The Agent Skills Directory

[SAFE]: No malicious patterns such as credential theft, persistence, or unauthorized command execution were detected. The skill instructions are consistent with the stated purpose of marketing operations management and originate from a known vendor.
[DATA_EXFILTRATION]: The skill references integrations with well-known technology companies and cloud services (e.g., Salesforce, HubSpot, Vercel, Slack) and official Airtable domains. There is no evidence of data being sent to suspicious, untrusted, or unknown external endpoints.
[PROMPT_INJECTION]: The instructions do not contain attempts to override agent safety guidelines, extract system prompts, or bypass content filters.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were found. Mentions of external application deployment (e.g., custom apps on Vercel) are descriptive of user-directed development workflows and do not involve the agent executing untrusted remote scripts.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from marketing requests and influencer pitches.
Ingestion points: Marketing Requests table via intake forms (documented in SKILL.md and references/sub-workflows.md).
Boundary markers: Not explicitly defined in the schema, though the skill recommends using structured fields.
Capability inventory: MCP tools for reading and writing Airtable records (e.g., list_records_for_table, update_records_for_table).
Sanitization: The skill explicitly mandates a 'copilot pattern' (AI drafts followed by human review) as a primary mitigation against the automated execution of potentially malicious instructions embedded in user data.

marketing-ops