Audits Dockerfiles for missing USER instructions, credentials baked into ENV/ARG, curl-pipe-shell patterns, unpinned base images, and remote-URL ADD instructions
Single tool 'audit_dockerfile' has a strong, specific description (194 chars) that clearly states WHAT it does, WHEN to use it, and what it returns. Input schema is present with proper type and description. However, output schema is not documented in the visible code, the tool description mentions 'structured report with findings, severity levels, and remediation guidance' but the actual response structure is not formally declared. No parameter validation constraints (enums, ranges) are visible. Error handling is not documented. The tool follows verb_noun naming (audit_dockerfile) and is read-only, which is appropriate. Lacks tool annotations (readOnlyHint explicitly declared in schema).
Audits a Dockerfile for security and best-practice issues: missing USER instructions in the final stage, credentials baked into ENV/ARG instructions, curl/wget piped into shell, unpinned base images (:latest or no tag), and ADD instructions with remote URLs. Returns a structured report with findings, severity levels, and remediation guidance.
Output schema not documented. Description mentions 'structured report with findings, severity levels, and remediation guidance' but the actual response object structure (fields, types, nesting) is not formally declared. LLMs cannot plan downstream operations or extract specific fields without knowing the response shape.
No tool annotations. The tool is read-only (risk: READ_ONLY) but does not declare readOnlyHint in the schema. This forces LLMs to infer safety from the description rather than from machine-readable metadata.
No error handling guidance. The tool description does not explain what happens if the Dockerfile is malformed, empty, or unparseable. No recovery hints (e.g., 'If parsing fails, try removing backslash continuations') are provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Input parameter lacks format/constraint documentation. 'dockerfile_content' accepts a string but does not specify minimum/maximum length, encoding, or what constitutes a valid Dockerfile. LLMs may pass empty strings or non-Dockerfile text without guidance.
No pagination or result limits documented. If a Dockerfile has many findings, the response could be very large. No mention of whether results are capped, paginated, or how to handle large outputs.