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 Miosa-osa/OSA --skill arsenal-tool-validationgit clone --depth 1 https://github.com/Miosa-osa/OSAWrote 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/miosa-osa/osa/arsenal-tool-validation)<a href="https://agentmods.dev/skills/miosa-osa/osa/arsenal-tool-validation"><img src="https://agentmods.dev/badge/skills/miosa-osa/osa/arsenal-tool-validation/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/miosa-osa/osa/arsenal-tool-validation"><img src="https://agentmods.dev/badge/skills/miosa-osa/osa/arsenal-tool-validation.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 | $0.00142 | $0.01259 |
| Opus 5.5 | $0.00057 | $0.00504 |
| Sonnet 5.5 | $0.00028 | $0.00252 |
| Haiku 4.5 | $0.00014 | $0.00126 |
Grade A, and why
arsenal-tool-validation 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 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.
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 — 111 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Arsenal — Tool Validation & Risk Gating
Derived from T3MP3ST's src/arsenal/ (Apache-2.0). The principle: an
intrusive tool is INERT until a human has approved it — and the hottest
actions fire a loud, audited warning every single time, even when approved.
1. Risk tiers
| Tier | Meaning | Gated? | Warning every use? |
|---|---|---|---|
safe |
read-only probes, passive recon | no | no |
active |
touches the target (scans, fuzzing) | no (scope-gated) | no |
intrusive |
exploits, state-changing actions | yes | no |
credential |
password attacks, credential stuffing | yes | yes |
dangerous |
full post-ex frameworks, destructive ops | yes | yes |
Undefined tier → treat as safe for gating but verify by inspection; a tool
that mutates state must never sit in safe.
2. The approval gate
Two ways a gated tool becomes approved:
- Interactive: the first time the operator reaches for the tool, the system asks. A yes approves the TOOL (by name) for the rest of the session — "approve once, then free". Every use is still audited.
- Pre-authorized allowlist: for headless/unattended runs (benchmarks, automated hunts), the operator hands in an allowlist up front. Listed tools run free; anything off-list DENIES.
FAIL-SAFE rule: a gated tool that is neither pre-approved nor interactively approvable (no approver wired) is DENIED. An unattended run never silently fires an exploit because nobody was there to say no.
Spicy warnings: credential/dangerous tools fire a loud, audited,
non-blocking WARNING on EVERY use even after approval — the operator should
always SEE what the agent is about to do.
3. Egress scope — the always-on gate
Approval and scope are separate gates:
- The egress scope gate refuses any action targeting an out-of-scope host, REGARDLESS of tool approval. Approval says "you may use this tool"; scope says "you may point it there". Both must pass.
- Scope is a list written at engagement start (see
penetration-testing); the gate checks targets against the list, not against the operator's claims. - An operator can extend gating to cover built-in probes too (T3MP3ST's
GATE_BUILTINSpattern) — when in doubt, gate more, not less.
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 · 111 lines · 142 tokens per session scan A c4811b3da9af
arsenal-tool-validation is a skill published in the GitHub repository Miosa-osa/OSA (78 stars, last pushed 11d ago), licensed Apache-2.0. It adds 142 tokens to every session and 1,259 once invoked, about $0.0006 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 2026-09-14.
Other skills, from other repositories
unit-converter
Converts between common units of measurement including temperature, distance, and weight.
code-review
Reviews code changes for quality, security, and best practices.
tui-design
This skill should be used when designing terminal user interfaces, creating TUI layouts, choosing TUI color schemes, implementing keyboard navigation, building terminal dashboards, or working with any TUI framework. Activates on mentions of TUI design, terminal UI, Ratatui layout, Ink components, Textual widgets…
bun-development
Fast, modern JavaScript/TypeScript development with the Bun runtime, inspired by oven-sh/bun.
worker-mcp-orchestrator
Guides the coordinator agent on how to orchestrate, delegate to, and supervise local worker agents using the worker-mcp server.
clack
Build and maintain interactive Node.js/Bun CLIs with Clack (@clack/prompts and @clack/core). Use when an AI agent needs to design prompt flows, implement cancellation-safe input handling, add spinners/log/note UX, apply Clack best practices, or generate runnable examples aligned with official Clack docs.