The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via remote MCP server configurations and OpenAPI specifications.
Ingestion points: Remote OpenAPI spec URLs and MCP server URLs provided in manifests (SKILL.md, references/manifest-schema.md).
Boundary markers: The SKILL.md contains explicit security warnings regarding remote specs, but no technical delimiters are enforced in the processing logic.
Capability inventory: Registered MCP servers provide tools that the agent can execute; the scripts/tfy-api.sh script enables the registration of these servers which then grant the agent new tool capabilities.
Sanitization: Content from remote specs is parsed by the platform gateway but not sanitized for instruction-like patterns before being converted into agent tools.
[EXTERNAL_DOWNLOADS]: The skill facilitates connections to remote MCP endpoints and fetches remote OpenAPI specifications at runtime for conversion into agent-accessible tools. It also references pinned container images from trusted registries like Amazon ECR and GHCR in references/container-versions.md.
[COMMAND_EXECUTION]: The skill uses Bash to execute helper scripts (scripts/tfy-api.sh, scripts/tfy-version.sh) for API communication and environment verification. It also leverages the TrueFoundry CLI (tfy) for applying resource manifests as described in references/cli-fallback.md.

mcp-servers