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 agentmods add skills/ninemindai/agentgem/verifynpx skills add ninemindai/agentgem --skill verifygit clone --depth 1 https://github.com/ninemindai/agentgemWhat 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 | $0.00025 | $0.00648 |
| Opus 5 | $0.00013 | $0.00324 |
| Sonnet 5 | $0.00005 | $0.00130 |
| Haiku 4.5 | $0.00003 | $0.00065 |
Grade C, and why
verify scanned grade C with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- **After pulling a refactor, clean `dist/` first** — `rm -rf dist tsconfig.tsbuildinfo Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- SSE surfaces are also curl-able directly, e.g. How it starts
The opening of the file, as written. The whole thing — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Verifying agentgem changes at the running surface
Build + launch
pnpm build # tsc -b AND the console SPA bundle — both required
AGENTGEM_HOME=$(mktemp -d) PORT=<unused-port> node dist/client.js
- Entry is
dist/client.js(the local-mode server) — NOTdist/index.js, which no longer exists after the server-library extraction into@agentgem/app. The banner isagentgem (client mode) listening at …. - After pulling a refactor, clean
dist/first —rm -rf dist tsconfig.tsbuildinfo packages/*/tsconfig.tsbuildinfo && pnpm build. Stale incremental output can leave a deaddist/index.jsimporting a since-removed package (e.g.@agentgem/deploy), which crashes on boot. - Always restart the server after a rebuild — it caches the console SPA HTML at boot, and
built-in miniapp constants (e.g.
EMBER_HTML) are module-level imports. - Pick a genuinely unused port and confirm ownership: with concurrent agent sessions in
sibling worktrees, another session's server may already hold your port and your
nodeexits after printing its banner. Checklsof -p $(lsof -tnP -iTCP:<port> -sTCP:LISTEN) | grep cwdpoints at YOUR worktree. AGENTGEM_HOMEisolates agentgem's own data; watched transcripts still come from the real~/.claude, so Watch/hygiene/Ember surfaces show real sessions.
Drive
- Console pages are hash routes:
http://localhost:<port>/#/arcade,#/watch,#/play, … - Browser consent for Play capabilities is per-origin
localStorage(agentgem:play:consent:<gem>:<cap>) — a reused port silently skips consent modals; clear those keys to test the consent path. - Sealed miniapp iframes are null-origin: you cannot reach inside from the top document. To
observe the broker wire, wrap
window.EventSourcein the top window (streams live there), or listen formessageevents (game→host half is visible onwindow). - SSE surfaces are also curl-able directly, e.g.
curl -N "http://localhost:<port>/api/watch/hygiene?file=<transcript>".
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.
- 3d ago First seen · 48 lines · 25 tokens per session scan C e292d8ebc37c
verify is a skill published in the GitHub repository ninemindai/agentgem (39 stars, last pushed 6d ago), licensed MIT. It adds 25 tokens to every session and 648 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it C with 2 findings (recursive force delete, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
webapp-testing
Start/reuse a local app, wait for readiness, inspect rendered state/console/network, act from observed selectors, and verify with evidence.
agent-browser
Use agent-browser CLI for website automation: navigate, click/fill, scrape, screenshot, test web apps, debug console/network.
playwright-e2e
End-to-end testing and click-verification of web apps with Playwright — install, write specs, drive authenticated pages, take screenshots, and run in CI. Use when verifying a web change actually works in a real browser, testing user flows, debugging a UI bug live, or confirming an authed/gated page before shipping.…
test-browser
Run browser tests on pages affected by current PR or branch.
browser-qa
Use this skill to automate visual testing and UI interaction verification using browser automation after deploying features.
vibe-e2e
Use when performing end-to-end verification before delivery. Launches the project, opens browser, verifies every page renders, every button works, every API responds. Triggers on "E2E", "end-to-end test", "真机验证", "交付验证", "跑一下看看", "打开浏览器验证".