What is launch-sandboxed?

A launcher that starts an autonomy-loop terminal inside an operating-system sandbox, a boundary that limits where it can write and which network destinations it can reach. It refuses to continue if the required sandbox program is unavailable.

Use it to check for the sandbox runtime, generate its settings, launch an interactive or one-command session, and expose the sandbox status to the loop's safety checks.

It confines automated work to approved locations and limits web access, reducing the damage an agent or fetched content could cause outside the project.

launch-sandboxed — Command by inferencegod

Install

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.

agentmods

npx agentmods add commands/inferencegod/autonomy-loop/launch-sandboxed

Clone the repo

git clone --depth 1 https://github.com/inferencegod/autonomy-loop

Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 529 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. Scan, not verified.

Origin original No closer match found in the catalogue.

Token cost

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	$0.00029	$0.00529
Opus 5	$0.00015	$0.00264
Sonnet 5	$0.00006	$0.00106
Haiku 4.5	$0.00003	$0.00053

Measured yesterday against content hash 2721829e06b4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

launch-sandboxed 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 yesterday.

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.

autonomy-loop/commands/launch-sandboxed.md · 10 lines

What it actually says

Launch this terminal inside Anthropic's sandbox-runtime (srt) so writes are confined to the workspace and egress is allow-listed. This is the SHIPPED sandbox tier the preflight's sandboxLive() probe attests to; once inside, hooks/sandbox-detect.mjs reports the boundary and hooks/preflight.mjs can clear the sandboxNotLive refusal.

Preconditions. Confirm srt is installed (command -v srt; install with npm install -g @anthropic-ai/sandbox-runtime, and on Linux also bubblewrap socat ripgrep). If it is missing, STOP and tell me to install it. Do NOT fall back to an unsandboxed shell (fail-closed).
Launch. Run the plugin wrapper ${CLAUDE_PLUGIN_ROOT}/sandbox/launch-sandboxed (optionally -- <command> to run one command instead of an interactive shell). It reads autonomy.config.json, generates the srt settings via ${CLAUDE_PLUGIN_ROOT}/sandbox/srt-settings.gen.mjs into $HOME/.config/autonomy-loop/srt-settings.json (OUTSIDE the workspace, so the agent cannot disable its own boundary: the Ona finding), and execs srt --settings <that file> with AUTONOMY_SANDBOX=1 exported.
What the boundary is. allowWrite = the workspace; denyWrite = the control plane (autonomy.config.json + hooks/ + every protectedPaths entry), so a bypassed gate inside the box still cannot edit-away its own gates. allowedDomains = a tight egress allowlist (forge api/git/lfs + package registries + the model API, no bare github.com apex); this same allowlist is the ceiling on the researcher role's web-fetch. Widen ONLY via AUTONOMY_EXTRA_DOMAINS, never by hand-editing the generated file.
Alternatives. For a full container boundary use the plugin .devcontainer/ (default-deny iptables/ipset firewall). For untrusted work, a gVisor/Firecracker microVM is the documented opt-in user-infra tier (see sandbox/README.md). This command NEVER provisions the microVM tier; that is your infrastructure to stand up.

Changes

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.

yesterday First seen · 10 lines · 29 tokens per session scan A 2721829e06b4

Subscribe to this mod's changes

launch-sandboxed is a command published in the GitHub repository inferencegod/autonomy-loop (1 stars, last pushed 2mo ago), licensed MIT. It adds 29 tokens to every session and 529 once invoked, about $0.0001 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 2026-08-31.

Autonomous Agents Code review Productivity Research ai-agents automation claude-code plugin

fest-show

Show festival progression (in-progress tasks, roadmap, and dependency view).

Obedience-Corp/festival · 17 tokens

test

Run Postman collection tests, analyze results, diagnose failures, and suggest fixes.

Postman-Devrel/postman-claude-code-plugin · 16 tokens

review

Review proposed ADRs for lifecycle progression — accept implemented decisions, deprecate stale ones, archive trivial ones.

dork-labs/dorkos · 21 tokens

types

Debug and fix TypeScript type errors with systematic analysis and expert guidance.

dork-labs/dorkos · 13 tokens

routine

Execute one maintenance routine defined in scv/routines/ .md (task + guardrails + exit-criteria contract), or list defined routines. SCV never schedules — pair with host features like /loop or cron yourself. Use whenever the user asks to run a recurring maintenance task, or asks what routines exist — not only when…

wookiya1364/scv-claude-code · 75 tokens

incident

Live incident response — intake a Sentry issue, stack trace, or symptom; gate mitigate-first vs root-cause-first; keep a UTC timeline. Use when production is broken RIGHT NOW; for the after-action write-up alone, use /orc:postmortem.

HigorAlves/orc · 54 tokens