farmslot

Drive another model through a tmux pane with guardrails. Use when Codex should orchestrate Claude, Codex, Grok, Cursor, or another agent inside an existing tmux session, verify pane state before sending input, and keep a feedback loop around observed pane output.

Build a validation recipe from PR intent, ticket ACs, or investigation findings

Help run or install a local recipe runtime without assuming the full Farmslot framework

Author or review a Farmslot backlog spec (.backlog/specs/**) — run deterministic spec-lint, check AC↔scope traceability and code refs, then file via backlog.create with the spec attached and verify the round-trip. For agents filing work the farm will later dispatch.

Review validation recipes and evidence quality without project-specific assumptions

Communicate complex agent output as compact, reviewable Farmslot operator packets with artifact anchors and safe action requests

Succinct release-needed signal and manual release-cut workflow for Farmslot workspaces (Command Center, gateway, protocol, companion, npm packages). Run release-status first; curate Unreleased bullets; cut with reviewed proposal.

Optional authoring review for Farmslot worker prompt templates — run deterministic contract/structure lint, then critique succinctness, contradictions, checklist flow, and stop conditions. Not part of slot lifecycle; for pack maintainers editing templates/worker/*.md.

Interactive tmux-only cross-runner review orchestrator for day-to-day PR hardening. Coordinates a worker pane, a fresh same/similar-runner review pane, and an optional different-runner review pane until all blocking findings, including nits, are fixed.

Recommend the smallest useful Farmslot recipe adoption level for a project

Judge recipe quality and evidence quality for recipe-cook style workflows. Use when an agent or developer needs a focused critique of recipe structure, AC coverage, evidence fit, flake risks, and the next best improvements.

Process pending farmslot retros in bulk — extract a digest, curate learnings into LEARNINGS.md, then resolve the decisions in one pass. Use when the retro inbox has accumulated more than ~5 unresolved items and per-retro accept/dismiss in the UI is too slow, or when you want to feed several runs' signal into the self-improvement loop at once.

Diagnose local readiness for recipe authoring and execution