A protocol for AI agents to enforce explicit state tracking, evidence-based completion, and the elimination of mid-task stalls.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install task-finish-contract
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 task-finish-contract 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 Task Finish Contract?
The Task Finish Contract is a specialized framework designed to solve the common problem of AI agents stalling mid-task or providing plans without follow-through execution. By mandating a strict format for tracking goals, progress, and immediate next steps, this skill ensures that every agentic interaction moves toward a verifiable conclusion. It is an essential component for developers using Openclaw Skills who require high reliability and transparency in autonomous workflows.
This skill transforms how agents communicate by requiring concrete evidence for every completed action. Instead of vague promises, the agent provides a structured output containing checklists and artifact paths, ensuring that 'finished' truly means finished. It acts as a contractual obligation between the user and the agent to maintain momentum and provide proof of work at every stage of the development lifecycle.
Task Finish Contract Use Cases
- Managing complex, multi-step technical tasks that require persistent state across multiple turns.
- Resuming interrupted agent sessions without losing context or repeating previous work.
- Ensuring high-fidelity handoffs between different specialized agents within Openclaw Skills.
- Auditing agent performance through structured evidence and artifact verification.
How Task Finish Contract Works
- The agent initializes a task by explicitly defining the Goal, current Progress, and the single Next concrete action.
- For every substantial step, the agent updates its internal memory or logs with the latest state transition.
- Upon task completion, the agent generates a DONE_CHECKLIST and provides a link or path to the EVIDENCE artifact.
- An Anti-Stall Rule is applied, preventing the agent from sending more than one planning-only response without execution evidence.
- The final output is verified against executable criteria to ensure no unresolved markers like TODO or PENDING remain in the workspace.
Task Finish Contract Setup
To integrate this protocol into your environment, add the state tracking requirements to your agent's core instructions. You can verify the integrity of your task logs using standard CLI tools commonly used with Openclaw Skills:
# Check for required state markers in the daily log
grep -E "Goal|Progress|Next" memory/$(date +%Y-%m-%d).md
# Verify that no TBD markers remain in your project artifacts
grep -r "TBD" ./src
Task Finish Contract Data Schema & Taxonomy
The skill organizes data using a structured Markdown schema to ensure human readability and machine parsability. All task logs are stored using relative paths within the workspace.
| Component | Purpose | Requirement |
|---|---|---|
| Goal | Defines the successful end-state | Mandatory |
| Progress | Summarizes actions already taken | Mandatory |
| Next | Specifies the immediate autonomous action | Mandatory |
| Evidence | Path, URL, or ID of the work produced | Required for 2+ steps |
| DONE_CHECKLIST | Granular breakdown of sub-task status | Required for 2+ steps |
Task Finish Contract Advanced Features
- Anti-Stall Enforcement: Restricts agents to a single planning-only reply to ensure continuous execution.
- Autonomous Step Chaining: Defines follow-up actions that run without user input to maintain workflow velocity.
- Evidence-Based Verification: Uses regex-based checks to programmatically confirm that all goals have been met.
- Privacy-First Logging: Ensures that evidence logs contain only relative paths and no sensitive credentials, adhering to Openclaw Skills safety standards.
SKILL.md
Loading
Related Openclaw Skills
Skill Governance
A sophisticated governance framework designed to manage the lifecycle, auditing, and routing of multiple AI agent capabilities.
Task Finisher
A protocol-driven skill that enforces completion standards and evidence-based verification for AI agent tasks.
Pre-Publish Privacy Scrub
A specialized security utility designed to scan, detect, and redact sensitive information from skill files before they are shared or published.
PowerShell Safe Chain
A standardized approach for chaining Windows PowerShell commands reliably using try/catch logic and error handling instead of brittle bash-style operators.
UI Design Optimizer
An automated design system generator that creates production-ready UI specs and starter code from local datasets.
Memory System Skill
A robust long-term memory management system for OpenClaw agents utilizing structured PostgreSQL and vector-based semantic search.
Featured*
