Listen

What This Skill Does

The Listen skill is an intelligent adaptive layer for OpenClaw's speech-to-text (STT) capabilities. Rather than relying on a static model, Listen actively monitors your interactions to identify transcription failures, recurring misinterpretations, and context-specific jargon that standard models often struggle to parse correctly. By maintaining a local repository of corrections, patterns, and prohibited terms, the skill creates a self-optimizing loop that improves transcription accuracy every time you interact with your agent.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/ivangdavila/listen

Once installed, ensure your config.md file is configured with your preferred STT provider credentials, and refer to criteria.md for specific formatting requirements regarding how the agent stores its learned data. The system will automatically generate the internal logs needed to track your unique speech patterns.

Use Cases

• Technical Professional: Dictating complex code snippets or technical architecture notes where library names and specific syntax are often misidentified by generic STT models.
• Legal & Academic Transcription: Ensuring that proper nouns, surnames, or niche academic terminology are captured correctly after the first or second correction.
• Voice-First Productivity: Creating a frictionless experience where the user does not need to repeatedly clarify acronyms or internal project codenames.
• High-Noise Environments: Tuning the agent to better understand your specific vocal cadence, even when surrounding ambient noise might cause traditional models to stumble.

Example Prompts

1. "OpenClaw, you heard me say 'Pydantic' but you wrote 'Pie dantic'. Please remember that for next time."
2. "Listen skill, stop correcting 'OpenClaw' to 'Open Claw' in my documents; that is a false positive."
3. "Show me the current patterns you have learned regarding my team's project names."

Tips & Limitations

• Consistency: The model requires at least two occurrences of a pattern before it confirms a permanent correction. Avoid 'training' the model with one-off speech errors.
• Config Management: Keep your config.md updated if you switch STT providers, as different underlying engines may require distinct correction strategies.
• Human Oversight: Periodically review the correction logs to ensure the agent hasn't 'learned' a bad habit based on a specific, non-representative dictation session.
• Scope: This skill is focused on improving transcript accuracy; it does not perform semantic analysis or sentiment processing, keeping its footprint lightweight and fast.