d /path &
animaworks-tool machine run --background "Consistency review..." -d /path &
wait

# Check results (Read files from state/cmd_output/)

When to Use

Scenario Suitable?
Multi-file code changes YES
Bug investigation / root cause analysis YES
Test code generation YES
Refactoring YES
Short questions NO (answer yourself)
Work requiring memory/messaging NO (do it yourself)

Notes

  • Machine tools have NO access to AnimaWorks infrastructure (no memory, messaging, or org info)
  • Rate limited (5 per session, 2 per heartbeat)
  • Background output streams to state/cmd_output/, check with Read/Glob