Skip to content

Add environment-classes section to Sandboxes from TokenJam explainer - #59

Merged
opencolin merged 1 commit into
mainfrom
claude/sandboxes-environment-classes
Jul 14, 2026
Merged

Add environment-classes section to Sandboxes from TokenJam explainer#59
opencolin merged 1 commit into
mainfrom
claude/sandboxes-environment-classes

Conversation

@opencolin

Copy link
Copy Markdown
Owner

Summary

Integrates TokenJam's "What are agent environments and sandboxes?" (Anil Murty, May 2026) into content/sandboxes.md.

New section: Environment Classes — Beyond Code Sandboxes

The page previously covered only code-execution sandboxes. The new section (placed right after "Why Sandboxes Matter") adds the other three classes from the article's taxonomy:

Class Representative infra New to the site?
Code sandboxes E2B, Modal already covered
Browser sandboxes Browserbase (+ Stagehand) ✅ first mention
Virtual desktops Computer Use, Operator ✅ first structured treatment
Simulation environments HUD, Inspect AI Sandboxing Toolkit ✅ first mention of HUD

Load-bearing ideas integrated

  • Environments and evaluation share infrastructure — same sandbox, deterministic data (fixed test cases, site snapshots, controlled time), full instrumentation. Framed as a vendor-selection criterion: "can I replay a production run as an eval." Cross-linked to Evals.
  • "Failures of grounding, not intent" — added to the Why Sandboxes Matter opening, plus the argument that read-only agents still need sandboxes (no-egress + capped budgets as "cheap insurance against failure modes you haven't thought of yet").
  • Virtual-desktop trade-offs — slower screenshot reasoning, higher visual-hallucination rates, UI drift breaking deterministic evaluation; "right tool when no API exists, wrong tool when one does."
  • Per-step isolation economics — Firecracker ~125ms boot / 3–10MB memory; fresh VM per run; "AWS Lambda's security model applied to agents"; with the isolation-is-a-spectrum caveat for regulated workloads.

Files

  • content/sandboxes.md — new section, updated index, grounding framing, Further Reading entry
  • content/changelog.md — entry under 2026-07-09
  • index.html — rebuilt

Test plan

  • New section renders between "Why Sandboxes Matter" and "Market Structure"
  • Index line includes Environment Classes anchor and it resolves
  • Evals cross-links resolve
  • TokenJam attribution links work

🤖 Generated with Claude Code

https://claude.ai/code/session_01C9gzSkgNa3yNtob4gBeeit


Generated by Claude Code

Integrates tokenjam.dev's "What are agent environments and
sandboxes?" (Anil Murty, May 2026) into sandboxes.md.

New "Environment Classes — Beyond Code Sandboxes" section covering
the three classes the page didn't address:
- Browser sandboxes (Browserbase + Stagehand; own profile/cookie
  jar per session)
- Virtual desktops (Computer Use / Operator) with the trade-offs:
  slower screenshot reasoning, higher visual-hallucination rate,
  UI drift breaking deterministic eval — "right tool when no API
  exists, wrong tool when one does"
- Simulation environments (HUD, Inspect AI Sandboxing Toolkit)

Two load-bearing ideas integrated:
- Environments and evaluation share infrastructure: same sandbox,
  deterministic data, full instrumentation — "can I replay a
  production run as an eval" is now framed as a vendor-selection
  criterion, cross-linked to evals.md
- "Failures of grounding, not intent" added to Why Sandboxes
  Matter, plus the read-only-agents-still-need-sandboxes argument

Also: Firecracker per-step isolation numbers (~125ms, 3-10MB),
"AWS Lambda's security model applied to agents," and the
isolation-is-a-spectrum caveat. Index and Further Reading updated;
changelog entry added.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C9gzSkgNa3yNtob4gBeeit
@vercel

vercel Bot commented Jul 14, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
agentic-engineering Ready Ready Preview, Comment Jul 14, 2026 9:46pm

@opencolin
opencolin merged commit 21e3e11 into main Jul 14, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants