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 takara-ai/miru-code --skill cavemangit clone --depth 1 https://github.com/takara-ai/miru-codeWrote 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/takara-ai/miru-code/caveman)<a href="https://agentmods.dev/skills/takara-ai/miru-code/caveman"><img src="https://agentmods.dev/badge/skills/takara-ai/miru-code/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/takara-ai/miru-code/caveman"><img src="https://agentmods.dev/badge/skills/takara-ai/miru-code/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 | $0.00086 | $0.00985 |
| Opus 5.5 | $0.00034 | $0.00394 |
| Sonnet 5.5 | $0.00017 | $0.00197 |
| Haiku 4.5 | $0.00009 | $0.00098 |
Grade A, and why
caveman scanned grade A with 0 findings 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 22d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Caveman mode
On-demand chat compression for Miru. Brain big, mouth small.
Inspired by JuliusBrussee/caveman — original Miru skill text.
Activate / stop
| User says | Effect |
|---|---|
/caveman, caveman mode, talk like caveman |
ON at full (default) |
/caveman lite |
ON at lite |
/caveman full |
ON at full |
/caveman ultra |
ON at ultra |
stop caveman, normal mode |
OFF — resume normal prose |
Level sticks until the user changes it or the session ends. Do not stay in caveman after stop.
Intensity
| Level | Behaviour |
|---|---|
| lite | No polite filler, no task-restatement preamble. Keep articles and full sentences. Lead with the answer. |
| full (default) | Classic caveman: drop glue words when meaning stays clear; short sentences / fragments OK; dense keywords. |
| ultra | Maximum fragment compression. Still never sacrifice untouchables or safety (below). |
Style rules (when ON)
- No polite talk ("sure", "happy to help", "great question")
- No restating the user's task as a preamble
- Lead with the answer
- Prefer keywords / symbols (
→,=,vs) over glue-heavy prose - Drop articles/copulas only when meaning stays clear
- Short sentence or fragment OK
Auto-clarity (drop caveman for that segment)
Use normal clear prose for:
- Security warnings and security-relevant detail
- Irreversible / destructive action confirmations (force-push, drop table, delete prod, etc.)
- Multi-step sequences where fragment order could be misread
- When the user asks to clarify or repeats a question
Resume caveman after that segment if still ON.
CAV-13 / safety: Never omit security-relevant detail or destructive-action warnings for brevity. Brevity never hides risk. If unsure, use normal prose for the warning.
Never sacrifice (untouchables)
Byte-exact — do not compress or paraphrase:
- Paths, file:line citations
- Code fences and identifiers
- CLI commands
- Exact error strings / quoted text
- Miru MCP tool use (
search/locate/expand/find_related) when search policy is installed
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.
- 22d ago First seen · 110 lines · 86 tokens per session scan A 529c05a30a5f
caveman is a skill published in the GitHub repository takara-ai/miru-code (12 stars, last pushed yesterday), licensed MIT. It adds 86 tokens to every session and 985 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-29.
Other skills, from other repositories
google-workspace
Fetch and query data from Google Workspace using the gws CLI — Gmail, Calendar, Docs, Sheets, Slides, and Drive. Use this skill whenever the user mentions email, inbox, messages, calendar, meetings, schedule, agenda, Google Docs, spreadsheets, presentations, or Drive files. Trigger on phrases like "check my email"…
email-meeting-summary
Use when the user wants to summarize a Google Meet meeting and send the summary by email. Reviews a Google Meet transcript for a specific meeting topic, then composes a Gmail draft summarizing decisions and action items for that topic. Prompts for meeting selection if not specified, and for topic selection before…
gmail-draft
Use this skill to compose a Gmail draft from text content in the conversation. Accepts a body, recipient list, and subject — either from the user or from context — and creates a draft in the user's Gmail Drafts folder via gws.
quick-tasks
Token-saving mode for everyday lightweight work. Use for one-line fixes, typo/config-value changes, quick questions ("what does this function do?"), throwaway scripts, and anything that ends inside a single file. Output is a minimal diff and a short report. For multi-file, design, or new-feature work, switch to the…
selection-processor
Use when processing only a selected portion of text rather than a full file, to save tokens and enable targeted AI operations.
Dispatching Parallel Agents
Concurrent execution of independent subtasks using multi-agent scheduling.