The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands and performs local file system operations.
Evidence: The skill uses npx skills find "<name>" in SKILL.md and references/conventions.md to query an external registry for naming collisions. It also writes content to SKILL.md and test files (e.g., tests/<name>/README.md) in the project's local or global directories.
Command Injection: The <name> parameter is interpolated directly into the shell command string npx skills find "<name>" without any instructions for sanitization or validation. This allows for potential command injection if a user or the agent provides a name containing shell metacharacters (e.g., ;, &&, |).
[REMOTE_CODE_EXECUTION]: The skill executes code fetched from an external registry at runtime.
Evidence: The instruction to use npx skills triggers the download and execution of the skills package from the npm registry. The execution is unpinned, meaning it will run the latest available version of the package, which presents a dependency risk.
[EXTERNAL_DOWNLOADS]: The skill initiates downloads from a well-known public registry.
Evidence: The npx command fetches the skills package from the official npm registry during the naming collision check.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its handling of untrusted input.
Ingestion points: SKILL.md Step 1 gathers task domains, use cases, and requirements from the user or context.
Boundary markers: Absent. The skill does not provide instructions to use delimiters or ignore embedded instructions when drafting the new skill file.
Capability inventory: The skill has the ability to write files to the system and execute shell commands (SKILL.md).
Sanitization: Absent. There is no requirement or step to validate or sanitize the gathered input before it is interpolated into the generated SKILL.md body.

skill-writing