The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from potentially untrusted external sources which can influence agent planning and decision-making loops.
Ingestion points: The agent reads project files, configuration files like suite.yaml, and the JSON output of the gimble runs command. It also analyzes screenshots of products under test using multi-modal models.
Boundary markers: There are no explicit instructions to the agent to treat external inputs as untrusted or to ignore instructions embedded within the data it processes.
Capability inventory: The agent has the ability to execute shell commands via gimble, steering sessions, and publishing results using the gh CLI.
Sanitization: The instructions do not specify any validation or sanitization requirements for the YAML/JSON inputs or project state data.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple CLI tools to perform its core functions.
Evidence: Instructions direct the agent to invoke gimble run, gimble steer, gimble watch, and gh (GitHub CLI) to perform reviews, steer sessions, and publish artifacts.
[DATA_EXFILTRATION]: The skill contains capabilities for transmitting data and artifacts to external locations.
Evidence: The gimble upload-artifact command and gh CLI are used to publish findings and screenshots to "configured public artifact destinations" and GitHub repositories. While intended for reporting, this constitutes a data transmission path from the local environment.

gimble-runs