The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses the local project's source code to perform fact-checking (R0 step). This is a legitimate function for a technical review tool and does not involve sending data to external servers.
[COMMAND_EXECUTION]: The skill utilizes standard system tools like grep and find via the agent's shell capabilities to locate terms and source code definitions within the local environment.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to process user-provided technical articles, it is susceptible to indirect prompt injection if the content being reviewed contains malicious instructions. However, the skill provides specific structured review steps which helps mitigate accidental execution of embedded instructions.
Ingestion points: User-provided article files (processed via Read and grep).
Boundary markers: Not explicitly defined in the provided instruction files.
Capability inventory: Read, Write, Edit, and Skill (Claude Code internal mechanism).
Sanitization: No specific input sanitization for the article content is implemented.
[METADATA]: The skill persists user preferences and review history to a local directory (.article-skills/) within the project root. This is used for context retention across review sessions.

article-review