The Agent Skills Directory

[NO_CODE]: The skill consists entirely of Markdown and YAML configuration files. There are no Python scripts, JavaScript files, or shell scripts associated with this skill, eliminating the risk of direct code execution or runtime vulnerabilities.
[DATA_EXFILTRATION]: No network operations (such as curl or wget) or sensitive file path references (such as .ssh or .env) were detected. The skill only processes user-provided manuscript text for reporting purposes.
[PROMPT_INJECTION]: The instructions focus on statistical integrity and transparency. There are no attempts to override agent safety filters or bypass system instructions. The 'Red lines' section in SKILL.md acts as a safety constraint by instructing the agent not to invent data.
[EXTERNAL_DOWNLOADS]: The skill references official Nature Portfolio URLs for documentation purposes. These are well-known, trusted academic sources and do not involve executable payloads.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data in the form of manuscript sections and reviewer comments as specified in SKILL.md and README.md.
Boundary markers: No explicit boundary markers or 'ignore embedded instructions' prompts are used for the input data.
Capability inventory: The skill has no identified capabilities to execute subprocesses, write to the filesystem, or perform network requests.
Sanitization: No specific sanitization or filtering logic is present for input data.
Risk Assessment: Given the lack of dangerous capabilities, the surface for indirect prompt injection is negligible, as the agent only generates text output for the user.

nature-statistics