The Agent Skills Directory

[SAFE]: The skill implements a structured workflow for research verification without exhibiting any malicious patterns. All identified behaviors, such as file system interaction and cross-skill tool calls, are directly relevant to its stated purpose.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface because it ingests and processes external literature abstracts and candidate paper data during its analysis phases.
Ingestion points: Technical claims and paper abstracts are retrieved via external tools such as superpower-writing:research-lookup and superpower-writing:literature-review (Phase B, SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or "ignore instructions" markers when processing the retrieved text from external papers.
Capability inventory: The skill utilizes file writing to the .writing/ directory and /tmp/, and can orchestrate the superpower-writing:collaborating-with-codex tool which interacts with a CLI bridge.
Sanitization: The protocol emphasizes citation integrity (using [UNVERIFIED] tags for unconfirmed papers in references/citation-discipline.md) but does not implement text-level sanitization or escaping of the ingested research content.

novelty-gap-check