The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local files and environment variables to construct experiment specifications and submits them to the Adaptyv Bio API.
Ingestion points: Reads protein sequences from candidates.json and project settings from environment variables like ADAPTYV_TARGET_ID (SKILL.md).
Boundary markers: None. The skill does not define specific delimiters or instructions to the agent to ignore potentially malicious embedded content within the sequence data or metadata.
Capability inventory: Uses the SDK and httpx to perform network operations, including experiment creation, submission, and result retrieval.
Sanitization: While the skill describes validation rules for amino acid strings (20 standard letters), it does not specify technical sanitization or escaping to prevent instructions within metadata fields from influencing the agent's behavior.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of the adaptyv-sdk directly from a GitHub repository.
Evidence: The command uv pip install "git+https://github.com/adaptyvbio/adaptyv-sdk.git@cdf207819ed5a58e0c127453626d2bce125c8064" (SKILL.md) downloads and installs code from an external source.
Risk Mitigation: The download is pinned to a specific commit hash (cdf207819ed5a58e0c127453626d2bce125c8064), which is a security best practice for supply chain integrity, though the source itself is a third-party vendor.
[COMMAND_EXECUTION]: The skill documentation includes shell commands for environment setup and API interaction.
Evidence: Includes uv pip install for package management and curl for direct REST API testing in SKILL.md.
[SAFE]: The skill provides best-practice guidance for handling sensitive information. It advises against hardcoding API keys, recommends using environment variables or .env files, and explicitly warns against printing or committing tokens to version control.

adaptyv