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 IgorGanapolsky/ThumbGate --skill thumbgate-guardgit clone --depth 1 https://github.com/IgorGanapolsky/ThumbGateWrote 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/igorganapolsky/thumbgate/thumbgate-guard)<a href="https://agentmods.dev/skills/igorganapolsky/thumbgate/thumbgate-guard"><img src="https://agentmods.dev/badge/skills/igorganapolsky/thumbgate/thumbgate-guard/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/igorganapolsky/thumbgate/thumbgate-guard"><img src="https://agentmods.dev/badge/skills/igorganapolsky/thumbgate/thumbgate-guard.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 6 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Tool Misuse · line 18 Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).Fix: Validate all tool parameters against an allowlist. Reject dangerous parameter values (shell=True, --force, -rf /) and use safe defaults.
- medium MCP Rug Pull · line 3 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
- medium MCP Rug Pull · line 24 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
- medium MCP Rug Pull · line 31 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
- medium MCP Rug Pull · line 40 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
- medium MCP Rug Pull · line 55 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
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 | $0.00201 | $0.00930 |
| Opus 5 | $0.00101 | $0.00465 |
| Sonnet 5 | $0.00040 | $0.00186 |
| Haiku 4.5 | $0.00020 | $0.00093 |
Grade A, and why
thumbgate-guard 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 7d 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 — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ThumbGate Guard
Escalate the agent's most recent mistake into an enforced ThumbGate prevention rule — a PreToolUse block gate that intercepts the same bad tool-call shape before it runs again, in this session and every future one (Claude Code, Cursor, Codex, Gemini, Amp, Cline).
This skill wraps existing ThumbGate capability and adds no new logic — it routes to the real
capture_feedback + force-promote path.
Workflow
- Name the mistake. From the recent conversation, state the specific bad action in one
sentence (e.g.
git push --force origin main,DROP TABLE users, deploy without tests). - Capture it with the
capture_feedbackMCP tool:signal: "down",context,whatWentWrong,whatToChange,tags. If the user only gave a vague signal, pass the recent turns asconversationWindow/chatHistoryfor history-aware distillation instead of refusing. - Promote it to an enforced gate:
This prints the newnpx thumbgate force-gate "<one-sentence context of the mistake>"gateIdand the active gate count. - Confirm enforcement. Show the user the promoted rule and state that it is now a PreToolUse block — not just a logged note.
Full capture_feedback field contract, force-gate flags, and the guided first-rule walkthrough
(npx thumbgate quickstart) are in
references/capture-and-promote.md.
Example
Input: "the agent force-pushed to main and overwrote a teammate's commit — never again"
Action:
capture_feedback→signal: down,context: "agent force-pushed to main, overwrote a teammate's commit",whatWentWrong: "destructive force-push to a shared branch",whatToChange: "block force-push to main",tags: "git,force-push".npx thumbgate force-gate "force-push to main overwrote a teammate's commit".- Report: the real
gateId+ "force-push to main is now blocked as a PreToolUse check."
Troubleshooting
- "Vague feedback rejected": add
whatWentWrong+whatToChange, or passconversationWindowso it can distill from history. force-gateprints nothing or errors: ThumbGate may not be wired — run the thumbgate-doctor skill.- Rule didn't fire next time: confirm it is active with the thumbgate-rules skill and that hooks are installed (thumbgate-doctor).
What ships with it
1 file 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.
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.
- 7d ago First seen · 58 lines · 201 tokens per session scan A a1f835f59450
thumbgate-guard is a skill published in the GitHub repository IgorGanapolsky/ThumbGate (26 stars, last pushed yesterday), licensed MIT. It adds 201 tokens to every session and 930 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 2026-09-03.
Other skills, from other repositories
agentguard
Runtime guardrails for AI coding agents. Stop loops, budget overruns, retry storms, and timeouts before they burn money. Zero dependencies, local-first, MIT licensed.
crm-pipeline-playbook
How to answer pipeline, renewal, and account-health questions using the CRM tools.
deciding-who-decides
Decides who holds authority for a change — the agent at the edge or a human gate — by matching decision rights to reversibility, evidence, and consequence, and names the escalation trigger. Use when an agent could act on something irreversible, trust-bearing, or thinly evidenced. Do not use for trivial reversible…
verifying-final-artifacts
Verifies the delivered artifact a consumer actually receives (a figure, chart, PDF, SVG, screenshot, compiled build, or deployed response) by regenerating it fresh and directly observing the output, not the source that produces it, and ends in one honest verdict. Use when the deliverable is a produced artifact and…
closing-stale-packets
Brings an abandoned or half-filled change packet to an honest terminal state by completing it, closing it with a recorded rationale, or deleting it. Use when ng status flags a scaffold or invalid packet, a long session left a draft behind, or work was abandoned mid-packet. Do not use for an actively in-progress…
vetting-outside-code-and-models
Checks dependencies, models, APIs, SaaS tools, generated files, and vendor claims against how you will actually use them, the proof you have, the gaps, and the release impact. Use when any of these affect evidence, permissions, data, releases, or public trust. Do not use for an internal cleanup with no outside…