Wreckit Skill (jmanhype Fork)
First time? Read SETUP.md first to install the Wreckit CLI from the jmanhype fork.
Connects Clawdbot to the jmanhype fork of Wreckit, the Autonomic Software Factory. This version is optimized for Cattle Architecture (Ephemeral Sprite VMs) and high-concurrency loops.
Script: skills/wreckit/scripts/run-wreckit.mjs
Engine: wreckit (Global CLI from jmanhype/wreckit)
-----|--------| | "Wreckit status" | List all items and their current states | | "Wreckit run [ID]" | Start the autonomous loop (Default: Cattle Mode) | | "Wreckit run [ID] mode pet" | Run locally without Sprites (Faster, less secure) | | "Wreckit dream" | Autonomous ideation (Self-Roadmapping) | | "Wreckit doctor" | Self-healing diagnostics and repair | | "Wreckit rollback [ID]" | Rollback a direct-merge item to its pre-merge state | | "Wreckit next" | Process the next incomplete item in sequence | | "Wreckit learn" | Extract patterns from completed items and compile skills | | "Wreckit summarize" | Generate visualization videos for completed items | | "Wreckit geneticist" | Analyze failure patterns and optimize system prompts |
The "Cattle" Advantage (Default)
Unlike standard agent loops, this version of Wreckit uses Ephemeral Sprite VMs. Every task spins up a fresh Firecracker microVM, executes code in isolation, and vanishes. No state contamination, maximum security.
The "Pet" Mode (Local)
If you need raw speed or are debugging locally, say "mode pet". This runs the RLM agent directly on the host machine.
Usage Guide
1. Checking Status
User: "Wreckit status" Bot: Lists all active items (Idea, Planned, In Progress, Done).
2. Running a Task
User: "Wreckit run 096" Bot: "Starting Wreckit run for item 096... [Streams logs] ... Done."
3. Dreaming (Autonomy)
User: "Wreckit dream" Bot: Scans your codebase for technical debt and creates new items automatically.
4. Rollback
User: "Wreckit rollback 096" Bot: Reverts item 096 to its pre-merge state using git operations.
With force: "Wreckit rollback 096 force" - Forces rollback even if item is not in 'done' state.
5. Next Item
User: "Wreckit next" Bot: Processes the single next runnable item with dependency satisfaction.
6. Learn (Pattern Extraction)
User: "Wreckit learn all" Bot: Extracts patterns from all completed items and compiles them into skills.
Specific item: "Wreckit learn item 096" By phase: "Wreckit learn phase done"
7. Summarize (Video Generation)
User: "Wreckit summarize all" Bot: Generates 30-second feature visualization videos for all completed items.
Specific item: "Wreckit summarize item 096" By phase: "Wreckit summarize phase done"
8. Geneticist (Prompt Optimization)
User: "Wreckit geneticist" Bot: Analyzes healing logs from the last 48 hours to identify and fix recurrent error patterns.
Custom window: "Wreckit geneticist time-window 24 min-errors 5"
Advanced Flags
Global Flags (apply to all commands)
--cwd <path>- Override the working directory--parallel <n>- Process N items in parallel (default: 1)--verbose- Enable verbose output--dry-run- Show what would be done without making changes
Mode Selection
--mode cattle(default) - Use ephemeral Sprite VMs--mode pet- Run locally without Sprites
Command-Specific Flags
Rollback
--force- Force rollback even if item is not in 'done' state
Learn
--item <id>- Extract patterns from specific item--phase <state>- Extract patterns from items in specific phase state--all- Extract patterns from all completed items--output <path>- Output path for skills.json (default: .wreckit/skills.json)--merge <strategy>- Merge strategy: append|replace|ask (default: append)--review- Review extracted skills before saving
Summarize
--item <id>- Generate video for specific item--phase <state>- Generate videos for items in specific state--all- Generate videos for all completed items
Geneticist
--auto-merge- Automatically submit PRs for optimized prompts--time-window <hours>- Analyze healing logs from last N hours (default: 48)--min-errors <count>- Threshold for recurrent pattern detection (default: 3)
Configuration
This skill assumes wreckit is installed globally or available in the path.
If running in Docker, ensure the container has access to the project directory.
Parallel Processing
When using --parallel flag with commands that process multiple items (like learn --all or summarize --all), the wrapper will process N items concurrently. For bot safety, default is 1 (sequential processing).
Working Directory Override
Use --cwd to specify a different project directory than the current working directory. This is useful for managing multiple projects or operating in a containerized environment.