The Agent Skills Directory

[COMMAND_EXECUTION]: Spawns python3 and duckdb subprocesses to execute database operations for logging call data and retrieving history (webhook.ts).- [COMMAND_EXECUTION]: Instructs the user to register a local service using the zo register-service CLI tool to maintain the webhook server as a background process (SKILL.md).- [DATA_EXFILTRATION]: Accesses Google OAuth credentials from a local file and communicates with Google's authentication endpoint to refresh access tokens for calendar management (webhook.ts).- [DATA_EXFILTRATION]: Sends call transcripts and summaries to the Zo Computer API to generate and deliver recap emails to the user (webhook.ts).- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by incorporating untrusted call transcripts and summaries directly into the system prompt for subsequent interactions (webhook.ts).- Ingestion points: Call transcripts and analysis summaries ingested from the Vapi webhook and stored in a local DuckDB database (scripts/webhook.ts).- Boundary markers: Uses textual headers like 'PREVIOUS CALL HISTORY' but lacks strict delimiters or explicit instructions to the AI to ignore instructions embedded within the history.- Capability inventory: Execution of shell commands and Python scripts via Bun.spawn, file system access for credential tokens and databases, and network communication with Vapi, Google, and Zo APIs.- Sanitization: Uses parameterized SQL queries to prevent database injection but does not sanitize the natural language content of transcripts before prompt interpolation.

vapi