# CLAUDE.md — Claude Code Instructions for WSC-MVC

Read `AGENTS.md`, then `SPEC.md`, then `IMPLEMENTATION_PLAN.md` before changing code. AGENTS.md contains the shared operating rules for all coding agents; this file is a concise Claude-specific execution entry point, not a competing architecture.

## Primary task
Implement WSC-MVC milestone by milestone: IIS/Classic ASP host, registered VBScript WSC components, and separate HTML templates. Start with the `/hello` vertical slice. Do not start with scaffolding every speculative class.

## Planning and context discipline
- At the start of each session, determine current milestone from implementation plan and actual tests, not from filenames alone.
- Inspect changed files and prior decisions; preserve user changes.
- For a task crossing architectural boundaries, first write a brief plan referencing the exact SPEC acceptance criteria.
- Use small edits and run the closest available test after each meaningful increment.
- If tool access lacks Windows or IIS, implement only what can be responsibly checked and clearly report Windows integration as NOT RUN.
- Do not turn guesses about WSC registration, COM marshaling, WSC XML semantics, IIS rewrite, or hot reload into established facts.

## Code review checklist
Confirm: VBScript `Option Explicit`; WSC `<public>` and script procedure agreement; stable COM identity; object/scalar assignment semantics; checked error paths; request-scoped state; safe response ownership; no URL-controlled arbitrary ProgID/method; no publicly accessible source/config/templates; tests updated; docs accurate.

## Before concluding
Summarize modified files, what actually ran, proof of acceptance or blocked tests, and the single next milestone. Never say 'done' for a host-dependent milestone without IIS evidence. Propose rule improvements with evidence; do not silently edit architectural guardrails.

## Useful prompt to resume
"Read AGENTS.md, SPEC.md, IMPLEMENTATION_PLAN.md, and docs/DECISIONS.md. Identify the first unmet acceptance criterion in the current milestone. Implement its smallest testable vertical slice, run all available tests, and report PASS/FAIL/NOT RUN with commands and evidence. Do not advance milestones until its gate passes."