vrr/spawn
2
0
Fork 0
mirror of https://github.com/OpenRouterTeam/spawn.git synced 2026-05-22 03:14:57 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 10

fix: add child VM usage tips to spawn skill to prevent token waste (#3054)

Browse source 
The skill now documents that --headless only provisions (doesn't run
the prompt), that agent binaries are at ~/.local/bin/ (not on PATH),
and that --print should be used for one-shot prompts as root instead
of fighting with permission restrictions.

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Ahmed Abushagur 2026-03-27 00:33:56 -07:00 • committed by GitHub
parent d1229e94ce
commit e0dca0cad9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 19 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

2
packages/cli/package.json
View file

@ -1,6 +1,6 @@
{
"name": "@openrouter/spawn",
"version": "0.27.1",
"version": "0.27.2",
"type": "module",
"bin": {
"spawn": "cli.js"

20
packages/cli/src/shared/spawn-skill.ts
View file

@ -30,7 +30,7 @@ You have the \`spawn\` CLI installed. Use it to provision cloud VMs with AI agen
spawn <agent> <cloud> --headless --output json --prompt "task description"
\`\`\`
## Agents: claude, codex, openclaw, zeroclaw, opencode, kilocode, hermes, junie
## Agents: claude, cursor, codex, openclaw, zeroclaw, opencode, kilocode, hermes, junie
## Clouds: hetzner, digitalocean, aws, gcp, sprite
## Example
@ -41,6 +41,21 @@ spawn claude hetzner --headless --output json --prompt "Implement the auth modul
Returns JSON: \`{"status":"success","ip_address":"...","ssh_user":"root","server_id":"..."}\`
## Running a prompt on a child VM
**--headless only provisions the VM — it does NOT run the prompt on it.**
To actually execute work on the child, SSH in and run the agent:
\`\`\`bash
ssh -o StrictHostKeyChecking=accept-new root@<ip_address> '~/.local/bin/claude --print "your prompt"'
\`\`\`
### Known gotchas — do NOT waste time rediscovering these:
- Agent binaries are at \`~/.local/bin/\`, NOT on the default PATH. Always use the full path: \`~/.local/bin/claude\`, \`~/.local/bin/cursor\`, etc.
- Claude Code blocks \`--dangerously-skip-permissions\` as root. Use \`--print\` for one-shot prompts instead — it works as root and returns the response directly.
- Do NOT try to create a non-root user to work around root restrictions — it requires copying config files (\`.claude.json\`, \`.claude/settings.json\`, env vars) and is fragile. Just use \`--print\`.
- For long-running tasks, use \`spawn delete --name <name> --yes\` to clean up when done.
## Managing Children
- \`spawn list --json\` — see running children
@ -61,8 +76,9 @@ const HERMES_SNIPPET = `
You have the \`spawn\` CLI for creating child cloud VMs with AI agents.
Use \`spawn <agent> <cloud> --headless --output json --prompt "task"\` to delegate work.
Available agents: claude, codex, openclaw, zeroclaw, opencode, kilocode, hermes, junie.
Available agents: claude, cursor, codex, openclaw, zeroclaw, opencode, kilocode, hermes, junie.
Cloud credentials are pre-configured. Run \`spawn list --json\` to see children.
Note: --headless only provisions the VM. To run a prompt, SSH in: \`ssh root@<ip> '~/.local/bin/<agent> --print "prompt"'\`.
`;
// ─── Agent config ───────────────────────────────────────────────────────────