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 bearlike/Assistant --skill mewbo-harnessgit clone --depth 1 https://github.com/bearlike/AssistantWrote 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/bearlike/assistant/mewbo-harness)<a href="https://agentmods.dev/skills/bearlike/assistant/mewbo-harness"><img src="https://agentmods.dev/badge/skills/bearlike/assistant/mewbo-harness/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/bearlike/assistant/mewbo-harness"><img src="https://agentmods.dev/badge/skills/bearlike/assistant/mewbo-harness.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00093 | $0.01202 |
| Opus 5 | $0.00046 | $0.00601 |
| Sonnet 5 | $0.00019 | $0.00240 |
| Haiku 4.5 | $0.00009 | $0.00120 |
Grade A, and why
mewbo-harness 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 9d 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 — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
How this harness behaves
Runtime semantics that no tool schema can carry. Everything here is a property of the ENGINE; anything specific to your session or your project is not stated here because it would be a guess — get those from the tool that owns them.
Tool results are capped, and a capped result is windowed
Every tool result is fitted to a per-tool character cap before you see it.
| Population | Cap when the tool declares none |
|---|---|
registry tools (ToolSpec.max_result_chars) |
2000 characters |
| session tools (plugin/per-session tools) | 200000 characters |
directly-bound tools (activate_skill, the spawn family) |
2000 characters, except activate_skill at 200000 |
| shell tools | 30000 characters (declared, not the default) |
When a result exceeds its cap it is windowed, not head-truncated: the head and
the tail are both kept and the middle is replaced by a marker of the form
[... N characters omitted ...]. So:
- A marker means you are missing the middle — not the end. The verdict of a command (traceback, exit banner, failure summary) survives.
- Do not treat a windowed result as complete. Narrow the read instead of
re-issuing the same call:
grepfor the line you need, or page.
A capped result is capped for YOU only. The transcript keeps its own, far larger snapshot, so a result you read in part is stored whole — which means the session page, the console and anyone reading the store see the complete text and have no way to tell that you did not. Nothing warns either side. If a decision turns on a result that carries a marker, say so rather than assuming the reader can see what you were missing.
Paging. read_file is line-windowed — pass offset (0-based start line) and
limit (max lines, default 2000) to walk a large file instead of re-reading it.
A tool that pages says so in its own schema; when it does not, narrow the query.
A timeout bounds the WAIT, not the work
Tool execution sits under a ceiling — 120 seconds unless the tool declares its own. A tool that hits it comes back as a timed-out call.
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.
- 9d ago First seen · 98 lines · 93 tokens per session scan A 6e9aef2d9caa
mewbo-harness is a skill published in the GitHub repository bearlike/Assistant (42 stars, last pushed yesterday), licensed MIT. It adds 93 tokens to every session and 1,202 once invoked, about $0.0005 per session on Opus 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 2026-08-30.
Other skills, from other repositories
playwright
Use when the task requires capturing or automating a real browser from the terminal.
pydantic-ai
Build production-ready AI agents with PydanticAI — type-safe tool use, structured outputs, dependency injection, and multi-model support.
html-artifacts
Author the HTML for a plan artifact, dashboard iframe, or Slack attachment — structure, design plan, available runtime, theming, and craft. Read this before writing HTML for saveplan, outputiframe, or slackattachhtml.
bootstrap-repo-analysis
First-time analysis of a repository with no prior reviewer outcomes. Crawl historical merged-PR review feedback with the gh CLI (plus any preloaded samples), extract the team's review norms, and synthesize the initial per-repo review-style prompt. Use this for a cold-start repo; use continual-learning instead once the…
prior-auth-packet-builder
Build a concise prior authorization packet from local case files and payer policy docs.
credit-note-fixer
Fix the tiny credit-note formatting bug and rerun the exact targeted test command.