seatbelt

seatbelt is a skill for Claude Code from adamlinscott/claude-skills. It costs 201 tokens per session (1,994 once invoked), scanned A, original, MIT.

A security setup for Claude Code that creates permission rules for a repository. It offers a developer mode with more freedom and a stricter vibe mode that blocks actions unless allowed.

In plain words
What is it for?
Use it to choose repository permissions before running Claude Code with fewer approval prompts.
Why use it?
It reduces the chance that the agent changes or removes something you did not intend, while still allowing more automatic work.

Skill for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: reads .claude/ paths; mentions CLAUDE.md; mentions subagents.

Part of the claude-skills plugin — 19 skills shipped together

Good fit Use it to choose repository permissions before running Claude Code with fewer approval prompts.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/adamlinscott/claude-skills/seatbelt
Install

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.

Any agent
npx skills add adamlinscott/claude-skills --skill seatbelt
Clone the repo
git clone --depth 1 https://github.com/adamlinscott/claude-skills

Made for: Claude Code.

Or install claude-skills, the plugin that ships this one along with the rest of its 19 skills.

Wrote 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.

agentmods badge for seatbelt

README.md
[![agentmods](https://agentmods.dev/badge/skills/adamlinscott/claude-skills/seatbelt/github.svg)](https://agentmods.dev/skills/adamlinscott/claude-skills/seatbelt)
Your own site
<a href="https://agentmods.dev/skills/adamlinscott/claude-skills/seatbelt"><img src="https://agentmods.dev/badge/skills/adamlinscott/claude-skills/seatbelt/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.

agentmods 80×15 button for seatbelt

Your own site · 80×15
<a href="https://agentmods.dev/skills/adamlinscott/claude-skills/seatbelt"><img src="https://agentmods.dev/badge/skills/adamlinscott/claude-skills/seatbelt.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 201 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,994 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00201 $0.01994
Opus 5 $0.00101 $0.00997
Sonnet 5 $0.00040 $0.00399
Haiku 4.5 $0.00020 $0.00199

Measured today against content hash c0fb9f9e6939, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

seatbelt 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 today.

The scan reads SKILL.md. This mod also ships 2 executable files (assets/guard.mjs, assets/guard.test.mjs), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

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.

skills/seatbelt/SKILL.md · 142 lines

How it starts

The opening of the file, as written. The whole thing — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Seatbelt

A seatbelt doesn't limit how fast you drive. It's what lets you drive fast at all.

This skill writes a per-person permission policy for the current repo so Claude can be given more autonomy, not less. It is the collection's only mutating skill — everything else here is read-only. For a read-only report on this repo's AI setup, use /context-audit instead; this skill deliberately has no dry-run mode, because that would just be /context-audit again.

What it honestly does: stops Claude over-reaching. What it does not do: stop a person who dismantles their own setup. Say this to the user; never let them believe otherwise.

Explicit invocation only (disable-model-invocation: true). Never offer or auto-run this because a user sounded worried, said Claude broke something, or challenged a change. A developer with their own permission setup would find an uninvited rewrite of it worse than the problem they were complaining about. It runs when someone types /seatbelt, and not otherwise.

Quick start

/seatbelt              # asks which mode, then sets up
/seatbelt --developer  # skip the question
/seatbelt --vibe       # skip the question
/seatbelt --allow "npm run deploy:staging"   # add one override
/seatbelt --uninstall  # remove exactly what the manifest records

--developer and --vibe together is an error, not a silent pick. --uninstall beats mode flags. With no TTY and no mode flag (headless, or running as a subagent), print what you would have asked and exit non-zero — never guess a mode.

Workflow

1. Ask the mode first

Before any grounding. A wall of stack-detection output is the worst possible first screen for a non-technical user, and the question costs nothing to ask early.

Frame by consequence, not identity. "I know what a PR is" biases everyone toward the mode with no brakes, because everyone wants fewer prompts. Use AskUserQuestion with:

  • Let Claude move fast — it can push, merge and deploy on its own. Pick this if you can read a diff and undo a bad merge.
  • Keep a human in the loop — Claude writes code freely, but anything permanent (merging, deploying, deleting) waits for you.

Read the full file on GitHub · 142 lines

Files

What ships with it

5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

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.

  1. today Changed · +1 tokens per session c0fb9f9e6939
  2. yesterday Changed · +4 tokens per session 8668f53d760e
  3. 12d ago First seen · 142 lines · 196 tokens per session scan A aeff57faae4c

Subscribe to this mod's changes

seatbelt is a skill published in the GitHub repository adamlinscott/claude-skills (3 stars, last pushed today), licensed MIT. It adds 201 tokens to every session and 1,994 once invoked, about $0.0010 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 2026-08-31.

Related

Other skills, from other repositories

implement

Use in the Implement phase whenever writing or editing production Java code, or fixing a bug, in a Spring/Spring Boot project. Enforces test-first (red-green-refactor), executes the approved plan step by step, and honors the project's path-scoped tech-stack rules and the task's enforcement set. Preloaded into…

taipt1504/claudehut · 73 tokens

review

Use in the Review phase before claiming any Java/Spring task is complete, fixed, or passing. Spawns the auditor subagents that check the implementation against every applicable skill, rule, and memory item, runs the test suite for fresh evidence, and re-spawns for at most two fix rounds before reporting what survives.…

taipt1504/claudehut · 81 tokens

claudehut-workflow

Use at the start of every session and whenever beginning a coding task in a Java/Spring backend - establishes the ClaudeHut 7-phase agentic workflow, the complexity-tier routing that lets small tasks skip deliberation phases, and the laws that govern which skills and rules must fire. Injected at session start; also…

taipt1504/claudehut · 86 tokens

discover

Use as the FIRST phase of EVERY coding task, before brainstorming - grounds the work in the existing codebase (entry points, key types, structure) and proves whether something reusable already exists. Produces the reuse-scan artifact the write gate requires and the reuse DECISION (adopt / extend / new). Runs inline on…

taipt1504/claudehut · 95 tokens

brainstorm

Use to generate and weigh solution approaches for a problem, after discovery has grounded the context. Produces two or more genuinely distinct options scored on trade-offs, recommends one, and (for code tasks) assembles the enforcement set the rest of the workflow audits against. General-purpose ideation — works for…

taipt1504/claudehut · 74 tokens

write-plan

Use in the Plan phase after the spec is approved - dispatches the planner agent to draft the executable plan from the standard template (decision summary, T-xxx task breakdown with test-first + verify per task), gets the user's approval, records the plan (opening the write gate), and mirrors the breakdown into Claude…

taipt1504/claudehut · 93 tokens