The Agent Skills Directory

[SAFE]: A thorough review of the skill's instructions and accompanying scripts found no evidence of prompt injection, data exfiltration, obfuscation, or persistence mechanisms.
[SAFE]: The skill uses local Bash scripts and artifacts (located in ~/.claude/skills/nanostack/) for telemetry and state management, which is standard behavior for its execution environment.
[SAFE]: The suggest-security.sh hook is a beneficial security utility that monitors the file diff for changes to authentication, payments, or infrastructure files and proactively suggests a security review to the user.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection because it processes untrusted code and project artifacts.
Ingestion points: Ingests changed files, plan artifacts via resolve.sh, and module briefs (diarizations) to guide the review.
Boundary markers: The instructions do not define explicit boundary markers to separate user code from agent instructions.
Capability inventory: Capabilities are restricted to local file inspection, JSON artifact generation via jq, and standard Bash operations.
Sanitization: The skill analyzes and reports on code issues but does not execute the code it reviews, limiting the risk of injection-led command execution.

review