The Agent Skills Directory

[DATA_EXFILTRATION]: The script may transmit sensitive session cookies to non-Atlassian domains.
In scripts/confluence-browser-fetch.js, the resolveInputToPageId function accepts a URL as input and performs a fetch request to it, including the Atlassian session Cookie header without verifying the destination domain.
The downloadAttachments function in the same file follows absolute URLs found in the _links.download metadata returned by the Confluence API and includes the Cookie header in those requests, which could lead to cookie theft if the API returns a malicious URL.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection vulnerability surface.
Ingestion points: Content is fetched from Confluence REST API and browser-rendered HTML in scripts/confluence-browser-fetch.js and stored in the raw/confluence/ directory.
Boundary markers: Absent. The fetched content is saved raw without delimiters or instructions to ignore embedded prompts.
Capability inventory: The skill performs file writes (fsp.writeFile) and network requests (fetch).
Sanitization: Filenames are sanitized for path safety using slugify and safeName in scripts/lib.js, but the page content remains unsanitized.
[EXTERNAL_DOWNLOADS]: The script downloads page data and attachments from Atlassian Cloud domains as part of its core functionality.
[COMMAND_EXECUTION]: The Node.js script executes filesystem management operations and network requests.
[SAFE]: The skill correctly uses slugify and safeName in scripts/lib.js to prevent path traversal vulnerabilities when creating directories and files based on external metadata.

confluence-browser-fetch