The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow involves running make check to validate newly created skills. This is a standard and expected validation procedure within the repository environment.
[PROMPT_INJECTION]: The skill processes user-supplied arguments and external information from web searches, creating a surface for indirect prompt injection.
Ingestion points: User-provided $ARGUMENTS and documentation retrieved from web research (SKILL.md, Step 2).
Boundary markers: The instructions do not define specific markers or delimiters to separate untrusted external data from the skill's internal logic.
Capability inventory: The skill has the ability to write files to the skills/ directory and execute local commands via make.
Sanitization: The workflow lacks specific requirements for sanitizing or escaping retrieved content before it is incorporated into the new skill files.
[DATA_EXFILTRATION]: The skill is instructed to recursively read the .local/ directory to gather project context. This access is restricted to the local workspace and is intended for research during the skill authoring process.

add-skill