The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local command-line tools solidcraft-cli (to process design history, run reproduction scripts, and take snapshots) and gh (to search and create issues on GitHub). These operations are well-scoped to the skill's primary purpose of bug reporting.
[DYNAMIC_EXECUTION]: The skill involves generating a JSON-based command script (repro.json) from a user's design history and executing it via solidcraft-cli run. While this involves executing instructions derived from user data, the operations are restricted to the SolidCraft CAD engine.
[INDIRECT_PROMPT_INJECTION]: The skill processes external design files and error logs which could potentially contain malicious instructions. The skill addresses this surface through the following:
Ingestion points: Loading command history from .solidcraft files and reading crash reports.
Boundary markers: Instructions require the agent to show the user the finished report and screenshots for approval before filing.
Capability inventory: The skill uses file system access for reproduction and gh for network-based issue submission to the vendor's repository.
Sanitization: The skill includes explicit instructions to strip sensitive information like home directories, usernames, and project names from messages and logs before they are included in a report.

solidcraft-reporting