The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external OAuth providers (e.g., GitHub, Google, Slack), creating a potential surface for indirect prompt injection if an attacker-controlled provider returns malicious instructions in metadata fields.
Ingestion points: External provider data enters the system through OAuth callbacks and API responses, specifically handled by verifier modules and stored as provider_data as described in references/authorization.md and references/providers.md.
Boundary markers: The skill emphasizes the use of intent_reference to tie completions to specific server-initiated actions and allowed_completion_origins to restrict the popup flow, reducing the risk of cross-tenant or unauthorized injections.
Capability inventory: The skill performs network operations via the Req library and executes database writes through Ecto.Multi (documented in references/connections.md and references/consumption.md).
Sanitization: The skill implements a robust security model for external data, including a mandatory allowlist for provider_data and the Tango.Validation.sensitive_key?/1 utility which automatically filters out secrets, tokens, and signatures from ingested metadata.

tango