A diagnostic and corrective skill focused on identifying and fixing AI outputs that appear correct but address the wrong task due to prompt ambiguity.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install output-wrong-task
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 output-wrong-task 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 Output Wrong Task?
The Output Wrong Task skill is a governance and quality assurance protocol designed for the Openclaw Skills ecosystem. It addresses a common failure mode in LLMs where the model produces well-formed, coherent output that fundamentally misses the user's intent. This typically occurs when a model latches onto a salient keyword or defaults to a high-probability interpretation of an ambiguous request, resulting in a convincing but incorrect deliverable.
By implementing this skill, developers can better monitor agent performance and ensure that the deliverables—whether code, documentation, or analysis—align strictly with the specific requirements of the prompt rather than a plausible-sounding alternative.
Output Wrong Task Use Cases
- Detecting when a model provides general documentation instead of the requested deep-dive debugging analysis.
- Identifying code generation that solves a simplified version of a complex architectural requirement.
- Catching instances where an agent answers only the first part of a multi-part prompt and ignores subsequent constraints.
- Verifying that automated outputs map back to the original goal before they are committed to a codebase.
How Output Wrong Task Works
- Monitoring: The skill monitors agent outputs for coherence versus intent alignment.
- Symptom Detection: It identifies 'latent ambiguity' by checking if the output matches a common interpretation that diverges from specific user constraints.
- Verification: It prompts the model to self-verify its understanding of the problem it just solved.
- Correction: It applies iterative refinement by restating concrete deliverables and anchoring the output format to eliminate room for substitution.
Output Wrong Task Setup
To integrate this protocol into your Openclaw Skills environment, add the skill definition to your local configuration:
# Navigate to your skills directory
cd ~/.openclaw/skills
# Create the skill file
touch output-wrong-task.md
Configure your agent's system prompt to include the mapping and verification steps outlined in the 'What to do' section of the skill documentation.
Output Wrong Task Data Schema & Taxonomy
| Feature | Description |
|---|---|
| Symptom Log | Tracks instances of mismatched intent for future prompt engineering. |
| Requirement Map | A Boolean check-list comparing the final output against the original multi-part requirements. |
| Constraint Anchor | Stores specific JSON schemas or function signatures that the output must adhere to. |
| Interpretation ID | Logs which 'incorrect' path the model took to help refine negative constraints. |
Output Wrong Task Advanced Features
- Automated multi-part task splitting to prevent silent dropping of requirements.
- Explicit interpretation ruling: automatically adds 'do not' clauses when recurring wrong-task patterns are detected.
- Structural anchoring: enforces rigid output formats (JSON, specific file paths) to reduce semantic drift.
- Cross-agent verification hooks to have a second model audit the intent alignment of the first.
SKILL.md
Loading
Related Openclaw Skills
Barcelona City Intelligence
A comprehensive data skill providing deep insights into Barcelona's tech ecosystem, economic drivers, and historical evolution.
Python Mutable Default Arguments Fix
A specialized tool for identifying and remediating Python functions that use mutable objects as default arguments to prevent silent state sharing.
Context Near Overflow
A guide for detecting and mitigating performance degradation caused by AI models reaching their context window limits during long sessions.
X/Twitter News Search
A high-performance search skill for X/Twitter that fetches raw tweet data and engagement metrics for AI agents.
Generate Product Hunt Launch Assets
An automated workflow to generate comprehensive copy, visuals, and metadata required for a successful Product Hunt launch.
Amsterdam Knowledge Skill
A comprehensive knowledge skill providing technical and historical insights into Amsterdam's evolution from a trading capital to a modern European tech hub.
Featured*
