The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources, creating a vulnerability to indirect prompt injection where malicious instructions could be embedded in data processed by the agent.
Ingestion points: The ANALYZER agent processes email reply text from the Gmail MCP. Lead and company data are ingested from Apollo MCP and Clay MCP as described in SKILL.md and reference/outreach.md.
Boundary markers: There are no specific delimiters or instructions for the agent to disregard embedded commands within the processed email text or metadata.
Capability inventory: The skill possesses significant capabilities, including updating the Epiphan CRM (HubSpot), managing Apollo email sequences via the apollo_emailer_campaigns_remove_or_stop_contact_ids tool, and writing files to the local system.
Sanitization: The provided instructions do not specify any validation or sanitization steps for the incoming text content before it is used for decision-making or CRM updates.
[COMMAND_EXECUTION]: The skill directs the agent to perform file system write operations in a specific local path.
Evidence: The Emit Outcome Sidecar section in SKILL.md instructs the agent to write a JSON object containing session metrics and status to ~/.claude/skill-analytics/last-outcome-sales-revenue.json upon completion. This involves writing to a hidden directory in the user's home folder.

sales-revenue