Percept Summarize

What This Skill Does

Percept Summarize, developed by jarvis563, acts as an automated cognitive layer for your OpenClaw agent. It functions as an intelligent post-conversation assistant that automatically processes audio transcripts once a 60-second silence period is detected. Rather than just transcribing, it performs high-fidelity entity extraction, relationship mapping, and action item identification. The skill utilizes a sophisticated 5-tier cascade system—combining exact, fuzzy, contextual, recency, and semantic search matching—to ensure that names, organizations, and project topics are correctly resolved and categorized. This data is structured into an internal SQLite database, allowing for powerful, complex queries regarding your past interactions.

Installation

To install this skill, ensure you have the percept-listen skill active to capture audio data. Run the following command in your OpenClaw CLI:

clawhub install openclaw/skills/skills/jarvis563/percept-summarize

Once installed, the agent will begin monitoring silence intervals to trigger the summarization pipeline automatically.

Use Cases

• Meeting Intelligence: Never miss a detail in a brainstorming session. Let the agent extract key decisions and action items while you focus on the discussion.
• Knowledge Retrieval: Quickly pull up the context of a project discussed weeks ago by querying the relationship graph.
• CRM Enhancement: Automatically map relationships between clients, internal teams, and mentioned projects.
• Project Management: Track follow-up tasks derived directly from verbal commitments made during voice calls.

Example Prompts

1. "What were the primary action items from the meeting with the marketing team yesterday?"
2. "Summarize my conversation with Sarah regarding the new deployment strategy."
3. "Who are the key stakeholders mentioned in my recent discussions about the AI project?"

Tips & Limitations

• Data Retention: Note that utterances are pruned after 30 days, while summaries persist for 90 days. Back up critical information if needed.
• Dependencies: This skill relies heavily on percept-listen. If the listening service is interrupted, summarization will not trigger.
• Optimization: For best results, use clear, distinct speech. The entity resolution's effectiveness depends on the quality of the transcription provided by the underlying listening skill.
• Database Access: You can perform complex manual audits via the local SQLite file or by accessing the Percept dashboard on port 8960.