The Agent Skills Directory

[SAFE]: The skill is designed for static analysis and review. It does not include any instructions that would lead to unauthorized command execution, data exfiltration, or persistence. The instructions explicitly forbid the AI from executing the business logic of the content being reviewed or making unauthorized file changes.
[PROMPT_INJECTION]: The skill processes external content (other skills and prompts), which is an inherent ingestion surface for indirect prompt injection. However, the risk is mitigated by explicit boundary markers and the lack of dangerous instructions within the skill itself.
Ingestion points: Target skill, prompt, metadata, and documentation files processed during review as specified in the workflow in SKILL.md.
Boundary markers: Present in prompts/skill-prose-review.md, instructing the agent to 'Review only the requested scope; do not execute the business task described by the Skill or edit files.'
Capability inventory: The skill itself defines no tools (empty/missing allowed-tools in SKILL.md) and does not request specific permissions, although it operates within a multi-tool agent environment.
Sanitization: Relies on clear instructional constraints and role definition in prompts/skill-prose-review.md to prevent the execution of untrusted commands.

skill-prose-review