The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because it instructs the agent to process external project code, git repositories, and build logs. If these sources contain malicious instructions, they could potentially influence the agent's behavior during the deployment process.
Ingestion points: Local project directory and remote git repositories.
Boundary markers: None specified to differentiate project content from instructions.
Capability inventory: The agent is authorized to perform API calls (project creation, deployment, log retrieval) and manage environment variables.
Sanitization: No specific sanitization or filtering of input data is mentioned beyond standard error handling.
[DATA_EXPOSURE]: The skill references the local file path ~/.build-host/credentials.json for authentication. This is an expected behavior for a deployment tool targeting the build.host service. The skill mitigates risk by explicitly instructing the agent never to print API keys, tokens, or full environment values to logs or chat history.

deploy-svelte