The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural language requests to generate automation blueprints and shell commands for recurring tasks. This pattern creates a surface for indirect prompt injection where malicious instructions in a user's task description could be persistently stored and executed within a cron job.
Ingestion points: The skill ingests untrusted natural language data from user prompts in SKILL.md (e.g., "automation-blueprint every weekday...") to define the logic of scheduled operations.
Boundary markers: There are no explicit delimiters or system instructions defined in the workflow to separate user-provided task content from the agent's control logic, increasing the risk of the agent following instructions embedded in the data.
Capability inventory: The skill provides access to the cronjob_manage tool and generates hermes cron create shell commands. These capabilities allow for the persistent execution of instructions and potential delivery of outputs to external platforms like Slack or Discord.
Sanitization: The analyzed files do not describe any sanitization, escaping, or validation steps to ensure that user-provided text is not treated as executable instructions when placed into the --prompt argument of generated CLI commands.

omh-automation-blueprint