The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions include 'pip install graphifyy', which appears to be a typosquatted variant of the legitimate 'graphify' package referenced in the project's resource section (github.com/safishamsi/graphify).
[REMOTE_CODE_EXECUTION]: The 'graphify install' command downloads and executes external code to configure the agent's environment, which can lead to arbitrary code execution if the source is compromised.
[COMMAND_EXECUTION]: The skill implements a persistence mechanism by adding a daily crontab entry to automatically run synchronization scripts at 10 PM.
[COMMAND_EXECUTION]: The '/save' command automates 'git commit' and 'git push' operations, which can lead to unintentional data exposure if sensitive information is committed and pushed to remote repositories without manual review.
[CREDENTIALS_UNSAFE]: The documentation guides users to export sensitive API keys (Anthropic and Moonshot) as environment variables, making them accessible to the agent and other processes.
[DATA_EXFILTRATION]: The automated sync process extracts full chat histories and moves them into a vault that is potentially synchronized with remote repositories via the automated Git workflow.
[PROMPT_INJECTION]: The skill processes untrusted chat export data and re-injects it into the agent context during session resumption via the '/resume' command, creating an indirect prompt injection surface. * Ingestion points: 'claude_to_obsidian.py' processes exported Markdown files from Claude. * Boundary markers: None identified in the provided processing scripts. * Capability inventory: Subprocess calls, Git operations, and broad file system access. * Sanitization: The script extracts keywords for tagging but does not sanitize or filter the main content of the notes.

claude-code-memory-obsidian