parallel-first.md

description	Enable a parallel-first policy for this thread and prefer delegate-parallel subagents when safe.
argument-hint	preferences (optional)

Enable parallel-first mode for this conversation thread.

PREFERENCES (optional) $ARGUMENTS

POLICY (sticky for this conversation)

1. Persist for this thread

   • Treat this message as a standing instruction for the rest of the conversation thread.
   • Apply it to future user requests unless the user explicitly disables it (e.g., "parallel-first off").

2. Parallelization gate (per request)

   • Before doing work, decide if the user request is safely parallelizable:
     • At least 2 independent tasks
     • Limited file overlap
     • Clear acceptance criteria / validation
     • Straightforward integration path
   • If not parallelizable: do NOT spawn subagents; proceed sequentially.

3. If parallelizable: use delegate-parallel workflow

   • Follow: skills/workflows/coordination/delegate-parallel/SKILL.md
   • Defaults (unless overridden by the user in plain language):
     • max_agents = 3
     • max_retries_per_task = 2
     • mode = patch-artifacts
   • Decompose into task cards, spawn subagents, require artifact-based delivery, integrate deterministically, validate, and iterate until accepted.

4. If underspecified: ask must-have questions first

   • Use: skills/workflows/conversation/ask-questions-if-underspecified/SKILL.md
   • Ask 1–5 “Need to know” questions with defaults.
   • Do not dispatch subagents until the user answers or approves assumptions.

5. Context hygiene

   • Keep main-agent chat output short and acceptance-focused.
   • Do not paste large diffs or logs into chat; write them to $AGENT_HOME/out/ artifacts and reference paths instead.

On enable, respond with:

• Confirmation that parallel-first mode is enabled for this conversation thread.
• The defaults you will use (max_agents / retries).
• How to disable it ("parallel-first off").