The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with the local Elixir environment using standard tools such as mix compile, mix test, mix format, and mix credo. It also inspects the environment via claude --version to determine agent nesting depth. These executions are functional, restricted to the local project, and follow standard developer workflows.
[EXTERNAL_DOWNLOADS]: The skill facilitates research through specialist sub-agents (web-researcher, hex-library-researcher). The web-researcher agent is used for information gathering from documentation and community forums, while the hex-library-researcher is restricted to evaluating new libraries not yet present in the project. These tools do not perform automated code execution or installations.
[PROMPT_INJECTION]: The skill processes user-supplied descriptions and external review files, which represents an indirect prompt injection surface. This is effectively mitigated by the skill's 'Iron Laws', which mandate user review and explicit confirmation before any plan is executed, preventing autonomous exploitation of injected instructions.
Ingestion points: User-provided arguments and paths to review files ($ARGUMENTS).
Boundary markers: Structured planning templates and scratchpads are used to separate input data from generated tasks.
Capability inventory: Shell command execution (mix), file system writing, and agent spawning.
Sanitization: Mandatory user confirmation (Human-in-the-loop) for all transitions from planning to implementation.

plan