Skip to content
Back to search

cavekit

Frozen — compressed spec-driven development plugin for Claude Code. Still works; active development moved to JuliusBrussee/caveman.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/juliusbrussee/cavekit with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/juliusbrussee/cavekit'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/juliusbrussee--cavekit/badge.svg)](/listings/juliusbrussee--cavekit)

HTML

<a href="/listings/juliusbrussee--cavekit"><img src="/api/listings/juliusbrussee--cavekit/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

One line, via the `skills` CLI:

npx skills add JuliusBrussee/cavekit

Installs nine skills into `~/.claude/skills/`: `spec`, `build`, `check` (the loop), `grill`, `research`, `review`, `deepen` (reach-for), plus `caveman` and `backprop` (the utilities). Claude activates each when its trigger context matches — e.g. "write a spec for…" invokes `spec`, a fuzzy idea invokes `grill`, a risky change before build invokes `review`. Claude Code picks them up on next launch.

Or via the Claude Code marketplace (also adds the `/ck:spec`, `/ck:build`, `/ck:check`, `/ck:grill`, `/ck:research`, `/ck:review`, `/ck:deepen` slash commands):

/plugin marketplace add juliusbrussee/cavekit
/plugin install ck@cavekit

Or clone directly:

git clone https://github.com/juliusbrussee/cavekit.git ~/.claude/plugins/cavekit

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 14 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. {
  2. "$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
  3. "name": "cavekit-marketplace",
  4. "description": "Cavekit — compressed spec-driven development for Claude Code",
  5. "owner": { "name": "Julius Brussee" },
  6. "plugins": [
  7. {
  8. "name": "ck",
  9. "description": "Full SDD loop (grill → spec → research → review → build) over one SPEC.md, caveman encoding, bug-to-spec backprop",
  10. "source": "./",
  11. "category": "development"
  12. }
  13. ]
  14. }
README.md · README · 192 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. > **Status: frozen (August 2026).** cavekit is no longer in active development.
  2. > Everything below still installs and works, but expect no new features or
  3. > fixes. Active development of the family lives in
  4. > [caveman](https://github.com/JuliusBrussee/caveman) and
  5. > [caveman-browse](https://github.com/JuliusBrussee/caveman-browse).
  6. <h1 align="center">cavekit</h1>
  7. <p align="center">
  8. <strong>compressed spec-driven development for claude code</strong><br/>
  9. <sub>one file · one loop · zero sub-agents</sub>
  10. </p>
  11. ---
  12. ## what this is
  13. Plan-then-execute forgets. SDD remembers — but most SDD frameworks bury
  14. that value under agent swarms, dashboards, and ceremony that costs more
  15. tokens than it saves.
  16. Cavekit is the simplest full loop: **grill → spec → research → review →
  17. build**, over one `SPEC.md` file, no sub-agents. Three commands you run
  18. every time; four more you reach for only when the change earns it.
  19. The spine is three properties that earn their tokens:
  20. - **durable spec** — `SPEC.md` at repo root survives context resets. It is
  21. the agent's long-term memory: lose the window, reload the spec, keep going.
  22. - **caveman encoding** — fewer tokens than prose. Symbols, fragments,
  23. pipe tables. That is the whole point.
  24. - **backprop reflex** — every test failure becomes a `§B` entry; classes
  25. of bug become `§V` invariants the spec never forgets.
  26. And one rule that keeps it from bloating into the frameworks it replaces:
  27. **right-size**. A one-line fix is just `/build`. The full chain is for
  28. genuinely uncertain or high-blast-radius work — never for a typo.
  29. ## commands
  30. **the loop** — run these every time:
  31. | cmd | job |
  32. |---|---|
  33. | `/ck:spec` | create / amend / backprop `SPEC.md`. Sole mutator. |
  34. | `/ck:build` | native plan → execute against spec. Names which test proves each `§V`. Auto-backprops on failure. |
  35. | `/ck:check` | read-only drift report. Lists §V / §I / §T violations. The drift detector. |
  36. **reach for these** — only when the change earns the ceremony:
  37. | cmd | job |
  38. |---|---|
  39. | `/ck:grill` | interrogate a fuzzy idea into a sharp `§G`/`§C`, one question at a time, before you spec. |
  40. | `/ck:research` | gather external knowledge into `§R` so build grounds in facts, not hallucinations. Every finding cites a source. |
  41. | `/ck:review` | adversarial senior review of the spec *before* build. Refutes, hardens `§V`, ends in a go/no-go gate. |
  42. | `/ck:deepen` | spare-budget design pass — make one shallow module deep. Behavior held, tests green before & after. |
  43. ## install
  44. One line, via the `skills` CLI:
  45. ```bash
  46. npx skills add JuliusBrussee/cavekit
  47. ```
  48. Installs nine skills into `~/.claude/skills/`: `spec`, `build`, `check`
  49. (the loop), `grill`, `research`, `review`, `deepen` (reach-for), plus
  50. `caveman` and `backprop` (the utilities). Claude activates each when its
  51. trigger context matches — e.g. "write a spec for…" invokes `spec`, a fuzzy
  52. idea invokes `grill`, a risky change before build invokes `review`. Claude
  53. Code picks them up on next launch.
  54. Or via the Claude Code marketplace (also adds the `/ck:spec`, `/ck:build`,
  55. `/ck:check`, `/ck:grill`, `/ck:research`, `/ck:review`, `/ck:deepen` slash
  56. commands):
  57. ```bash
  58. /plugin marketplace add juliusbrussee/cavekit
  59. /plugin install ck@cavekit
  60. ```
  61. Or clone directly:
  62. ```bash
  63. git clone https://github.com/juliusbrussee/cavekit.git ~/.claude/plugins/cavekit
  64. ```
  65. ## format
  66. See [`FORMAT.md`](./FORMAT.md). Sections: §G goal, §C constraints, §I
  67. interfaces, §R research (optional, pipe table), §V invariants, §T tasks
  68. (pipe table), §B bugs (pipe table). Each verb owns specific sections —
  69. no verb rewrites a section it does not own.
  70. ## files
  71. ```
  72. FORMAT.md spec schema + caveman encoding + sectioned ownership
  73. commands/ seven thin slash-command entry points → the skills (loop + reach-for)
  74. skills/spec spec mutator — sole writer
  75. skills/build plan-execute, verification contract
  76. skills/check drift report
  77. skills/grill sharpen a fuzzy idea → §G/§C before spec
  78. skills/research external knowledge → §R, every finding sourced
  79. skills/review adversarial senior review of the spec → hardens §V
  80. skills/deepen spare-budget design pass — make one module deep
  81. skills/caveman encoding utility
  82. skills/backprop bug → spec protocol (six steps)
  83. ```
  84. ## non-goals
  85. - no sub-agents. Main Claude does the work.
  86. - no dashboards. `cat SPEC.md` is the dashboard.
  87. - no parallel workers. One thread, one spec, one diff.
  88. - no JSON / YAML spec bodies. Markdown + pipe tables.
  89. - no hooks, no orchestration binaries, no TypeScript helpers.
  90. ---
  91. ## older cavekit (the Hunt lifecycle, v3.1.0 and earlier)
  92. The previous generation is **not deprecated** — it is frozen at tag
  93. [`v3.1.0`](https://github.com/juliusbrussee/cavekit/tree/v3.1.0) and
  94. remains a fully working plugin.
  95. **What it is**:
  96. > Spec-driven AI development with an autonomous execution loop. Four-command
  97. > Hunt lifecycle (`/ck:sketch` → `/ck:map` → `/ck:make` → `/ck:check`),
  98. > plus `/ck:ship`, `/ck:review`, `/ck:revise`, `/ck:status`, `/ck:design`,
  99. > `/ck:research`, `/ck:init`, `/ck:config`, `/ck:resume`, `/ck:help` — 16
  100. > slash commands total. 12 named sub-agents. Per-task token budgets,
  101. > stop-hook state machine, model-tier routing, auto-backpropagation from
  102. > test failures, tool-result caching, Codex peer review, Karpathy
  103. > behavioral guardrails, caveman token compression, knowledge-graph
  104. > integration, and design-system enforcement. Parallel wave execution and
  105. > team mode.
  106. **Pick v3.1.0** if you want the full autonomous loop, parallel agents,
  107. peer review, or design-system workflow. **Pick v4** if you want the
  108. distilled loop — one spec, no orchestration, right-sized ceremony.
  109. ### install the older version
  110. Marketplace:
  111. ```bash
  112. /plugin marketplace add juliusbrussee/cavekit@v3.1.0
  113. /plugin install ck@cavekit
  114. ```
  115. Git:
  116. ```bash
  117. git clone -b v3.1.0 https://github.com/juliusbrussee/cavekit.git
  118. ```
  119. Full docs live at the tag — `git checkout v3.1.0` and read the README
  120. there for command reference, skill catalog, and the Hunt lifecycle guide.
  121. ### choosing, or moving
  122. See [`UPGRADE.md`](./UPGRADE.md). Honest framing:
  123. - Stay on v3.1.0 if your project has active `context/kits/` investment.
  124. - Move to v4 if you want fewer moving parts and smaller token bills.
  125. - It is a **two-way door** — `SPEC.md` is plain markdown; nothing traps
  126. you in either direction.
  127. ## ecosystem
  128. Cavekit is one rock in the caveman family:
  129. | repo | what | status |
  130. |---|---|---|
  131. | [caveman](https://github.com/JuliusBrussee/caveman) | output compression skill + engine — *why use many token when few do trick* | live |
  132. | [caveman-browse](https://github.com/JuliusBrussee/caveman-browse) | token-efficient browser automation | live |
  133. | [cavegemma](https://github.com/JuliusBrussee/cavegemma) | Gemma 4 31B fine-tuned on caveman pairs — *why prompt every turn when weight remember* | labs |
  134. | [cavemem](https://github.com/JuliusBrussee/cavemem) | cross-agent persistent memory — *why agent forget when agent can remember* | frozen |
  135. | **cavekit** *(you here)* | spec-driven build loop — *why agent guess when agent can know* | frozen |
  136. ## philosophy
  137. > The spec is the only artifact that earns its tokens. Everything else
  138. > that costs tokens must either save more tokens later, or the user's
  139. > attention, or it gets cut.
  140. See [`CHANGELOG.md`](./CHANGELOG.md) for the full v3 → v4 break.
  141. ## license
  142. MIT.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.