Home / All Skills / Workflow & Core Tools / caveman-help
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · juliusbrussee/caveman

caveman-help

Quick-reference card for caveman modes, skills and commands. Trigger: /caveman-help or "caveman help".

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit
Author juliusbrusseeGitHub juliusbrussee/caveman ↗Stars 105691License MIT（仓库根 LICENSE + 根内 scope note；LICENSING.md 表格明确 `skills/` | MIT | Existing Caveman skill stays MIT and untouched.）。注意本仓库为拆分授权：engine/ proxy/ rewriter/ browse/ mcp/ shrink/ 等 Engine-linked 目录为 BSL-1.1（LICENSE.BSL）；GitHub API 的 license.spdx_id 返回 NOASSERTION，即 GitHub 无法判定单一 SPDX。Commit 15581d1400
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

This is a purely display-oriented instruction card: the SKILL.md body is a static cheat sheet (6 modes + 4 skills + deactivation methods + language rules + default mode configuration), containing no scripts, tool calls, or file write instructions.

skills/caveman-help/SKILL.md
Display this reference card when invoked. One-shot — do NOT change mode, write flag files, or persist anything. Output in caveman style.
What it is doing: narrowing the skill's execution semantics down to "output the contents of this file in caveman style". Everything else consists of human-readable tables (Modes / Skills / Deactivate / Language / Configure Default Mode), with nowhere requiring the agent to run commands, read files, or alter state.

"One-off, no side effects" is explicitly stated in two places: within the SKILL.md body itself, and in the repository's built-in skill behavior comparison table (under the output contract and side effects columns).

docs/technical/skills-hooks-and-plugins.md
| `caveman-help` | One-shot command card | Does not change active mode |
Comparing it against others in the same table, caveman-stats (reads local session logs + writes a status suffix) and caveman-compress (reads and writes selected local files): caveman-help's "side effects" column is the cleanest of all, declaring only that it does not alter the active mode.

The "default mode" configuration taught by the card is an actual mechanism rather than documentation rhetoric: the host's hook code features a four-tier resolution chain where env vars have the highest priority, falling back to 'full'.

src/hooks/caveman-config.js
// Resolution order for default mode: // 1. CAVEMAN_DEFAULT_MODE environment variable // 2. Repo-local config (checked-in, per-project default):
The card only covers the first two levels: "env > user config file > full". However, the code actually resolves: env > repo-level .caveman/config.json or .caveman.json (traversed upward from cwd) > user config ($XDG_CONFIG_HOME or ~/.config or %APPDATA%) > 'full'. See verification.second_pass for details.

Mode semantics are implemented by the primary caveman skill in the same repository; this card serves merely as its index: modes are "sticky" until modified by the user or the session ends.

skills/caveman-help/SKILL.md
Mode stick until changed or session end.
Consistent with the Persistence section of skills/caveman/SKILL.md ('Default style for this whole session, every response, until user say "stop caveman" or "normal mode"'). The card is responsible for informing the user of the available tiers (lite/full/ultra + three wenyan variants), while the primary skill handles execution.

The card intentionally preserves the user's language: it compresses style, not language; technical terms, code, commands, and error strings are retained verbatim.

skills/caveman-help/SKILL.md
Keep user's language by default. User write Portuguese → reply Portuguese caveman. Compress the style, not the language. Technical terms, code, commands, commit types, and exact error strings stay verbatim unless user ask for translation.
Consistent with the primary skill (which additionally forbids adding words merely to "sound like caveman"). This rule is crucial for usability in multilingual teams: terminology is not translated, allowing outputs to be directly consumed by tools or colleagues.

The trigger surface consists of an explicit user invocation (/caveman-help) plus a natural language alias, both documented in SKILL.md description and the human-facing README.

skills/caveman-help/README.md
Also triggers on "caveman help", "what caveman commands", "how do I use caveman".
The SKILL.md frontmatter specifies 'Trigger: /caveman-help or "caveman help"', while the README expands this to three conversational trigger phrases.

Distribution path: this skill is not included in the Claude Code plugin mirror (plugins/caveman/skills/ only contains caveman / cavecrew / caveman-stats / caveman-compress); instead, it reaches various hosts via standalone hook+skill install paths, `npx skills add`, and CLI skill directory manifests for hermes/opencode.

bin/install.js
const HERMES_SKILL_DIRS = ['caveman', 'caveman-commit', 'caveman-review', 'caveman-help', 'caveman-stats', 'caveman-compress', 'cavecrew'];
CLAUDE.md explicitly details this division of labor ('Skills not in this table (`caveman-commit`, `caveman-review`, `caveman-help`, `caveman-stats`) are not mirrored into the Claude Code plugin distribution by CI.'); preserved_skill_ids in skills/registry.json also includes caveman-help.

2Core Capabilities

01Single-screen quick reference: 6 mode tiers and their trigger phrases (lite / full / ultra / wenyan-lite / wenyan / wenyan-ultra)
02Quick lookup for sibling skills and their output contracts (caveman-commit / caveman-review / caveman-compress / caveman-help itself)
03Provides deactivation and recovery phrases to prevent users from being trapped in a "sticky" mode
04Two configuration methods for the default mode (CAVEMAN_DEFAULT_MODE environment variable, user config file defaultMode field)
05Using "off" to disable automatic activation at session launch while retaining manual /caveman invocation
06Language preservation rules: compress style without switching languages; technical terms, code, commands, and error strings remain verbatim
07Does not alter session state (no mode switching, no flag file creation, no disk writes); strictly read-only Q&A

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Can be auto-loaded
For Administrators
Register source and version
Evidence & Triggers (File Paths & Excerpts)
  • SKILL.md: Pure prompt instructions and rule text; no script execution or external impact.
Audit Notes: Purely instruction-based, zero scripts, zero network access, zero credentials, and zero writes, independently verified in the repository's behavior comparison table as not altering session state. The only consideration is that the card guides users to configure environment variables/config files (which is usage guidance, not an action performed by the skill itself).

5Second-Pass Verification

  • [ok] 卡片属实：一次性展示、不改模式、不落盘 — SKILL.md 正文首段与 docs/technical/skills-hooks-and-plugins.md 的副作用列两处独立印证；目录内无任何可执行脚本。
  • [discrepancy] 默认模式配置链（卡片声称 env > config file > full） — 卡片写 'Resolution: env var > config file > `full`.'，实际代码是四级且含仓库内配置：① CAVEMAN_DEFAULT_MODE → ② 仓库内 .caveman/config.json 或 .caveman.json（从 process.cwd() 向上最多 64 级查找，拒绝符号链接）→ ③ 用户 config（$XDG_CONFIG_HOME/caveman/config.json，或 ~/.config，或 Windows %APPDATA%）→ ④ 'full'（src/hooks/caveman-config.js:15-26 与 findRepoConfigPath 注释）。卡片漏了「仓库内配置」这一层，团队若用 .caveman.json 固定项目默认档，卡片会让人误判优先级。非安全影响，属文档简化。
  • [ok] wenyan 档位命名一致性（卡片 vs 主 skill vs 代码） — 卡片列 wenyan-lite/wenyan/wenyan-ultra（并称 'Wenyan-Full | /caveman wenyan'）；代码 VALID_MODES 为 ['off','lite','full','ultra','wenyan-lite','wenyan','wenyan-full','wenyan-ultra',...]，卡片触发词全部落在集合内；主 skill SKILL.md 的切换串亦为 '/caveman lite|full|ultra|wenyan-lite|wenyan-full|wenyan-ultra|off'。
  • [ok] 外部依赖为零 — 对两个文件扫描 URL/CLI/包/网络 token 零命中；卡片末行的 'Full docs: https://github.com/JuliusBrussee/caveman' 是给人点的文档链接，不是运行时调用点。
  • [ok] 无脚本执行 / 无凭证读取 / 无文件写入 — skills/caveman-help 下仅两个 markdown；仓库级 shell/node 代码（bin/install.js、src/hooks/*）属于 caveman CLI 本体，不由本 skill 触发，除 SKILL.md 文字外无任何指向它们的调用指令。
  • [ok] 声明 'Saves ~46% input tokens'（指向 caveman-compress）是否有出处 — skills/caveman-compress/README.md 基准表平均值为 46%（Average 898 → 481，46%），卡片引用与邻居 skill 自述一致。
  • [ok] pin commit 与上游状态 — 本地 HEAD == origin/main == 15581d14007fd01fb3f132016741962f34936ca2（2026-09-07T08:19:43+00:00, 'chore: sync SKILL.md copies [skip ci]'）。gh 显示仓库 pushed_at 晚于该时间，故本报告结论仅对该 pin commit 有效，不跨版本推广。

6Verdict & Scenarios

  • Zero-cost, zero-risk discoverability entry point: a single markdown card allows users to view all modes and commands within three seconds, with no scripts, no network requests, and no writes.
  • Decoupled from the execution layer: the card acts purely as an index while actual mode behaviors reside in the primary caveman skill within the same repository; modifying the card carries low blast radius as it does not affect behavior.
  • Multilingual-friendly: explicitly defines "compress style, not language" and retains terms/code/error strings verbatim, ensuring cross-lingual teams do not lose technical precision due to compression.
  • Independently documented and verifiable side-effect surface: the repository's own behavior comparison table labels it as "Does not change mode", allowing users to select tools according to their side-effect profile.
    • Best for: Suitable for any team or individual using the caveman ecosystem (modes compressing output tokens): as a quick-reference card when forgetting slash commands, explaining available tiers and sibling commands to new members, or showing others how to fix the default mode or disable auto-activation. Also serves as an ideal "zero-asset skill" reference for skill authors.
    Not for: Not suitable for scenarios where it is expected to "switch modes"—it is an informational card, not a toggle switch (SKILL.md explicitly states 'Does not change mode'); not suitable for those requiring exact details of the configuration resolution chain (read src/hooks/caveman-config.js directly, as the card omits the in-repo configuration tier); not suitable for environments lacking the primary caveman skill, as the described tier behaviors cannot be executed.
    Install Agent-Ready
    ① Verified Mirror 2026-09-15
    Option A · Direct DownloadDownload caveman-help.tar.gz
    sha256: 742f9e1fe5d8d598…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 15581d1400.