The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes scripts/discover-rust-gates.mjs to parse repository configuration files, including Cargo.toml, Justfile, mise.toml, Makefile, and GitHub workflows (.github/workflows/*), to identify valid build and test commands. This creates a surface for indirect injection if a repository is malicious.
Ingestion points: The skill reads contents of repository-controlled files using fs.readFileSync within scripts/discover-rust-gates.mjs.
Boundary markers: The script outputs raw command strings found in the files without explicit boundary markers or warnings to ignore embedded instructions when presenting them to the agent.
Capability inventory: The agent is instructed to use these discovered commands for repo-native verification and possesses the capability to execute shell commands and modify code.
Sanitization: The discovery script extracts commands using regular expressions but does not perform safety validation on the extracted strings. If a malicious repository contains harmful commands within its workflows or task runners, the agent could potentially be misled into suggesting or executing them.

rust-expert