The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill performs invasive credential discovery across the local filesystem. The scripts/confluence_auth.py and scripts/confluence_api.py files are designed to read sensitive configuration files belonging to various AI IDEs and extensions, including Claude Desktop, Cursor, Windsurf, Antigravity, and VS Code extensions like Cline and Roo-Cline, to extract Confluence Personal Access Tokens.
[CREDENTIALS_UNSAFE]: The utility script scripts/confluence_api.py includes a discover-pat command that outputs discovered Personal Access Tokens (PATs) in plain text directly to the console output (stdout), which could expose them in logs or to other local processes.
[COMMAND_EXECUTION]: The Mermaid diagram rendering functionality in scripts/render_mermaid.py and scripts/mermaid_renderer.py executes the external mmdc command-line utility via subprocess.run to process generated diagram files.
[PROMPT_INJECTION]: The skill processes untrusted content from Confluence and Markdown files, creating an indirect prompt injection surface. Ingestion occurs during page downloads (HTML/Storage format) and uploads (Markdown), with no explicit boundary markers or validation logic to prevent embedded instructions from influencing the agent's behavior.
[SAFE]: Disabling SSL certificate verification (ssl.CERT_NONE and verify=False) is implemented in scripts/confluence_api.py and scripts/download_confluence.py. While a security best-practice violation, this is a common requirement for interacting with internal enterprise Data Center instances on private networks.

confluence-dc