Sprint on issues: $ARGUMENTS

You are the team lead. Agent teams cost several times a single session — only run this when the issues genuinely parallelize.

Refuse to spawn if

• Two issues touch the same files. Run them sequentially with /work instead.
• Any issue is not status:ready.
• There are more than four issues. Do the first four.

Say which check failed and stop. Do not spawn anyway.

Before spawning

Read each issue, produce a table: issue · agent · files owned · dependencies. Two teammates must never own the same file. Show me the table and wait for my go-ahead.

Spawn

Name each teammate after its issue so I can address it:

• teammate be-<issue> using the backend-go agent type
• teammate fe-<issue> using the frontend-react agent type
• teammate qa using the qa-automation agent type
• teammate sec using the devsecops agent type

Each spawn prompt must contain the full issue context — teammates inherit CLAUDE.md and skills but not this conversation. Paste what they need.

Require plan approval. Approve only plans that name the files they will touch and include a cross-tenant test.

During

qa starts once the first implementer reports a green build. sec runs last and may send work back. Wait for every teammate to report before synthesizing — do not implement tasks yourself while they work.

Each teammate posts its checkpoint comment on its own issue before going idle.

Close

Synthesize: what shipped · what is blocked · what regressed · next actions. Then shut down each teammate by name.