The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill defines logic for processing sensitive identity artifacts such as authorization codes and access tokens, creating an interface where data provided by an external source is incorporated into cryptographic proofs.
Ingestion points: Functions defined in SKILL.md such as createDPoPProof and calculateAuthCodeJti ingest external parameters including authCode, accessToken, and jti.
Boundary markers: The instructions do not specify explicit delimiters or "ignore" instructions for these inputs within the generated JWT payload, relying instead on the cryptographic structure of the DPoP proof.
Capability inventory: The skill utilizes crypto.subtle.sign for ECDSA signatures, crypto.getRandomValues for randomness, and SHA-256 for hashing operations.
Sanitization: The implementation employs sanitizeHTU for URI normalization and applies SHA-256 hashing to tokens (calculateATH) and authorization codes before they are used as claims, which limits the potential for direct payload injection.
[External Tool Integration]: The skill mentions the use of a Model Context Protocol (MCP) server to facilitate access to developer documentation. This involves fetching and executing a utility package.
Evidence: The skill references npx -y @google/mcp-developer-knowledge-server as a method to query documentation.

dpop-adoption