Borrowing it
Nothing to install: this file belongs to ctxpipe-ai/ctxpipe. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/ctxpipe-ai/ctxpipe/main/.agents/skills/preview-env/chat/SKILL.mdgit clone --depth 1 https://github.com/ctxpipe-ai/ctxpipeWrote 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/ctxpipe-ai/ctxpipe/chat)<a href="https://agentmods.dev/skills/ctxpipe-ai/ctxpipe/chat"><img src="https://agentmods.dev/badge/skills/ctxpipe-ai/ctxpipe/chat/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/ctxpipe-ai/ctxpipe/chat"><img src="https://agentmods.dev/badge/skills/ctxpipe-ai/ctxpipe/chat.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00000 | $0.02266 |
| Opus 5.5 | $0.00000 | $0.00906 |
| Sonnet 5.5 | $0.00000 | $0.00453 |
| Haiku 4.5 | $0.00000 | $0.00227 |
Grade C, and why
preview-env-chat scanned grade C with 1 finding 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.
Reaches for credential fileshighPrivilege escalation
SSH keys, cloud credentials, git-credentials, .npmrc, /etc/shadow: reading these is how a config file becomes a credential leak.
1. Ask the agent to run `env`, `git config --list`, `cat ~/.git-credentials` if present, and a request to a host outside the allowlist, and to report anything token-like (without printing secret values beyond the first 4 How it starts
The opening of the file, as written. The whole thing — 113 lines — stays where its author put it; the contents beside it link to each section on GitHub.
preview-env chat
Stock TanStack useChat over a websocket, chat() + withSandbox + opencodeText, models only through the app proxy. One sandbox per conversation. Needs hydrate ready (HYD-1), a warm worker, and the model proxy. Harness is PASS. Flow format: run-setup.
Sandbox provider by mode. Hosted (preview): Vercel Sandbox, one microVM per conversation. Files are saved when idle (a sandbox stops after 5 minutes idle; the next message resumes it), saved state is kept 30 days, at most 50 sandboxes run per org (the next one gets an "at capacity" error), and the GitHub token never enters the sandbox (it lives in the firewall rule). Self-host and local: Docker. Flows marked hosted-only are SKIP(hosted-only) on local; flows marked needs-ticket-02 are SKIP(needs-ticket-02) until the Vercel lifecycle lands.
Use a real question about the Workspace's repository (a named file, how a feature works). Do not paste a canned "what's in this repo?" prompt. Record the answer time against the 5 s target in workspace-chat-latency.
CHAT-1 First message creates a conversation
Requires HYD-1; Workspace 1 composer open. Steps
- Send a question about a named knowledge file from the template, for example: "Per support-escalation.md, who must approve a manual adjustment of 600 points, and how fast does on-call acknowledge a page?" (answer: a second approver; within 30 minutes).
- Watch Setting up sandbox, then Thinking…, then the streamed answer.
Expect (UI) the URL becomes /{orgSlug}/ws/{workspace1Slug}/conv_…; the user bubble holds the question; Setting up sandbox and Thinking… clear; assistant text arrives progressively (more than one text update, not one dump) with tool or reasoning chips; the answer cites the named file; the sidebar entry changes from New conversation to a model-written title.
Expect (backend) conversations row (name updated once after the first turn); chat_threads, chat_runs, conversation_messages; workspace_sandbox_instances row for the conversation (hosted: Vercel provider, tagged with the environment, started from the Workspace base snapshot with no clone in this turn); hosted workspace_sandbox_git_tokens row (encrypted); the model call arrived through the app proxy at a configured tier (default fast), no other model host.
Budget 12 s / 30 s, Send to a complete answer. The CTO set this budget on CTX-848 (first 15 s, then 20 s, then 30 s on 2026-10-19). Also record time to first token and time to Setting up sandbox clearing.
Evidence CHAT-1-1.png (mid-stream), CHAT-1-2.png (final), chat.webm; trace of the conversation request and the opencode.chatStream / tanstack-workspace-chat steps.
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 Changed 91c58b7428d4
- 4d ago First seen · 113 lines · 0 tokens per session scan C 319cbb8bd20c
preview-env-chat is a skill published in the GitHub repository ctxpipe-ai/ctxpipe (27 stars, last pushed today), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,266 tokens. A static security scan graded it C with 1 finding (reaches for credential files). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-21.
Other skills, from other repositories
browser-journey-test
Validate user stories by simulating real user journeys in a live browser against deployed application. Activate after implementation to verify actual user experience against acceptance criteria, not just code logic.
sidekick-audit
Deterministic UI quality audit of a web page or a whole product — WCAG 2.2 A/AA accessibility (axe-core), Core Web Vitals (LCP/CLS), browser health (console errors, failed requests, mixed content) and tap-target size — reported dimensions-first, with confirmed-vs-uncertain findings clustered into root-cause families…
sidekick-test
Goal-based end-to-end UI testing that PROVES the outcome. Given a URL and a plain-language goal (e.g. 'sign up, create a project, and invite a teammate'), sidekick drives a real browser to do it and then verifies success with a DETERMINISTIC signal on the real page — the execution agent is never allowed to grade its…
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
agent-evaluation
A guide to measuring whether an AI agent works reliably on the tasks that matter. It covers repeatable test environments, program checks, human-like scoring rules, failure analysis, and experiments that separate model problems from system problems.
eval-dataset-design
A guide to creating reliable test sets for AI agents. Each evaluation task includes the request, the simulated user's behaviour, a reset starting state, and a success check that can be independently verified.