An execution-time skill that enforces atomic, single-action tool calls for AI agents to prevent silent stalls and parser crashes during complex workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install decomtangle
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install decomtangle using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is DecomTangle?
DecomTangle is an execution-time discipline skill designed for OpenClaw-style agents to govern the structural shape and boundaries of tool calls. Multi-step workflows often tempt local and open-weight models to output an entire procedure—such as a complex bash script, loop, or long API sequence—within a single tool call's arguments. This mega-tool-call anti-pattern frequently breaks strict JSON tool-call parsers, resulting in HTTP 500 errors or silent agent stalls.
Rather than serving as a high-level goal or TODO planner, this skill focuses strictly on runtime execution. It enforces an intentional step-observe-step loop, guaranteeing that every single tool call represents exactly one observable action. By breaking procedures down so that N steps equal N calls, DecomTangle ensures that open-source LLMs can successfully parse arguments and dynamically adjust their next action based on immediate system feedback.
DecomTangle Use Cases
- Automating multi-step browser interactions where subsequent actions depend entirely on the state of the previous page load.
- Running sequential API request chains where downstream operations require parsing data returned from upstream mutations.
- Performing complex database migrations or system operations through shell tools without risking opaque parsing timeouts.
- Enhancing execution reliability for local or open-weight models (like Qwen or specialized OSS models) prone to parser-fragile JSON emissions.
How DecomTangle Works
- Analyze Objective Boundary: The agent identifies a multi-step or stateful procedure and maps out the distinct physical actions required before executing.
- Isolate and Call Singular Tool: The agent executes exactly one atomic action (e.g., a standalone file write or single navigation) containing the bare minimum arguments required.
- Capture and Observe State: The system halts to process the tool output, allowing the agent to evaluate the concrete result, status code, or live context.
- Determine Next Step dynamically: Based on the observed outcome, the agent decides on the exact shape of the subsequent action rather than following a blind, pre-scripted path.
- Confirm and Verify Side Effects: Side effects are treated only as attempted until a final fresh observation confirms the new persistent state in the external system.
DecomTangle Setup
To deploy this execution discipline within your agent runtime environment, include the package components in your agent's system prompt or tool-governance framework. This skill operates as an abstract behavioral constraint and requires no specific binary installations or credential sets.
Verify your repository structure includes the core references and examples:
# Check for required Openclaw Skills files
ls -l references/decomposition-heuristics.md
ls -l references/atomic-call-checklist.md
ls -l examples/good-multicalendar-atomic.md
Ensure your agent system architecture links these heuristics to the tool execution loop, applying the six-point pre-flight checklist immediately prior to dispatching any tool payloads.
DecomTangle Data Schema & Taxonomy
DecomTangle organizes its behavioral assets and reference schemas into structured markdown files that define and enforce tool-calling boundaries:
| File path | Purpose | Core Content Type |
|---|---|---|
references/decomposition-heuristics.md |
Boundary Guideline | Contains the observable-action test, decision-point test, and an anti-pattern catalog. |
references/atomic-call-checklist.md |
Validation Schema | A six-point pre-flight checklist evaluated prior to executing any outward tool call. |
examples/good-multicalendar-atomic.md |
Reference Workflow | A step-by-step trace of a proper navigate-snapshot-select-poll-act workflow execution. |
examples/bad-mega-script-stall.md |
Incident Post-Mortem | Annotated failure trace of the verbatim mega-script pattern causing parser crashes. |
DecomTangle Advanced Features
- The Complexity Tripwire: Provides an explicit mental heuristic that flags multi-line scripting, nested quotes, or complex heredocs inside JSON parameters as a signal to decompose the step further.
- Minimized Argument Footprint: Promotes a strict data strategy where optional arguments are explicitly omitted to prevent type slips (like quoting booleans) that trigger parser rejections.
- Pacing Isolation: Standardizes wait states by turning pauses into standalone, atomic actions (e.g., isolated sleep tasks) rather than embedding sleep-and-act loops into standard functional arguments.
- Local Failure Isolation: Restricts errors to a single localized action step, enabling the agent to fix and retry that specific step without invalidating or restarting the entire transaction chain.
SKILL.md
Loading
Related Openclaw Skills
OppHub OPC Skill
An automation and integration client for OppHub OPC matchmaking, designed with zero interactive prompts for autonomous AI agents.
Douyin Creator Center Video Upload
An automation tool that uploads local video files to the Douyin Creator Center and manages the publication workflow via a Chrome DevTools Protocol connection.
Live Highlight Slicer
An automated utility designed to detect high-energy moments in stream recordings and slice them into highlight videos.
File Hash Verifier
Verify files with SHA-256 checksums to detect corruption, tampering, and release drift in seconds.
Novel Bug Checker
An automated diagnostic utility designed to audit narrative text for logical loopholes, character inconsistency, and pacing issues.
novel-master
An advanced engineering framework for generating publication-grade long-form Chinese novels using distributed character state machines and compressed narrative graphs.
Featured*
