/refine

name: cli-anything-web:refine
description: Refine an existing cli-anything-web CLI by recording additional traffic and expanding command coverage. Invokes the gap-analyzer skill as its first step, then implements missing endpoints.
argument-hint: <app-path> [focus-area]
allowed-tools: Bash(*), Read, Write, Edit, mcp__chrome-devtools__*

CRITICAL: Read HARNESS.md First

**Before refining, read `${CLAUDE_PLUGIN_ROOT}/HARNESS.md`.** All new commands and tests must follow the same standards as the original build. HARNESS.md is the navigational overview; the rules themselves are defined in `${CLAUDE_PLUGIN_ROOT}/skills/shared/CONVENTIONS.md`.

CLI-Anything-Web: Refine Existing Harness

Read the methodology SOP: @${CLAUDE_PLUGIN_ROOT}/HARNESS.md

Target: $1 Focus area: $2

Process

> **Skills used:** `gap-analyzer` (mandatory FIRST step), `methodology` > (pipeline), `capture` (if re-recording)

1. **Gap analysis (FIRST step — always)**: Invoke the `gap-analyzer` skill (`${CLAUDE_PLUGIN_ROOT}/skills/gap-analyzer/SKILL.md`) with `APP_PATH=<app>/agent-harness`. It diffs the captured/documented API surface against the implemented commands and client methods. Every reported gap must cite its evidence: the endpoint entry in `<APP>.md` and/or `traffic-capture/traffic-analysis.json` (with `raw-traffic.json` hit counts for priority). If a focus area is specified, filter the report to that domain. 2. **Present gap report**: Show the user the gap analysis results and confirm which gaps to address before proceeding with any recording or implementation 3. **Record new traffic**: Use playwright-cli (see HARNESS.md Phase 1) or chrome-devtools-mcp fallback 4. **Analyze new endpoints**: Add to API map in `<APP>.md` 5. **Implement new commands**: Add to existing command groups or create new ones 6. **Update REPL help**: Edit `_print_repl_help()` in `<app>_cli.py` to reflect every new command and option added (CONVENTIONS.md §REPL Rules help-sync rule). Users typing `help` in the REPL must see all available filters and options. 7. **Update tests**: Add unit + E2E tests for new commands 8. **Run full test suite**: Ensure no regressions 9. **Update TEST.md**: Document new coverage

Rules

• NEVER break existing commands
• NEVER change existing command signatures
• ADD new commands and options only
• Run full test suite after changes
• Update `<APP>.md` with new endpoints
• **Always update `_print_repl_help()` to match the actual command surface**

Success Criteria

• All identified gaps have been addressed or explicitly deferred
• No existing commands are broken or have changed signatures
• New commands follow CONVENTIONS.md (exceptions, --json envelope, REPL rules)
• Full test suite passes (including new tests)
• TEST.md updated with new test coverage
• `<APP>.md` updated with new endpoints
• **REPL `help` output reflects all new commands and key options**

Notes

• Refine is **incremental** — it only adds, never removes commands
• Always **present the gap report** before implementing changes
• Run the full test suite after changes to ensure no regressions