The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the open shell command (macOS) to launch a web browser to display the dashboard.html file created during the optimization process.
[EXTERNAL_DOWNLOADS]: The generated dashboard is designed to load the Chart.js library from an external Content Delivery Network (CDN) at runtime.
[REMOTE_CODE_EXECUTION]: By fetching and executing scripts from a CDN within the browser context of the generated dashboard, the skill introduces a remote code dependency that executes when the dashboard is opened.
[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface because it processes untrusted content from external "target skill" files and user-provided "test inputs" to autonomously mutate instructions.
Ingestion points: Reads content from user-specified file paths (Target skill) and test input scenarios.
Boundary markers: Absent; the skill does not use delimiters to isolate target skill content from its own internal reasoning or mutation logic.
Capability inventory: File system read/write (modifying the target SKILL.md), shell command execution (open), and the ability to run other skills repeatedly.
Sanitization: None; the agent is instructed to analyze failures and directly form hypotheses for prompt mutations based on the content of the target skill and its failed outputs.

autoresearch