Capture HTTP and terminal evidence as reproducible screenshot bundles
Two tools with clear, actionable descriptions and well-structured input schemas. Tool names follow verb_noun convention (capture_*). Schemas include type definitions and parameter descriptions. However, output schemas are not documented in the visible code, and error handling lacks recovery guidance. Parameter descriptions are adequate but could be more prescriptive about constraints and formats. No tool annotations (readOnlyHint/destructiveHint) despite one tool being READ_ONLY and the other IRREVERSIBLE.
Execute a non-interactive command in a host PTY. This has the current user's full host permissions. Captures merged terminal output and paginated screenshots.
Send one raw HTTP/1.1 request over TCP or TLS, preserve the received bytes, and render paginated request/response evidence screenshots.
Output schemas not documented. LLMs cannot infer what fields capture_http_evidence and capture_command_evidence return, forcing them to guess about downstream data availability.
Tool annotations missing. capture_http_evidence is marked READ_ONLY and capture_command_evidence is IRREVERSIBLE in the metadata, but no readOnlyHint or destructiveHint annotations are present in the schema to guide LLM behavior.
Error handling lacks recovery guidance. No evidence of actionable error messages that tell LLMs what to do next (e.g., 'timeout_ms must be 1-300000' is good, but no guidance on retry strategy or fallback actions).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2025-06-18+ | v2 |
Parameter descriptions lack format/constraint details. 'target' and 'command' descriptions do not specify minimum length, character restrictions, or format patterns that LLMs need to validate input.
No pagination or result-limiting guidance. If HTTP responses or terminal output can be large, no documentation of how results are truncated, paginated, or capped.