The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It processes content from an external issue tracker (ticket titles, bodies, and comments) that may be modified by other users, and uses this information to drive agent behavior and subagent invocation.
Ingestion points: Issue tracker map and ticket bodies accessed during the 'Work through the map' phase.
Boundary markers: None; the agent is instructed to fetch and act on the full body of any related ticket.
Capability inventory: Invoking subagents (/research, /prototype), writing comments, closing issues, and performing environment-modifying tasks.
Sanitization: No sanitization or validation of the ticket content is implemented.
[CREDENTIALS_UNSAFE]: The instructions for 'Task' tickets explicitly direct the agent to record sensitive 'facts' like the 'credentials location' in the resolution comments on the issue tracker, which may be accessible to unauthorized parties.
[COMMAND_EXECUTION]: The skill facilitates automated 'Task' tickets that can perform 'manual work' such as 'provisioning access' or 'moving data'. In 'AFK' mode, these tasks are driven by the agent alone, potentially leading to unauthorized or unreviewed system modifications.

wayfinder