The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install a community-developed NPM package, confluence-cli (npm install -g confluence-cli). This tool is not an official Atlassian product, and its maintenance and security are managed by a third-party developer rather than a recognized trusted organization.
[CREDENTIALS_UNSAFE]: The instructions guide the agent and user to manage sensitive Atlassian API tokens and email addresses using environment variables (CONFLUENCE_API_TOKEN, CONFLUENCE_EMAIL). The skill also suggests persisting these credentials in shell profile files such as ~/.bashrc or ~/.zshrc, which are sensitive file paths often targeted for credential harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill provides the agent with the capability to read and search content from external Confluence instances. If the content being processed contains malicious instructions, it could potentially influence the agent's downstream actions.
Ingestion points: Data retrieved from confluence read and confluence search commands.
Boundary markers: None identified in the prompt instructions to isolate external content from agent directives.
Capability inventory: The agent can execute shell commands via the bash tool and perform network operations through the CLI tool.
Sanitization: No specific sanitization or validation of the retrieved Confluence content is mentioned.

confluence-cli