The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing external 'briefs' through an LLM-based implementer, creating a vulnerability surface for instruction injection.
Ingestion points: Untrusted instructions enter the system via the brief.txt file (referenced in SKILL.md and dispatch-and-poll.md).
Boundary markers: Documentation suggests using structured headers (GOAL, CHANGE, GATES) but correctly warns that these do not constitute a security boundary.
Capability inventory: The implementer (Aider) possesses file-system write access and the ability to execute repository-defined shell commands via linting and testing hooks.
Sanitization: The skill relies on manual human review of diffs ('Review and Land' policy) rather than programmatic sanitization of the brief content.
[EXTERNAL_DOWNLOADS]: The instructions recommend the installation of the aider-chat Python package from PyPI. This is a well-known open-source tool for AI-assisted development.

aider-delegate