New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Caveman Search

by ragnar-pwninskjold in ragnar-pwninskjold/tech-snacks

A skill your agent uses when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something.

MITAuto-check passedProductivity & Automation

Install Caveman Search

skills CLI
$ npx skills add ragnar-pwninskjold/tech-snacks --skill caveman-search -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install ragnar-pwninskjold/tech-snacks caveman-search --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/ragnar-pwninskjold/tech-snacks.git skills-src && mkdir -p .claude/skills && cp -r skills-src/plugins/tech-snacks/skills/caveman-search .claude/skills/caveman-search && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
caveman-search
GitHub stars
137
Token cost
~1.2k tokens
SKILL.md length
600 words
Files
3 (incl. references)
Skills in repo
7
Repo updated
First seen
Licence
MIT

At a glance

A skill your agent uses when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something.

  • Works in 6 steps: Detect mode → Escalation gate (local-first) → Parallel fan-out → …
  • The user wants a terse
  • SKILL.md covers What this does, Flags, Step 1 — Detect mode and Step 2 — Escalation gate…, plus 5 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Caveman Search is an agent skill from ragnar-pwninskjold/tech-snacks. Use when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something. Triggers on "caveman search", "caveman research", "/caveman-search", and on tooling/architecture questions where the user wants the direct answer, not a survey. Runs decomposed research (local repo + live web), verifies, and renders in a compressed caveman voice that leads with the load-bearing truth.

Its SKILL.md is about 1.2k tokens, which your agent loads only when the skill is triggered. The skill folder holds 3 other files, including reference files (for example `references/output-format.md` and `references/voice-spec.md`).

It sits in Productivity & Automation, covering Web search. The repository describes itself as: A skill library for vibe coders that make cool stuff. The licence is MIT.

When your agent uses it

  • The user wants a terse
  • Research-backed answer to a what tech/tools/stack do I need for X question
  • Asks how to approach building something
  • Caveman research

Example prompts

  • “what tech/tools/stack do I need for X”
  • “caveman search”
  • “caveman research”
  • “/caveman-search”

Workflow steps

6 steps, taken from the step headings in SKILL.md.

  1. Detect mode
  2. Escalation gate (local-first)
  3. Parallel fan-out
  4. Verify pass (only if --verify)
  5. Render
  6. Output

What it can do on your machine

Read from SKILL.md and the folder at commit 2d9a73b. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Caveman Search loads about 1.2k tokens when it runs, and up to ~3.4k if it reads all its reference files. Until then it costs about 121 tokens; SKILL.md has 600 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~121
When it runs · the whole SKILL.md, loaded when a task matches
~1.2k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~3.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from ragnar-pwninskjold/tech-snacks at commit 2d9a73b, republished under its MIT licence (© ragnar-pwninskjold). 600 words, ~1,185 tokens.

Download SKILL.mdSave it as .claude/skills/caveman-search/SKILL.md (or your agent's skills folder). This skill also uses 2 other files; get the full folder from GitHub.
name
caveman-search
description
Use when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something. Triggers on "caveman search", "caveman research", "/caveman-search", and on tooling/architecture questions where the user wants the direct answer, not a survey. Runs decomposed research (local repo + live web), verifies, and renders in a compressed caveman voice that leads with the load-bearing truth.

Research-backed answers in compressed caveman voice. Decompose research like Compound Engineering; render like Matt Pocock; always lead with the big truth.

What this does

  1. Detects whether the question needs LOCAL repo research, EXTERNAL web research, or BOTH.
  2. Runs a local-first escalation gate — skips web research when the repo already answers it.
  3. Fans out to parallel research agents.
  4. Optionally runs an adversarial verify pass (--verify).
  5. Renders the result in caveman voice, big-truth-first.

Flags

  • --verify (or --deep): run the adversarial verifier on each key recommendation.
  • --hard: maximum voice compression (see voice spec).
  • No flags: light-medium voice, no verifier.

Step 1 — Detect mode

Inspect the question and current context:

  • Mentions "this/here/our code", repo file names, or "add X here" -> repo mode.
  • Names external tech, asks "what should I use", greenfield, or "if I wanted to build" -> web mode.
  • Both signals, or genuinely ambiguous -> both.
  • Pure general-topic with no repo angle -> web only (do not waste a repo scan).

Also read the audience tier from the question. DEFAULT to builder and bias toward simple — recommend the simplest tool that clears the bar, not the most rigorous one. Escalate to expert ONLY on explicit signals: "ML team", "production-grade", "at scale", "I already use X", named low-level libraries, researcher-depth framing. Signals like "assume X is solved, I just care about Y" or "I want to ship" mean: stay simple, give one clear path. Pass the tier to the web researcher in Step 3.

Step 2 — Escalation gate (local-first)

If repo mode ran:

  • Spawn caveman-repo-scout first (or as part of the parallel fan-out) and read its local_coverage.
  • If local_coverage >= 3 AND the topic is NOT high-risk (auth, payments, data migration, external API integration) -> answer from local research, skip web.
  • Otherwise escalate: also run web research.
  • High-risk topics ALWAYS escalate to web, regardless of local_coverage.

Step 3 — Parallel fan-out

Dispatch the selected agents in a SINGLE message so they run concurrently (Task tool, one call per agent):

  • repo mode -> tech-snacks:research:caveman-repo-scout
  • web mode -> tech-snacks:research:caveman-web-researcher
  • both -> both, in parallel.

Pass each agent the topic (cleaned of flags). Pass the web researcher the audience tier from Step 1 so it surfaces the right depth of tools (simple/builder = batteries-included; expert = composable stacks ok).

Show full SKILL.md (233 more words)Show less

Step 4 — Verify pass (only if --verify)

Each caveman-web-researcher recommendation is already a self-contained claim. For each one, dispatch tech-snacks:research:caveman-verifier in parallel (one call per claim, single message). Reconcile:

  • verdict holds -> keep, unmarked.
  • verdict weakened -> keep with the verifier's correction folded in.
  • verdict wrong -> drop it; if the verifier supplied a real answer, use that instead.

Step 5 — Render

Read these two files and follow them exactly:

  • references/voice-spec.md — the voice TRANSFORM (not a description — apply it). Big Truth First is mandatory.
  • references/output-format.md — the section skeleton.

The digest you got from the researchers is verbose. Do NOT echo its register. Transform it through the voice spec — drop articles/filler/hedging, force fragments, lead with the Big Truth. Then run the voice spec's pre-output check before sending: if sentences read full and smooth like a report, you drifted — rewrite them. Keep the researcher's per-job 2-3-option structure; do not re-expand it into a buffet.

Apply the Auto-Clarity Exception from the voice spec for any security warning, irreversible action, or ordered sequence.

Step 6 — Output

Emit the response per the output skeleton: big truth -> tools -> shape (if architectural) -> bottom line -> sources (if web mode). Annotate corrected claims inline if --verify ran.

Honesty rule

If research found nothing solid, say so terse and name the closest option plus its gap. Never invent confident-sounding tool names — in this voice a wrong claim reads as authoritative, which is the worst failure mode.

© ragnar-pwninskjold, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files (references) in plugins/tech-snacks/skills/caveman-search of ragnar-pwninskjold/tech-snacks.

  • SKILL.md
  • references/output-format.md
  • references/voice-spec.md

Open the folder on GitHubat commit 2d9a73b

Compare with similar skills

Caveman Search next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Caveman Search compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Caveman Search this skillragnar-pwninskjold/tech-snacks137—~1.2kAutomated safety check: PassMIT
Brave Searchbadlogic/pi-skills2.6k5 repos~592Automated safety check: PassMIT
Enterprise AI Scenario MapMetaInFLow/Enterprise-ai-scenario-map-skill632—~1.8kAutomated safety check: PassMIT
Web Searchjjyaoao/HelloAgents3.2k1 repos~5.6kAutomated safety check: PassMIT
Ddg SearchTheSyart/claude-agent-examples4071 repos~493Automated safety check: PassNone
Local Web SearchuluckyXH/OpenMOSS1.3k—~392Automated safety check: NotesMIT

Similar skills

  • Brave Search

    badlogic/pi-skills

    Web search and content extraction via Brave Search API. An agent skill from badlogic/pi-skills.

    2.6k GitHub starsUsed in 5 repos~592 tokens
    Productivity & AutomationAuto-check passed

  • Enterprise AI Scenario Map

    MetaInFLow/Enterprise-ai-scenario-map-skill

    企业AI场景地图生成报告工具。通过 web-search 深度调研企业信息，按照V2.1标准模板生成结构化AI应用场景地图报告，包含企业画像、业务诊断、行业实践、AI场景全量表、实施路径等完整内容。

    632 GitHub stars~1.8k tokensUpdated 6 mo ago
    Productivity & AutomationAuto-check passed

  • Web Search

    jjyaoao/HelloAgents

    Implement web search capabilities using the z-ai-web-dev-sdk.

    3.2k GitHub starsUsed in 1 repo~5.6k tokens
    Productivity & AutomationAuto-check passed

  • Ddg Search

    TheSyart/claude-agent-examples

    Web search without an API key using DuckDuckGo Lite via webfetch.

    407 GitHub starsUsed in 1 repo~493 tokens
    Productivity & AutomationAuto-check passed

  • Local Web Search

    uluckyXH/OpenMOSS

    A skill your agent uses when the user asks for web search that should run via the local-160 Responses API with websearch tool (base URL like https://proxy.example.com, model gpt-5.2-codex(xhigh)).

    1.3k GitHub stars~392 tokensUpdated 3 mo ago
    Productivity & AutomationAuto-check: notes

  • Ask Search

    ythx-101/ask-search

    Web search via self-hosted SearxNG. An agent skill from ythx-101/ask-search.

    538 GitHub stars~332 tokensUpdated 6 mo ago
    Productivity & AutomationAuto-check passed

More from ragnar-pwninskjold/tech-snacks

  • Prd To UX

    ragnar-pwninskjold/tech-snacks

    A skill your agent uses when translating a PRD, feature spec, or raw product idea into screen-level prompts for web UX generators (Google Stitch, Figma AI, Pencil.dev, Claude Design, v0, or similar).

    137 GitHub stars~1.3k tokensUpdated 1 mo ago
    Auto-check passed

  • Scaffold Claude

    ragnar-pwninskjold/tech-snacks

    A skill your agent uses when the user wants to create, draft, or scaffold a CLAUDE.md (or AGENTS.md) file for their project.

    137 GitHub stars~1.7k tokensUpdated 1 mo ago
    Auto-check passed

  • UI Cloner

    ragnar-pwninskjold/tech-snacks

    A skill your agent uses when user provides a URL and wants to replicate or clone a website's UI, design, or visual style for their own product.

    137 GitHub stars~1.1k tokensUpdated 1 mo ago
    Auto-check passed

  • Mine Claude Md

    ragnar-pwninskjold/tech-snacks

    Mine recent Claude Code sessions for non-obvious, multi-file CLAUDE.md candidates, adversarially verify them, and propose paste-ready additions.

    137 GitHub stars~1k tokensUpdated 1 mo ago
    Auto-check passed

  • React Refactor Tournament

    ragnar-pwninskjold/tech-snacks

    Review React/Next.js code against the real vercel-react-best-practices skill, backlog the performance findings keyed to actual rule ids + impact tiers, rank the most over-subscribed tiers, then fix…

    137 GitHub stars~1.3k tokensUpdated 1 mo ago
    Auto-check passed

  • Lite Prd

    ragnar-pwninskjold/tech-snacks

    A skill your agent uses when the user hands you a vague or half-formed feature ask and wants it turned into a lightweight PRD, or asks to "write a PRD", "spec this feature", "flesh this out", or…

    137 GitHub stars~853 tokensUpdated 1 mo ago
    Auto-check passed

Categories

Questions about Caveman Search

What does Caveman Search do?

A skill your agent uses when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something. Caveman Search is an agent skill from ragnar-pwninskjold/tech-snacks. Use when the user wants a terse, opinionated, research-backed answer to a "what tech/tools/stack do I need for X" question, or asks how to approach building something.

When should I use Caveman Search?

Caveman Search fits situations like: the user wants a terse; research-backed answer to a what tech/tools/stack do I need for X question; asks how to approach building something; caveman research.

How do I install Caveman Search in Claude Code?

Run `npx skills add ragnar-pwninskjold/tech-snacks --skill caveman-search -a claude-code`. Or copy the skill folder (plugins/tech-snacks/skills/caveman-search in ragnar-pwninskjold/tech-snacks) into .claude/skills/caveman-search in your project. Claude Code loads it when a task matches its description.

How do I install Caveman Search in Codex?

Run `npx skills add ragnar-pwninskjold/tech-snacks --skill caveman-search -a codex`. Or copy the skill folder (plugins/tech-snacks/skills/caveman-search in ragnar-pwninskjold/tech-snacks) into .agents/skills/caveman-search in your project. Codex loads it when a task matches its description.

Can I use Caveman Search in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add ragnar-pwninskjold/tech-snacks --skill caveman-search -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/caveman-search, .gemini/skills/caveman-search, .github/skills/caveman-search and .opencode/skills/caveman-search in your project.

What does Caveman Search need to run?

SKILL.md names no scripts, command-line tools or credentials: Caveman Search is instructions for the agent only.

Does Caveman Search access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Caveman Search safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Caveman Search use?

Caveman Search is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Caveman Search use?

About 1.2k tokens (SKILL.md is roughly 4.7k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 2.3k tokens, read only when the agent opens those files.

What are the alternatives to Caveman Search?

Skills that share tags, products or a category with Caveman Search: Brave Search (badlogic/pi-skills, 2.6k stars), Enterprise AI Scenario Map (MetaInFLow/Enterprise-ai-scenario-map-skill, 632 stars), Web Search (jjyaoao/HelloAgents, 3.2k stars) and Ddg Search (TheSyart/claude-agent-examples, 407 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Caveman Search?

ragnar-pwninskjold (a GitHub user) maintains it in ragnar-pwninskjold/tech-snacks, which has 137 GitHub stars. The repository holds 7 skills in this directory. The repository was last updated on August 18, 2026.

Source: ragnar-pwninskjold/tech-snacks on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.