outreach-icp-finder
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, specifically the text content of replies from outreach leads, which could potentially contain malicious instructions intended to influence the agent during the labeling phase.
- Ingestion points: Untrusted data enters the context through CSV file imports or via MCP tool calls such as
get_conversation_messageswhich fetches raw reply text. - Boundary markers: The instructions direct the agent to reason over JSON output from a deterministic script, providing a layer of separation, but the agent is still required to manually read and label a sample of the raw text replies.
- Capability inventory: The skill has the capability to read local files, invoke MCP tools to fetch communication logs, and trigger sibling skills through the
sendPromptmechanism. - Sanitization: The skill uses a structured Python engine to normalize labels and compute statistics, but the agent's interpretation of raw lead messages lacks formal sanitization against adversarial text.
- [COMMAND_EXECUTION]: The skill executes a bundled Python script (
scripts/analyze.py) to perform data processing and statistical analysis on the provided outreach datasets.
Audit Metadata