The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external Cosense (formerly Scrapbox) pages. This content could contain malicious instructions designed to manipulate the agent's behavior, especially since the skill has capabilities to modify data.
Ingestion points: Data enters the agent's context through commands like browsePage, readPage, and searchFullText as described in SKILL.md and read-page.md.
Boundary markers: There are no explicit instructions or delimiters defined to isolate the untrusted page content from the agent's system instructions.
Capability inventory: The skill possesses significant capabilities including editing pages (submitEdit), deleting files (deleteFile), and uploading content (uploadFile) as detailed in edit-page.md and upload-file.md.
Sanitization: The instructions do not specify any sanitization or validation of the content retrieved from external pages before it is processed or summarized.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch external data and recommends installing external dependencies.
Findings: Uses curl to fetch images and videos from Gyazo as noted in read-page.md.
Findings: Instructs the user or environment to install the @helpfeel/cosense-cli package from the npm registry.
[COMMAND_EXECUTION]: The skill operates by executing various subcommands of the cosense CLI tool.
Findings: Invokes shell commands such as cosense browsePage, cosense previewEdit, and cosense uploadFile to interact with the Cosense platform.

cosense