Kit — Agent Context

Product (2.x): Kit is the control room for parallel agent work.
Dispatch many agents. Watch them in one place. Nothing ships unproven.

Not the product: skill marketplace, registry social, Ink TUI (archived).
Skills remain local substrate — what you dispatch, not the point.

Full PRD: docs/dev/PRD-1.0.md
Who builds what: docs/dev/BUILD-ASSIGNMENT.md
Active surface tasks: tasks/plan.md, tasks/todo.md

Multi-agent roles (CEO / Power / Luna)

See docs/dev/ORCHESTRATION.md.

Luna storm: parallel explore/implement shards; Power synthesizes; CEO decides/merges.
Workflow sketch: .grok/workflows/kit-luna-storm.rhai (local; .grok may be gitignored).
Quality ladder (long-running): docs/dev/WORKFLOW-QUALITY.md + workflow kit-quality-cycle
(audit → skeptic → plan → implement → prove). Git copy: docs/dev/workflows/kit-quality-cycle.rhai.
Craft ladder (design | product first): workflow kit-craft-cycle before TUI/motion slices. Git copy: docs/dev/workflows/kit-craft-cycle.rhai.
CEO brief template: docs/dev/tasks/CEO-BRIEF-next.md.

Claude-grade coding practice (always): Karpathy surgical diffs and tests-as-success; addyosmani spec → plan → TDD → five-axis review; public contracts change only with a CHANGELOG entry and a test; motion is live work (RUNNING/GATING spinner) plus the resting fox's tail wag in the empty Control Room (kit-tui/src/fox.rs), all on the one clock.

Engineering skills (always)

This repo installs addyosmani/agent-skills under .agents/skills/ (24 skills).

Before non-trivial work, load using-agent-skills and route:

Core operating rules (from using-agent-skills)

1. Surface assumptions before implementing.
2. Stop on confusion — do not guess contracts.
3. Push back on approaches that break kill criteria or crate boundaries.
4. Simplicity first — boring over clever.
5. Scope discipline — only touch the task's crate/files.
6. Verify with evidence — cargo test / clippy / snapshots; "looks right" is never done.

Who may edit what

Any agent may edit any crate (kickoff spec K15, docs/superpowers/specs/2026-10-26-kit-kickoff.md). Two rules guard the public surface:

• A change to a kit-core public type, a --json shape, KIT.toml / kit.toml, kit.lock or the receipt format needs a CHANGELOG.md entry and a test that pins the new shape.
• The bar for every change is docs/QUALITY.md: tests at the right layer, a snapshot for every screen change, CI green on all three operating systems.

TUI non-negotiables:

• One clock: only AppEvent::AnimationTick (no other timers)
• Hit-testing from the same ratatui rects the renderer drew (F6)
• Navigation mutates Screen in the reducer; engine work is Action
• Reduced motion: NO_COLOR, KIT_MOTION=off

Commands

# Run the product
cargo run -p kitctl -- --demo    # Control Room + PRD fixture
cargo run -p kitctl              # empty Control Room
cargo run -p kitctl -- doctor

# Verify
cargo test -p kitctl-tui
cargo clippy -p kitctl -p kitctl-tui --all-targets -- -D warnings
cargo fmt --all --check
cargo test --workspace

Living status: docs/dev/CURRENT.md

Rust workspace: crates/kit-{core,agents,gate,tui,cli}
The Kit 0.1 TypeScript tree was removed from trunk in A0.0; it is kept at tag legacy-0.1-final.

Milestone map (surface focus)

Visual / interaction language

• Control Room is the default (kit → table of runs)
• One interaction voice across screens (same footer grammar, Esc back)
• Failures surface first error line in the table; detail holds full gate log
• Attach: Esc detaches without kill; q disabled while attached