implement-batchlisted
Install: claude install-skill ghozimahdi/gm-aiagent-plugins
Use the current user request as this skill's input. In Claude Code invoke it as
`/ufil:implement-batch`; in Codex invoke it as `$ufil:implement-batch`. Resolve
`UFIL_ROOT` to the plugin root containing this skill; Claude Code may provide
`CLAUDE_PLUGIN_ROOT`, while Codex can resolve it from the installed skill path.
Use the client's native subagent capability; do not hardcode one client's tool
call schema.
Implement multiple features in parallel using auto-scaled agents.
Arguments: <requested arguments> (space-separated feature names or ticket IDs)
## Step -1: Initialize Serena (auto-onboard if needed)
Run in order at session start, BEFORE spawning any agents:
1. Call `mcp__serena__initial_instructions`.
2. Call `mcp__serena__check_onboarding_performed`.
3. **If not onboarded:**
- Probe for code: `find . -maxdepth 3 -type f -name '*.dart' -not -path '*/.*' | head -1`
- If a Dart file is found → call `mcp__serena__onboarding` (one-time per project).
- If empty → SKIP onboarding; tell the user: "Serena onboarding skipped — no Dart code detected. Will auto-run on the next implement-batch workflow once code exists." Subagents will fall back to built-in search and read tools.
4. **If already onboarded:** proceed.
Onboarding must happen BEFORE parallel agents spawn so they inherit a ready symbol index instead of racing to onboard concurrently. To force re-onboarding, invoke the `serena-refresh` skill.
## Project Type Detection (MUST DO FIRST)
- **Modular**: `packag