Agent Packs
Run a pipeline of specialist agents — spec, code, review, verify — supervised live from your phone or the web dashboard.
- Turns your task into testable acceptance criteria
- Implements them test-first
- It never sees the coder’s chat
- Verifies every criterion and writes the report
Describe one task and Agent Packs runs it as a sequence of specialist stages — spec, code, review, verify — instead of one long agent conversation. Each stage starts in a fresh context (the reviewer judges the code, not the coder’s reasoning) and ends in a verified handoff: a real commit — validated against the repo — the diff, and your project’s checks run for real, green or red. You supervise the whole assembly line from your phone or the web dashboard (session menu → Agent Packs).
Each specialist works in a clean conversation. The reviewer never sees the coder’s chat, so it judges the code on its merits.
A stage only passes the baton with a real commit, the diff summary, and your project’s checks actually run — green or red. Every role closes with a written handoff the next role receives verbatim.
The Reviewer records every finding as data — id, severity, file and line, resolution — in REVIEW-FINDINGS.pack.json, committed with its stage. QA verifies each finding by id, and the app shows them on the handoff card.
Watch stages advance live. When a specialist asks you something the run pauses; answer in the stage chat and Resume continues that same conversation. Retry runs the stage again fresh, told what went wrong the first time.
The run’s ledger lives in the repo. If the CLI restarts mid-stage, the pipeline comes back stalled with the reason — not lost — and Resume, Retry, Skip and Abort keep working.
Quick Pack (Coder → Reviewer) is free for tight fix-and-review loops; the Full Pack pipeline (Specifier → Coder → Reviewer → QA) is Pro.
- 1Open a connected session → ⋯ → Agent Packs
Agent Packs lives in the session menu of any connected session.
- 2Pick a pack and read each role’s brief
Quick (Coder → Reviewer) or Full (Specifier → Coder → Reviewer → QA) — each role card explains what that specialist does.
- 3Describe the task and run
One task description is all the pipeline needs — the Specifier turns it into acceptance criteria for the rest.
- 4Watch the pipeline
Stages advance live. Expand any handoff to see the commit, diff, and check results — and act when a specialist asks.
Requires a recently deployed session (latest CLI). Stages run one at a time, so a Full Pack run is a “kick off and supervise” workflow — start it, then watch the handoffs land.
FAQ
Related
Ready to try it?