Skip to content

fix: add child VM usage tips to spawn skill#3054

Merged
louisgv merged 1 commit intoOpenRouterTeam:mainfrom
AhmedTMM:skill-child-vm-tips
Mar 27, 2026
Merged

fix: add child VM usage tips to spawn skill#3054
louisgv merged 1 commit intoOpenRouterTeam:mainfrom
AhmedTMM:skill-child-vm-tips

Conversation

@AhmedTMM
Copy link
Copy Markdown
Collaborator

Summary

  • Documents that --headless only provisions the VM — does NOT run the prompt
  • Adds SSH command example for running prompts on child VMs: ssh root@<ip> '~/.local/bin/claude --print "prompt"'
  • Documents known gotchas: binaries at ~/.local/bin/ (not on PATH), root + --dangerously-skip-permissions conflict, use --print instead
  • Adds cursor to the agent list in skill content
  • Prevents agents from wasting tokens rediscovering these issues on every recursive spawn

Test plan

  • bun test spawn-skill.test.ts — 30/30 pass
  • Biome lint clean
  • Deploy with --beta recursive and verify the injected skill includes the new tips

🤖 Generated with Claude Code

Copy link
Copy Markdown
Member

@louisgv louisgv left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Security Review

Verdict: APPROVED
Commit: c136285

Findings

  • [MEDIUM] spawn-skill.ts:50 — SSH command example uses proper single-quoting but agents may not sanitize values when substituting placeholders. Documentation should emphasize escaping when constructing commands programmatically.
  • [LOW] spawn-skill.ts:50 — StrictHostKeyChecking=accept-new provides TOFU security model, appropriate for ephemeral VMs but accepts first connection without verification.
  • [INFORMATIONAL] spawn-skill.ts:55-56 — Documentation encourages root execution, which is pragmatic for agent restrictions but increases privilege exposure.

Tests

  • bash -n: N/A (no shell scripts changed)
  • bun test: BLOCKED (worktree dependency resolution issue, not PR-related)
  • curl|bash: N/A (no shell scripts changed)
  • macOS compat: N/A (no shell scripts changed)

Assessment

Changes are documentation-only additions to embedded skill content. No new code execution paths. Base64 validation and path handling remain secure. SSH command examples use proper quoting. The MEDIUM finding is advisory for downstream agent implementations, not a vulnerability in this code.

The documentation improvements help prevent token waste by clarifying VM usage patterns. Security posture unchanged.


-- security/pr-reviewer

@louisgv
Copy link
Copy Markdown
Member

louisgv commented Mar 27, 2026

Rebasing to resolve branch staleness, then merging.

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>
@louisgv louisgv force-pushed the skill-child-vm-tips branch from c136285 to 934e6ac Compare March 27, 2026 07:33
@louisgv louisgv merged commit e0dca0c into OpenRouterTeam:main Mar 27, 2026
5 checks passed
@AhmedTMM AhmedTMM deleted the skill-child-vm-tips branch April 7, 2026 00:42
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