The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands via scripts/tfy-api.sh to interact with the TrueFoundry API and scripts/tfy-version.sh to detect installed software versions.
The tfy-api.sh script includes validation to prevent path traversal and only allows a set of standard HTTP methods.
It parses .env files using a line-by-line method rather than source, which is a safer practice.
[EXTERNAL_DOWNLOADS]: The skill contains instructions in references/prerequisites.md to automatically install the truefoundry CLI and pydantic via standard package managers (pip or uv). These dependencies are well-known and originate from the tool's vendor.
[PROMPT_INJECTION]: The skill incorporates robust defensive instructions to mitigate indirect prompt injection and unauthorized actions. It mandates user confirmation before selecting a workspace and explicitly warns the agent against fetching or ingesting content from third-party release pages (GitHub, HuggingFace) because they are untrusted and could contain adversarial instructions.
[CREDENTIALS_UNSAFE]: While the skill manages an API key, it provides clear instructions to use secret references (tfy-secret://) instead of hardcoding raw credentials in deployment manifests.

truefoundry-applications