The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from model provider error messages and hints to determine recovery steps. \n
Ingestion points: The agent is instructed to read the error and hint fields from job_get and jobs_wait outputs, which are sourced from external model providers (file: SKILL.md). \n
Boundary markers: There are no instructions to use delimiters or ignore embedded instructions within these error fields. \n
Capability inventory: The skill utilizes tools like model_run (execution of generations), recommend, job_get, and usage (file: SKILL.md). \n
Sanitization: No specific sanitization or validation of the error text is mentioned before the agent interprets it.
[EXTERNAL_DOWNLOADS]: The skill contains instructions for installing additional sibling skills from the vendor's repository. \n
Evidence: npx skills add scenario-labs/skills --skill <name> (file: SKILL.md). This download targets the official repository of the skill's author.
[COMMAND_EXECUTION]: The instructions involve the use of various platform-specific CLI tools and API calls to manage generation jobs and query usage data. \n
Evidence: Calls to job_get, usage, teams_list, and model_run are integrated into the recovery workflows (file: SKILL.md).

scenario-moderation