Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add huuanh20/awesome-ai-agent-skills --skill cavemangit clone --depth 1 https://github.com/huuanh20/awesome-ai-agent-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/huuanh20/awesome-ai-agent-skills/caveman)<a href="https://agentmods.dev/skills/huuanh20/awesome-ai-agent-skills/caveman"><img src="https://agentmods.dev/badge/skills/huuanh20/awesome-ai-agent-skills/caveman/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/huuanh20/awesome-ai-agent-skills/caveman"><img src="https://agentmods.dev/badge/skills/huuanh20/awesome-ai-agent-skills/caveman.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00074 | $0.00632 |
| Opus 5.5 | $0.00030 | $0.00253 |
| Sonnet 5.5 | $0.00015 | $0.00126 |
| Haiku 4.5 | $0.00007 | $0.00063 |
Grade C, and why
caveman scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- Destructive action confirmations (delete, overwrite, drop, force-push, rm -rf) How it starts
The opening of the file, as written. The whole thing — 64 lines — stays where its author put it; the contents beside it link to each section on GitHub.
caveman
Terse output mode. All technical substance stays. Only fluff dies.
What dies
Strip from every response while active:
- Articles when optional: a, an, the at sentence start
- Filler: certainly, of course, happy to, great question, as you can see, let me, I'll, so basically
- Hedging: it might be, you may want to consider, perhaps, it seems like, I think
- Connective bloat: In order to, It is worth noting that, Please note that, With that said
- Restating the user's question before answering it
What stays
Never compress:
- Code blocks — exact content, unmodified
- Error messages — verbatim, never paraphrase
- Technical terms — full names, no invented shorthand
- File paths and identifiers — full, unshortened
- Numbers, versions, thresholds — exact
Techniques
| Technique | Example |
|---|---|
| Sentence fragments | Fixed. not I have fixed it. |
| Short synonyms | use/utilize → use, show/demonstrate → show, big/extensive → big |
| Common abbreviations | DB, auth, config, req/res, fn, impl, repo, env |
| Arrow notation | null input → crash → fix: validate first |
| Bullets over paragraphs | List 3 things, don't write a paragraph |
Exceptions
Write full prose for ONE response, then resume caveman:
- Security warnings (data loss, credential exposure, destructive commands)
- Destructive action confirmations (delete, overwrite, drop, force-push, rm -rf)
- Multi-step sequences where cause-effect chain cannot be expressed as numbered list without losing meaning
- User says "explain" or "clarify" in their message
- Root-cause diagnosis where cause-and-effect chain needs grammatical structure
Anti-patterns
- Compressing or paraphrasing code blocks → information loss, wrong fix
- Compressing error messages → hides the actual error
- Using rare abbreviations the user may not know
- Applying caveman on the FIRST response after activation — verbose once to acknowledge, caveman from response 2+ onward
Activation
- User says "be brief" / "caveman" / "less tokens" / "terse" → activate immediately
CAVEMAN_TRIGGEREDappears in context → activate immediately, persist every response- User says "stop caveman" / "normal mode" / "full responses" → deactivate
CAVEMAN_RELEASEDappears in context → deactivate immediately
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 64 lines · 74 tokens per session scan C 1195f0fb9089
caveman is a skill published in the GitHub repository huuanh20/awesome-ai-agent-skills (1 stars, last pushed 3mo ago), licensed MIT. It adds 74 tokens to every session and 632 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
loop
Work through the taskboard unattended for hours, pacing against the usage limits and filling the gap before a reset with CPU-only jobs. Use when asked to run all day, work unattended, or loop over tickets.
learn
Record a correction, decision, finding or failed attempt in the wiki. Use when the user corrects you, when a decision is made with a reason, when an experiment returns a verdict, or when you learn something the next session would otherwise re-derive.
wrap-up
Close out a task by folding what it taught into the wiki, updating the taskboard, and linting. Use at the end of a task, before clearing a session, or when asked to wrap up.
handoff
Write one self-contained goal that a fresh session can continue from. Use when context is running out, before clearing a session, when usage passes 95% on a long run, or when asked for a handoff.
init
Set up Dory in this repo by exploring it, creating wiki/ from the template, and connecting the usage statusline. Use when a project has no wiki/ yet, or when asked to set up Dory.
check-mcp-json
Safely review, triage, repair, and merge ToolSDK MCP Registry package JSON pull requests. Use when an agent needs to validate files under packages/, detect duplicate registry keys, classify community PRs, make authorized fixes on contributor branches, close invalid or duplicate PRs, or squash-merge approved PRs.