The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a suite of specialized CLI tools (such as confluence content, confluence space, and confluence tree) to perform operations. These are vendor-provided tools for interacting with the Atlassian ecosystem.
[DATA_EXFILTRATION]: The instructions describe a workflow that involves reading content from remote Confluence instances and saving it to local files for processing (references/editing.md). This is a legitimate part of the editing lifecycle but involves data movement between the remote service and local storage.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted data from Confluence pages that could be modified by third parties.
Ingestion points: Data enters the agent via confluence content get and search commands (defined in references/content.md).
Boundary markers: No specific delimiters are defined to separate untrusted content from instructions.
Capability inventory: The agent has the ability to execute commands and write local files (SKILL.md).
Sanitization: No explicit sanitization of fetched content is described before the agent processes it.

twg-confluence