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/agentcomputerai/torch/costconpx skills add AgentComputerAI/torch --skill costcogit clone --depth 1 https://github.com/AgentComputerAI/torchWhat 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 | $0.00099 | $0.02376 |
| Opus 5 | $0.00049 | $0.01188 |
| Sonnet 5 | $0.00020 | $0.00475 |
| Haiku 4.5 | $0.00010 | $0.00238 |
Grade B, and why
costco scanned grade B 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 2d 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.
Sends data to an external URLmediumData exfiltration
A POST to an outside endpoint may be telemetry or may be exfiltration; either way the mod talks to somewhere, and you should know where.
const res = await fetch("https://gdx-api.costco.com/catalog/search/api/v1/search", { method: "POST", Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
description: Proven scraping playbook for costco.com category listings (e.g. /laptops.html, /computers.html). Akamai Bot Manager hard-blocks bare curl with HTTP 403 on the HTML pages, but Costco's internal catalog API at How it starts
The opening of the file, as written. The whole thing — 153 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Costco (costco.com)
Costco's category HTML pages (
/laptops.html,/computers.html, etc.) sit behind Akamai Bot Manager and return 403 to anything that doesn't look like a fully-warmed browser. The actual product grid is rendered client-side by a Next.js app at/consumer-web/search/prd/catalog-usbc/that POSTs to a backend search API. That backend API is public — no cookies, no tokens, no bot manager — so the correct strategy is to replay it directly withfetchand ignore the HTML page entirely.
Detection
| Signal | Value |
|---|---|
| CDN / WAF | Akamai (server: AkamaiGHost, bm_ss/bm_s/akavpau_* cookies) |
| Frontend | Next.js SPA at /consumer-web/search/prd/catalog-usbc/ |
| Bare curl on HTML | 403 Forbidden (384-byte Akamai challenge body) |
Real Chrome (127.0.0.1:9222) on HTML |
200, but product DOM is empty until hydration |
| Backend API | https://gdx-api.costco.com/catalog/search/api/v1/search (POST) |
| API auth | None — no cookies, no tokens |
| API bot manager | None — plain fetch with a UA and 3 client headers works |
Architecture
GET /laptops.htmlreturns an Akamai-protected Next.js shell. The DOM has zero product tiles atdomcontentloaded.- The shell hydrates and fires a single
POST https://gdx-api.costco.com/catalog/search/api/v1/searchwith a JSON body containingfilterBy: ["attributes.category_uri: ANY(\"laptops\")"]andpageCategories: ["laptops"]. - The response is a full search result with
searchResult.results[](each containingproduct,variantRollupValues,rating) plus facets and pagination (totalSize,nextPageToken). - Individual tile hydration additionally fires per-item
gdx-api.costco.com/catalog/product/product-api/v1/products/summary?items=...calls — skip those, everything you need is already in the search response.
Strategy used
- Phase 0 (curl): 403 on
/laptops.html. Akamai. Noted. - Phase 1 (framework): Next.js, but the initial HTML has no
__NEXT_DATA__product payload and no/_next/data/JSON routes for this sub-app. Skipped. - Phase 2 (browser recon only): Connected to real Chrome via
127.0.0.1:9222, navigated to/laptops.html, captured network. Foundgdx-api.costco.com/catalog/search/api/v1/searchwith the POST body and headers. Immediately abandoned the browser. - Phase 3: Replayed the API directly with
fetch. No cookies. No browser. ~1.7s for the whole category.
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.
- 2d ago First seen · 153 lines · 99 tokens per session scan B cd1e507152b7
costco is a skill published in the GitHub repository AgentComputerAI/torch (5 stars, last pushed 4mo ago), licensed MIT. It adds 99 tokens to every session and 2,376 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it B with 2 findings (sends data to an external url, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
tiny-web-crawler
Crawl from one or more starting web pages, fetch readable content, search within pages, follow relevant links, and stop when the requested information is found or a bounded limit is reached.
playwright-cli
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
computer-use
Read and drive native desktop applications through the accessibility layer — list on-screen apps, snapshot one window as a numbered element tree, then click / type / set a value / scroll / drag / run a named action, by element index or by screen coordinates. Use for work in a desktop app rather than a web page. Full…
feature-demo-recording
Record a demo video of a web feature from a real browser. Two modes -- a NARRATED film where measured voiceover drives the timeline (designed slides, subtitles, punch-in camera, rendered from an HTML timeline), and a SILENT evidence clip for a PR or a QA pass. Use when the user asks to record a video, demo, or screen…
browserstack
../../../engineering-team/playwright-pro/skills/browserstack/SKILL.md.
browser-recording
Record a browser flow as a video/GIF for evidence — animations, transitions, and multi-step interactions that a still screenshot cannot prove. Drives the project's own Playwright through a bundled runner, then converts to mp4 + GIF via ffmpeg. Use when the user asks to record a demo, capture a GIF or video of the UI…