The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and processing external data from remote OpenAPI specifications and user-supplied hosted STDIO configurations.
Ingestion points: OpenAPI specification URLs and mcpServers JSON configuration objects defined in SKILL.md and references/mcp-yaml-variations.md.
Boundary markers: The skill explicitly instructs the agent to show a plain-English summary, a full YAML manifest or diff, and the exact CLI command to be executed, requiring explicit human confirmation before any applying changes.
Capability inventory: The skill uses the tfy command-line interface and a local helper script (tfy-api.sh) to perform network operations and update server registry entries.
Sanitization: The instructions require verifying that spec URLs are trusted and discourage the use of raw secrets, though the content within the specifications themselves is not automatically sanitized by the skill.
[COMMAND_EXECUTION]: The skill uses shell commands via the tfy CLI and a local API script to interact with the platform. These capabilities are constrained by safety rules that prohibit destructive operations like DELETE and require confirmation-gated dry-runs before final execution.
[CREDENTIALS_SAFE]: The skill implements best practices for secret management by utilizing tfy-secret:// references and per-user placeholders ({{API_KEY}}) instead of hardcoding sensitive data. It also explicitly forbids the agent from revealing or printing real credentials unless requested by the user.

truefoundry-mcp-servers