The Agent Skills Directory

[SAFE]: The skill performs legitimate auditing tasks as described. Its operations are restricted to reading project documentation and writing report files. No malicious behavior, obfuscation, or unauthorized access attempts were detected.
[NO_CODE]: The skill consists entirely of instructional markdown and does not include any scripts, binary files, or executable code snippets.
[DATA_EXFILTRATION]: The skill references configuration and workflow files within the user's home directory (~/.claude/gsd-core/). These paths are specific to the 'open-gsd' environment and are used to provide the necessary context for the audit. No sensitive system credentials or network exfiltration patterns were found.
[PROMPT_INJECTION]: Analysis of the indirect prompt injection surface associated with processing external project files.
Ingestion points: The skill reads AI-SPEC.md and previous execution data from the local repository.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the audit prompts.
Capability inventory: The skill has access to Bash, Write, Read, and Grep tools to analyze data and create the EVAL-REVIEW.md report.
Sanitization: No explicit content validation or escaping of the ingested documentation is described in the workflow.

gsd-eval-review