The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Elixir build and test tools, including mix compile, mix test, mix format, and mix ecto.migrate, as part of its core feature implementation and verification loop.
[COMMAND_EXECUTION]: Arbitrary Elixir code is executed using project_eval for feature-level behavioral smoke tests. The instructions specify using transactions and rollbacks to avoid persistent side effects during these evaluations.
[COMMAND_EXECUTION]: The skill integrates with local helper scripts such as format-elixir.sh, iron-law-verifier.sh, and error-critic.sh to enforce coding standards and analyze failures.
[EXTERNAL_DOWNLOADS]: The skill uses mix deps.get to resolve and download project dependencies from official repositories.
[PROMPT_INJECTION]: An indirect prompt injection surface is present because the skill processes instructions from markdown files within the .claude/plans/ directory.
Ingestion points: Plan and scratchpad files located at .claude/plans/*/plan.md and .claude/plans/*/scratchpad.md as described in SKILL.md.
Boundary markers: Absent; the instructions are consumed as natural language without explicit delimiters to separate task content from agent instructions.
Capability inventory: Project-wide file writes, arbitrary subprocess execution via the mix CLI, and dynamic Elixir evaluation via project_eval as detailed in execution-guide.md.
Sanitization: Absent; the skill does not perform escaping or validation of the natural language instructions read from plan files.

work