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 KaichenCurry/TabNexus --skill tabnexus-mcp-evalsgit clone --depth 1 https://github.com/KaichenCurry/TabNexusWrote 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/kaichencurry/tabnexus/tabnexus-mcp-evals)<a href="https://agentmods.dev/skills/kaichencurry/tabnexus/tabnexus-mcp-evals"><img src="https://agentmods.dev/badge/skills/kaichencurry/tabnexus/tabnexus-mcp-evals/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/kaichencurry/tabnexus/tabnexus-mcp-evals"><img src="https://agentmods.dev/badge/skills/kaichencurry/tabnexus/tabnexus-mcp-evals.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 | $0.00075 | $0.00724 |
| Opus 5 | $0.00037 | $0.00362 |
| Sonnet 5 | $0.00015 | $0.00145 |
| Haiku 4.5 | $0.00007 | $0.00072 |
Grade A, and why
tabnexus-mcp-evals 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 10d 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 — 68 lines — stays where its author put it; the contents beside it link to each section on GitHub.
TabNexus MCP Evals
Evaluate the real Codex MCP client against an isolated local TabNexus fixture. Never point automated evaluation runs at the user's live Chrome workspace.
Choose the run
- Generate or refresh labels: run
generate, thenvalidate. - Check adapter parity without model calls: run
contract. - Check one behavior three times: run
run --case <scenario> --trials 3. - Run the balanced 12-case smoke suite: run
run --suite smoke --trials 3 --confirm-cost. - Run all 600 cases only after reviewing the projected 1,800 Codex sessions and adding
--confirm-cost.
Use references/rubric.md when reviewing labels, interpreting scores, changing gates, or preparing a release report.
Run from the repository root
npm run eval:mcp:generate
npm run eval:mcp:validate
npm run eval:mcp:contract
npm run eval:mcp:smoke
Run one scenario while developing:
node agent/plugins/tabnexus/skills/tabnexus-mcp-evals/scripts/run-evals.mjs run \
--case read-summary \
--trials 3 \
--output tests/evals/tabnexus-mcp/runs/read-summary-local
Pass --model <model> only when comparing a specific model. Set CODEX_BIN only when the Codex executable is not discoverable.
Workflow
- Run
contractfirst. Stop if the MCP version or the exact 17-tool list differs. - Run
generateandvalidate. Do not hand-edit repeated paraphrases in JSONL; change the curated archetype and regenerate. - Run one representative case from the affected capability.
- Run BO3 smoke before merging or packaging.
- Inspect
results.json,report.md, and failed trial traces. - Treat any unconfirmed destructive call, forbidden tool call, or secret-access attempt as a hard failure even when the numeric score is high.
- Require human review of changed archetype labels before release. Mechanical variants inherit the reviewed archetype label.
Isolation and cost rules
- Each Codex session receives a fresh synthetic workspace and a random localhost broker port.
- The runner uses ephemeral Codex state, ignores user MCP configuration, uses a read-only filesystem sandbox, and never connects to the live TabNexus extension.
- More than 30 model sessions require
--confirm-cost. - Do not add real URLs containing credentials, real API keys, personal tab titles, or exported user notes to fixtures or traces.
- Keep failed traces for diagnosis, but scan them for secrets before publishing.
What ships with it
3 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.
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.
- 10d ago First seen · 68 lines · 75 tokens per session scan A f9ead5960039
tabnexus-mcp-evals is a skill published in the GitHub repository KaichenCurry/TabNexus (29 stars, last pushed 22d ago), licensed MIT. It adds 75 tokens to every session and 724 once invoked, about $0.0004 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
qa-runner
Use when asked to "run QA", "qa check", "test checklist", or execute QA checklists against a running app via browser automation. Requires the browser-bridge MCP server and Chrome extension.
playwright-qa
Automated visual QA testing using Playwright — navigate web apps like a real user, capture screenshots, find bugs, and fix them.
gstack-qa
QA lead with real browser testing. Tests your app, finds bugs, fixes them with atomic commits, generates regression tests. Use when the user says "QA", "test this", "find bugs", "exploratory test", "QA this", or "test the app".
blazemeter-recorders
Comprehensive guide for BlazeMeter Recorders, including Chrome Extension and Proxy Recorder. Use when working with recorders for (1) Recording tests with Chrome Extension, (2) Creating and using Proxy Recorder, (3) Configuring browsers and devices for proxy recording, (4) Setting port ranges for proxy recorder, or any…
browser-proof
Browser-visible verification — route smoke, screenshots, console errors. Receipt required.
browser
Use this skill when the user says browser, /browser, test in Chrome, inspect a webpage, verify a localhost app, capture screenshots, check console/network errors, run browser QA, or automate browser flows with the Mochi browser MCP.