The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from the repository, including source code, commit history, and pull request descriptions, which could contain malicious instructions designed to influence the agent.
Ingestion points: Codebase files, repository history (commit messages, blame), pull request descriptions, and design documentation (READMEs, ADRs).
Boundary markers: Includes specific instructions to stay within "Boundaries" (explain only, never run code) and use "inference labels" to distinguish facts from inferences.
Capability inventory: Capability to modify documentation files, update PR descriptions, and create temporary simulation files.
Sanitization: The instructions do not explicitly describe sanitization or escaping mechanisms for the processed data.
[DYNAMIC_EXECUTION]: The skill generates interactive "micro-world" simulations in the form of self-contained HTML/JavaScript files to illustrate code logic.
Evidence: Instructions in references/micro-world.md describe creating these throwaway files within a .tmp/ directory.
[COMMAND_EXECUTION]: The skill uses command-line tools to update repository metadata and documentation.
Evidence: references/durable-home.md references the use of git and gh commands for persistence tasks, requiring explicit user approval.

explain-code