The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands such as elm-review and elm-format --yes. These are standard tools in the Elm ecosystem used for static analysis and code style enforcement. The instructions emphasize using actual authorized paths and discourage silent execution or automatic installation of tools.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection as it ingests and processes untrusted content from .elm source files and elm.json configurations. An attacker could embed malicious instructions in code comments or project metadata to influence the agent's behavior. However, the risk is mitigated by the skill's focus on structured tool output (exit codes, specific failure types) and the requirement for human review of any code changes (diffs).
Ingestion points: Elm source code files (.elm) and project configuration files (elm.json).
Boundary markers: The skill uses a structured output template and distinct reporting categories (PASS/FAIL/UNVERIFIED) to separate data processing from analysis.
Capability inventory: File system read access for analysis and write access via elm-format. Shell command execution for linting tools.
Sanitization: The skill relies on the native Elm toolchain (elm-review) for parsing, which acts as a validation layer for the content being analyzed.

codeguard-elm