The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the Salesforce CLI (sf) and local Node.js scripts to manage component metadata and query Salesforce org information.
Evidence: scripts/namespace.mjs uses execFileSync to run sf commands. It includes a regex-based allow-list (SAFE_SF_ARG_RE) to mitigate shell injection on Windows systems.
Evidence: Various reference files (e.g., references/create-component-metadata.md, references/get-object-documentation.md) instruct the agent to execute scripts like query-versions.mjs and create-component.mjs.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and parse external Salesforce metadata files (.md-meta.xml), which could contain malicious instructions designed to influence the agent's behavior during the planning or summary phases.
Ingestion points: scripts/read-component.mjs reads and parses user-specified XML metadata files from the local filesystem.
Boundary markers: The instructions in references/read-component-metadata.md do not specify delimiters or instructions to ignore potential commands embedded in the component descriptions or metadata fields.
Capability inventory: The skill has the ability to write files (scripts/create-component.mjs) and execute shell commands (scripts/namespace.mjs).
Sanitization: While the scripts handle XML escaping, they do not perform semantic sanitization to detect or neutralize natural language prompt injection within the ingested data.

avonni-dynamic-components