The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests untrusted external content (e.g., LinkedIn posts, company news, and enrichment data) and interpolates it into AI-generated email snippets.
Ingestion points: Data enters the agent context in SKILL.md (Step 5 and Step 7) through enrichment lookups for LinkedIn posts and company news.
Boundary markers: The skill uses fallback=none markers within lead tokens (Step 9.8) to handle missing or potentially problematic data.
Capability inventory: The skill has the capability to write to the Audiences database (Step 8) and modify sequence steps in the Campaigns module (Step 9).
Sanitization: Mitigations are present in SKILL.md (Step 9.8), which instructs the AI to only restate facts and not infer intent, and in Step 9.10, which requires manual previews and automated spam checks before finalizing drafts.
[COMMAND_EXECUTION]: The skill utilizes the clay CLI to perform searches, manage workflows, and update campaigns. These operations are core to the vendor-provided functionality (clay-run) and include multiple mandatory user-confirmation halts before any mutations are executed.

icp-outbound-campaign