playbook-social-link-finding
Pass
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTML content fetched from external websites, which is then analyzed by an LLM to verify profile ownership.
- Ingestion points: Raw HTML content is retrieved from external domains using the
scrape-websitetool as specified inclay-table.md(Column 3) andclay-workflow.md(Node 3). - Boundary markers: The verification prompt in
SKILL.md(Section 6) uses the delimiter "PER-ROW DATA (appended last)" to distinguish system instructions from external evidence. - Capability inventory: The skill utilizes network-enabled tools (SERP and enrichment APIs) and executes logic within Python and JavaScript nodes to process data and output verified social URLs.
- Sanitization: The skill uses regular expression patterns in
clay-workflow.md(Node 4) to extract specific URL slugs, which provides a layer of filtering by only passing specific matched strings and snippets to the model, though this does not fully mitigate instructions hidden within page titles or metadata. - [COMMAND_EXECUTION]: The
clay-workflow.mdfile provides CLI command examples for deploying the skill to the Clay platform usingclay workflowsandclay routines. These are documentation for manual setup and are not executed dynamically or autonomously by the agent to target the user's host environment.
Audit Metadata