The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the retrieval and decryption of data (blobs) from the Walrus public storage network. Because the content of these blobs is provided by external users, it represents an untrusted ingestion point where an attacker could embed malicious instructions to influence agent behavior upon processing.\n
Ingestion points: The skill uses walrusClient.walrus.readBlob to fetch external content from the Walrus network (seal-sdk.md).\n
Boundary markers: There are no instructions provided to the agent to treat the contents of decrypted blobs as untrusted or to use specific delimiters when processing the retrieved text.\n
Capability inventory: The skill involves writing to a distributed storage network (walrus.writeBlob), executing cryptographic functions (sealClient.encrypt, sealClient.decrypt), and configuring off-chain computation environments (Nautilus).\n
Sanitization: The instructions do not include steps for sanitizing or validating the content of decrypted blobs before the agent uses them in further tasks.\n- [EXTERNAL_DOWNLOADS]: The skill directs the user to fetch documentation, software packages, and container images from various sources. These sources are either well-known public registries or vendor-managed domains associated with the skill's author.\n
Evidence: References to docs.wal.app, seal-docs.wal.app, and the container registry ghcr.io/mystenlabs/seal/key-server.\n- [COMMAND_EXECUTION]: The documentation includes standard commands for setting up the development environment by installing specific libraries.\n
Evidence: Instructions to run npm install @mysten/seal @mysten/sui found in SKILL.md and seal-sdk.md.

walrus-data-security