The Agent Skills Directory

[SAFE]: The skill implements a well-defined, multi-phase pipeline (Discover, Validate, Analyze, Report) dedicated to image auditing and accessibility verification.
[COMMAND_EXECUTION]: The skill utilizes Bash for standard local file operations such as retrieving file metadata (ls -la, stat) and optionally uses identify (from ImageMagick) for image dimension analysis. These are legitimate, non-malicious uses of command-line tools within the scope of the skill's purpose.
[PROMPT_INJECTION]: Instructions are focused and directive, with no evidence of attempts to override safety guidelines, bypass filters, or extract system prompts.
[DATA_EXFILTRATION]: No network operations, external requests, or credential-accessing patterns were identified. All analysis is performed on local workspace files.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it reads and parses external content files (Markdown, HTML) for image references. However, the risk is mitigated by the use of specific regex patterns for extraction and the non-executable nature of the subsequent processing phases.
Ingestion points: Content files (Markdown, HTML) searched via Grep.
Boundary markers: None explicitly defined for untrusted data beyond regex patterns.
Capability inventory: Bash, Write, Edit, Grep, Glob.
Sanitization: No specific sanitization or escaping of extracted content is mentioned.
[SAFE]: The skill includes explicit safety instructions in Phase 4, mandating that the process remains read-only and requires explicit user consent for any optimization or file modification tasks.

image-auditor