The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes instructions from a backlog (‘vision/backlog’) and a methodology file (‘agents/coordinator-yolo.md’). If these files are influenced by untrusted external sources, such as pull request descriptions or issue comments integrated into the backlog, the agent could execute malicious instructions autonomously.
Ingestion points: vision/backlog, agents/coordinator-yolo.md (via Step 1 and Step 4 prompt).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are defined for the ingested backlog tasks.
Capability inventory: The skill can perform file system modifications (via git commits) and execute arbitrary build/test commands as part of its validation cycle.
Sanitization: No explicit sanitization or filtering of the backlog content is mentioned; the skill relies on automated build/test failure detection to roll back changes.
[COMMAND_EXECUTION]: The skill uses shell commands to manage branches (git checkout, git rev-parse) and executes project-specific build and test suites autonomously. While these are standard development operations, the lack of human-in-the-loop validation for the tasks being executed increases the risk that a task could result in unintended command execution if the backlog is compromised.

coordinator-yolo