The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/fetch-x.sh manages browser session cookies stored at ~/.config/llm-wiki/x-profile/cookies.json. It includes a command to harvest cookies from a running Chrome instance using agent-browser --auto-connect cookies. Handling and storing session cookies for external services represents a significant credential management risk.\n- [COMMAND_EXECUTION]: The skill utilizes several shell and Python scripts (scripts/setup.sh, scripts/connect-obsidian.sh, scripts/fetch-x.sh) to perform file system operations, directory creation, and interactive browser control via the agent-browser tool. These scripts process user-provided paths and arguments to manage the wiki infrastructure.\n- [EXTERNAL_DOWNLOADS]: The skill performs network requests to api.vxtwitter.com (a third-party service) to retrieve tweet content. Additionally, scripts/download-images.py downloads image files from arbitrary, unverified URLs provided through standard input.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from external websites and social media platforms to update its persistent wiki files, creating an indirect prompt injection surface.\n
Ingestion points: scripts/extract-html.py (local HTML files) and scripts/fetch-x.sh (external X/Twitter content).\n
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' instructions for the ingested text during the wiki update process.\n
Capability inventory: File system write access (wiki updates), network communication, and script execution capabilities.\n
Sanitization: Absent. The scripts extract and pass text content without sanitizing or escaping potential prompt injection payloads that might be embedded in the source material.

llm-wiki