The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external, user-controlled content in Confluence, creating a potential surface for indirect prompt injection.\n
Ingestion points: The skill fetches page content, comments, and space instructions (AGENTS.MD) from Confluence as documented in SKILL.md and references/content.md.\n
Boundary markers: Instructions mandate that the agent reads selected content before performing summaries or edits and requires following space-specific protocols that define casing, structure, and canonical sources.\n
Capability inventory: The skill utilizes the 'twg' CLI for comprehensive content management, including creating, updating, moving, and deleting pages, blogposts, and spaces.\n
Sanitization: Security is maintained through process-level guardrails, including explicit user authorization for all mutations and the use of snapshot tokens to prevent race conditions during edits.\n- [SAFE]: The skill implements extensive security best practices for managing enterprise collaboration data.\n
[SAFE]: Destructive operations (archive, trash, purge) are protected by a two-stage check: explicit user authorization and a required '--yes' CLI flag for execution.\n
[SAFE]: The skill enforces a 'read-save-edit-update-verify' workflow for all non-trivial body edits to ensure data integrity and auditability.\n
[SAFE]: The instructions clearly distinguish between different content types and permission levels, ensuring the principle of least privilege is applied to sharing and access mutations.

twg-confluence