The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes external plugin manifest and configuration files, such as VS Code's package.json, IntelliJ's plugin.xml, and Zed's extension.toml. This creates an attack surface where malicious instructions could be embedded within the metadata or content of these user-supplied files, potentially targeting the AI agent during the analysis or build process.
Ingestion points: Scripts like check_vscode_manifest.mjs, check_intellij_plugin_xml.py, and check_zed_extension.py parse user-provided project files.
Boundary markers: There are no explicit delimiters or warnings to the agent to ignore instructions embedded in the analyzed files.
Capability inventory: The skill's intended workflow involves the agent executing build, test, and packaging commands, which requires significant filesystem and execution privileges.
Sanitization: The diagnostic scripts do not perform sanitization of external text before it is presented to the agent.
[SAFE]: The skill makes numerous references to official technical documentation and specifications, which are recognized as trusted and well-known sources.
The skill references the OSGi Core specification on docs.osgi.org, which is the official foundation site for OSGi standards.
Documentation links target official vendor sites including code.visualstudio.com (Microsoft), plugins.jetbrains.com (JetBrains), and zed.dev (Zed Industries).

develop-editor-plugins