The Agent Skills Directory

[DATA_EXFILTRATION]: The skill processes local data files including targeted_candidates.json and brief.md, and outputs to triaged_candidates.json. There are no network operations, attempts to reach external domains, or access to sensitive system credentials (e.g., .ssh, .aws).
[REMOTE_CODE_EXECUTION]: No patterns for downloading and executing remote scripts or packages were found. The skill does not use shell commands, eval, or subprocess calls.
[PROMPT_INJECTION]: The skill uses instructional logic to define priority and routing. While it identifies a user-provided brief.md as the source of truth for client rules, it does not contain instructions to bypass safety guidelines, reveal system prompts, or ignore core behavioral constraints.
[INDIRECT_PROMPT_INJECTION]: The skill processes data that may originate from external sources (news excerpts).
Ingestion points: Signal excerpts and titles are read from targeted_candidates.json.
Boundary markers: Instructions do not define explicit delimiters to separate external data from system instructions.
Capability inventory: The skill lacks dangerous tools or capabilities, such as network access, system file writes, or command execution, limiting the potential impact of data-driven instruction override.
Sanitization: No specific sanitization logic is described for the processed text.
[OBFUSCATION]: All instructions and JSON templates are provided in clear text. No Base64 encoding, hex escapes, or hidden characters were detected.

newsjack-triage