The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by instructing the agent to read and process potentially untrusted content from existing workspace documents.
Ingestion points: The read_doc tool is used to ingest existing content into the agent's context (SKILL.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to differentiate between the ingested document content and system instructions.
Capability inventory: The agent possesses capabilities to publish_doc, update_doc, and set_access (SKILL.md), which could be manipulated by malicious instructions within a read document.
Sanitization: There is no evidence of sanitization or validation of the content retrieved from documents before it is processed.
[EXTERNAL_DOWNLOADS]: The skill refers to an external command-line tool for versioning and publishing.
Evidence: Mentions installation of basalt-cli via the NPM registry (npm i -g basalt-cli). This is a vendor-owned resource associated with the author 'cofoundy'.

basalt