The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from potentially untrusted external sources like CRM platforms and analytics webhooks to perform attribution analysis.
Ingestion points: Data enters the agent's context through CRM records (HubSpot, Salesforce), analytics platforms (GA4, PostHog), and third-party webhook payloads (SavvyCal, Stripe) as detailed in BY-BUSINESS-TYPE.md and FIRST-PARTY-STITCH.md.
Boundary markers: The skill emphasizes using a consistent 'source of truth' to reconcile data but does not prescribe specific prompt delimiters for the ingested content.
Capability inventory: The skill provides instructional code for data integration but does not execute subprocesses or dynamic scripts directly on the host system.
Sanitization: Instructions include specific routines for email normalization and 'anonymity guards' to prevent the leakage of personally identifiable information (PII) across domains.
[DATA_EXFILTRATION]: Technical guides in FIRST-PARTY-STITCH.md describe patterns for transmitting event data and user identifiers to analytics providers.
Evidence: Provided code snippets demonstrate sending identified user events to a remote analytics host using variables for the endpoint and API key.
Mitigations: The skill follows best practices by instructing users to avoid logging PII and provides logic to ensure only anonymous identifiers are passed in URL parameters.
[DYNAMIC_EXECUTION]: The skill provides logic for decorating links and processing webhooks at runtime.
Evidence: FIRST-PARTY-STITCH.md includes JavaScript for a link decorator that modifies outbound URLs based on current session state and a webhook handler for merging user identities.

marketing-attribution