The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted design documents and performs code verification using file system tools. An attacker could potentially embed paths to sensitive files within a document to attempt unauthorized file access during the verification step.
Ingestion points: Full content of the user-provided design document (SKILL.md, Step 2).
Boundary markers: Absent.
Capability inventory: Read, Grep, and Glob operations (SKILL.md, Step 4).
Sanitization: Absent.
Note: The skill includes explicit instructions to 'Bound verification to the modules the document references' and forbids whole-tree scans, which serves as a mitigation against unauthorized file exploration.
[EXTERNAL_DOWNLOADS]: The file references/sources.md contains references to reputable external technical resources including the Fuchsia project and the Pragmatic Engineer blog. These are well-known technical documentation sources used here to establish the provenance of the review rubric.

design-review