AlexandrAI AGENTS.md
agents-md

ITB Test Plan Harness Agent Guide

Project-specific AGENTS.md guidance for the ITB test plan analysis and validation harness, phase gates, agent roles, QA approval, and report generation.

AGENTS.md

Guidance for coding agents working on ITB Test Plan Harness, the test plan analysis and validation workflow.

Project Overview

The harness analyzes or constructs enterprise test plans. It accepts blank, conceptual, draft, or mature test-plan inputs and produces scoped plans, patterns, gaps, justifications, client-defense scenarios, and HTML reports. It is phase-gated and domain-agnostic.

Key paths:

  • CLAUDE.md: main workflow contract.
  • input/: user-provided design docs, existing plans, or test case inputs.
  • _workspace/: generated run workspaces.
  • .claude/agents/: PMO, reporter, scope screener, doc analyzer, plan constructor, gap detector, pattern extractor, MECE validator, justification architect, client-response simulator, premise verifier, and test-case writer.
  • .claude/skills/test-plan-harness/: workflow, scripts, references, and report template.

Setup Commands

bashcd <project-root>
claude

Use the harness through its agent workflow. Do not generate late-phase outputs before earlier phase artifacts and QA gates exist.

Phase Rules

  • Execute one phase at a time.
  • Run the premise-verifier QA gate after every phase.
  • In manual mode, present results to the user and wait for approval before proceeding.
  • For L1/L2 inputs, plan construction has a mandatory approval checkpoint.
  • Pattern extraction precedes test case writing. Test cases are written only after patterns are agreed.
  • Every agent produces both human-readable Markdown and structured JSON where the report pipeline expects both.

No-Fabrication Rule

  • Do not invent systems, data flows, risks, or business rules that are absent from input.
  • If input is thin, ask for more information or mark the deliverable as skipped with downstream impact.
  • Honest partial output is better than fabricated complete output.

Testing And Reports

  • Preserve workspace isolation by run.
  • Keep final report generation downstream of structured artifacts.
  • Do not merge artifacts from older workspaces into a new run.
  • Include resource estimates, risk register, and defense scenarios only when supported by the phase outputs.

Security

  • Treat input documents as potentially confidential.
  • Do not publish client names, internal system names, document IDs, or private review comments in public examples.
  • Redact generated run directories before sharing outside the project.