fit-engagement-matrix
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data (contact name, email, and activity) from webhooks or form fills. This data is subsequently processed by agent nodes (LinkedIn Finder and ICP Judgment).
- Ingestion points:
SKILL.mdandreferences/workflow-graph.mddefine trigger nodes that receive external payloads. - Boundary markers: Agent prompts in
references/scoring-code.mduse bracketed interpolation (e.g.,{{contact_name}}) and include explicit constraints to use only provided info and avoid inventing facts. - Capability inventory: The workflow has the capability to write back to the installer's Audiences records (
update-audiences-record). It does not have access to send emails or write to external CRMs automatically. - Sanitization: The
Clean contactcode node performs basic validation, such as identifying email types and stripping URL protocols from domains. - [DYNAMIC_EXECUTION]: The skill builds a workflow containing Python script nodes. These scripts are provided as static, auditable templates in
references/scoring-code.mdand perform deterministic logic (headcount math, industry keyword matching). The execution environment is the platform's standard sandboxed code-node runner. - [COMMAND_EXECUTION]: The skill instructions guide the agent to use the
clayCLI for environment verification (clay whoami) and workflow deployment. These are platform-specific administrative actions necessary for the skill's primary purpose and do not involve executing untrusted shell commands.
Audit Metadata