The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from an external feedback stream which could lead to manipulation of agent behavior.
Ingestion points: As described in Step 1h of SKILL.md, the skill reads friction reports filed by other agents from the meta/feedback stream using the exarchos_event query tool.
Boundary markers: The instructions do not specify the use of delimiters or specific warnings to ignore instructions embedded within the processed message content.
Capability inventory: The skill has a write capability through exarchos_orchestrate, which is used to automatically draft and file GitHub issues based on the findings (Step 6).
Sanitization: There is no mention of sanitizing or escaping the message content from the feedback stream before interpolating it into issue drafts or using it for root cause categorization.
[COMMAND_EXECUTION]: The skill facilitates actions that interact with external project management systems.
The skill instructs the agent to use exarchos_orchestrate with the create_issue action. While this is a functional requirement for the skill's purpose, it represents a capability to affect external state based on the interpretation of ingested data.

dogfood