Self-driving dev harness for MCP-compatible agents. Verifiable, budgeted, persistent dev harness with inference from host agent via MCP sampling/createMessage.
Radiant Harness exposes 4 tools with basic schemas and descriptions. All tools have names starting with action verbs (read_, write_, search_, run_) and descriptions present. However, descriptions are terse (10-60 chars), parameter descriptions lack detail on constraints/formats, and output schemas are not documented. The write_file and run_gate tools perform destructive operations but lack confirmation/dry-run patterns. No tool annotations (readOnlyHint/destructiveHint) are present. Error handling guidance is absent, tools do not indicate what the LLM should do on failure.
Read the contents of a file at the given path. Path must be inside the project directory.
Run a quality gate command (go test, go vet, etc.). Returns stdout/stderr and exit code.
Search the project for a regex pattern. Returns matching lines with file:line:column format.
Write content to a file at the given path. Creates parent directories as needed. Path must be inside the project directory.
Output schemas not documented. LLMs cannot infer what fields to expect from tool responses, forcing them to guess at downstream tool chaining and data extraction.
Destructive tools (write_file, run_gate) lack confirmation/dry-run patterns. Agents can accidentally overwrite files or run unintended commands without a safety gate.
Parameter descriptions lack constraint details. 'path' parameter does not specify format, length limits, or allowed characters. 'pattern' in search_code does not document Go regexp syntax constraints or error cases.
No error recovery guidance. Tools do not indicate whether failures are retryable, user-fixable, or fatal. LLMs receive no actionable next steps on failure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Tool annotations missing. No readOnlyHint, destructiveHint, or idempotentHint attributes present. LLMs cannot determine which tools are safe to retry or have side effects.