The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from various user channels to update profiles or execute erasure requests, which creates a surface for indirect prompt injection if an AI agent is utilized to interpret or automate these requests.
Ingestion points: Request payloads in the /api/me/profile and /api/me/erasure endpoints, as well as natural language requests from informal channels (WhatsApp, social media) described in the operational workflow.
Boundary markers: The implementation relies on standard API request boundaries; no explicit LLM delimiters or 'ignore' instructions are present in the provided data-processing templates.
Capability inventory: Modifies user records in the database, revokes authentication sessions, and triggers data propagation to external service operators.
Sanitization: The code implements robust sanitization by whitelisting specific fields (EDITABLE_FIELDS) and enforcing type checks, which effectively limits the impact of malicious inputs.
[DATA_EXFILTRATION]: The skill includes an automated mechanism to propagate data changes to third-party services like Stripe, Mixpanel, and Customer.io. These are documented as necessary steps for regulatory compliance (LGPD Art. 18, § 6º) and target well-known, trusted service providers.
[CREDENTIALS_UNSAFE]: Security-sensitive values, including the HMAC key used for anonymizing backup data, are correctly accessed via environment variables (process.env.BACKUP_SUPPRESSION_KEY_B64) rather than being hardcoded in the scripts.

lgpd-dsar