The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The mandatory workflow in SKILL.md directs the agent to ingest content from the project's Cargo.toml file using the explore-codebase tool. This creates a surface where malicious content within project files could influence agent behavior.
Ingestion points: Local Cargo.toml file read via fuse-ai-pilot:explore-codebase as specified in SKILL.md.
Boundary markers: The skill does not provide specific delimiters or instructions for the agent to disregard embedded content within the ingested file.
Capability inventory: The skill utilizes tools with network access (research-expert, mcp__context7__query-docs) and template-based file creation capabilities.
Sanitization: There is no mention of filtering or validation for the data ingested from the project codebase.
[COMMAND_EXECUTION]: The references/templates/cargo-toml-stack.md template configures a TcpListener to bind to 0.0.0.0. This configuration causes the resulting application to listen on all network interfaces, which may expose the service to unintended external traffic if the environment is not properly secured.

rust-ecosystem-crates