The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a consumer for data from external integrations like GitHub, Zendesk, and error tracking systems, which creates a surface for indirect prompt injection attacks.
Ingestion points: Data enters the context via tools such as inbox-reports-list, inbox-reports-retrieve, and posthog:execute-sql (in SKILL.md), which ingest summaries and signals generated by external sources.
Boundary markers: The instructions do not specify the use of delimiters or specific safety prompts to isolate untrusted external content.
Capability inventory: The skill possesses significant capabilities including claiming reports, linking pull requests, modifying project source configurations, and initiating background agent tasks via inbox-report-actions-run (in SKILL.md).
Sanitization: There are no defined procedures for sanitizing or validating the content of external signals before they are used to inform agent actions.
[PROMPT_INJECTION]: The skill includes instructional guidance on how to report system-held reports to the user, specifically advising against misattributing system judge decisions to the user. This is a benign directive aimed at ensuring technical accuracy in reporting and does not constitute a malicious concealment or jailbreak attempt.

inbox-exploration