Cavekit v2

Compressed spec-driven development for Claude Code.

One file. Three commands. Zero sub-agents.

WHY

Plan-then-execute forgets. SDD remembers — but most SDD frameworks bury that value under agent swarms, dashboards, and ceremony that wastes tokens.

Cavekit v2 keeps only what earns its place:

• Durable spec — SPEC.md at repo root survives context resets.
• Caveman encoding — ~75% fewer tokens than prose. Symbols, fragments, pipe tables for repeating records.
• Backprop reflex — every test failure becomes a §B entry; classes of bug become §V invariants the spec will never forget.

That's the whole pitch.

COMMANDS

cmd	job
/ck:spec	create / amend / backprop SPEC.md. Sole mutator.
/ck:build	native plan → execute against spec. Auto-backprops on failure.
/ck:check	read-only drift report. Lists §V / §I / §T violations, suggests remedy.

FORMAT

See FORMAT.md. Fixed sections: §G goal, §C constraints, §I interfaces, §V invariants, §T tasks (pipe table), §B bugs (pipe table). Caveman prose plus two pipe tables. Nothing else.

FILES

FORMAT.md          spec schema + caveman encoding rules
commands/          three markdown command prompts
skills/caveman     encoding skill
skills/backprop    bug → spec protocol (six steps)

NON-GOALS

• No sub-agents. Main Claude does the work.
• No dashboards. cat SPEC.md is the dashboard.
• No parallel workers. One thread, one spec, one diff.
• No JSON / YAML spec bodies. Markdown + pipe tables.
• No hooks, no orchestration binaries, no TypeScript helpers.

INSTALL

Add as a Claude Code plugin via marketplace, or clone into ~/.claude/plugins/. Commands become available as /ck:spec, /ck:build, /ck:check.

HISTORY

v1 (main branch) had 16 commands, 12 agents, 21 skills, a Go binary, shell hooks, and a dashboard. v2 replaces all of it with this.

MIT.