The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation highlights that plugin configurations may contain !!js expressions. These expressions are explicitly evaluated by the Loader during the activation of child plugins, which constitutes a dynamic execution surface within YAML configurations.
[INDIRECT_PROMPT_INJECTION]: The skill outlines a workflow where the agent creates bundle files (package.json, cordis.patch.yml) in a workspace and then installs them using the plugin_manager. This sequence creates a vulnerability where untrusted data processed by the agent could influence the content of the bundle, leading to the execution of unintended code in the host process.
Ingestion points: The agent writes configuration and manifest files (e.g., review-preset/package.json, review-preset/cordis.patch.yml) based on instructions or data.
Boundary markers: No specific delimiters or warnings for the agent to ignore embedded instructions within these workspace files are described.
Capability inventory: The skill utilizes the plugin_manager (specifically the install_bundle action) which has the capability to execute code in the Host process. It also utilizes file-read and assumes access to workspace file-writing.
Sanitization: The instructions do not specify any validation or sanitization steps for the content of the generated patches before installation.
[COMMAND_EXECUTION]: The plugin_manager tool's install_bundle action is used to execute plugin code in the Host process. The skill correctly notes that this action requires high-level access or explicit user approval due to its security implications.

editing-cordis-compositions