The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data, such as user-provided usage examples and external skill packages, which could contain hidden instructions targeting the agent.
Ingestion points: Ingests external SKILL.md files, repository conventions, and representative prompts from users.
Boundary markers: No explicit delimiters or boundary markers are instructed to separate ingested data from the agent's core instructions.
Capability inventory: The skill possesses extensive capabilities including file writing/editing, shell command execution, and script execution.
Sanitization: There are no instructions for sanitizing or escaping the external content before interpolation.
[DYNAMIC_EXECUTION]: The skill facilitates the generation and testing of deterministic scripts as part of the skill creation and validation process.
Evidence: The instructions explicitly mention moving deterministic work into a scripts/ directory and running those scripts to verify behavior (e.g., "Run and test deterministic scripts that changed").
[COMMAND_EXECUTION]: The skill acknowledges and defines side effects that include running shell commands and opening applications.
Evidence: The 'Define side effects and output' section states, "State whether the skill answers, audits, edits files, runs commands, opens applications, or changes external state."

skill-creator