Skip to content

AGENTS.md

Latest commit

 

History

History
39 lines (27 loc) · 1.23 KB

AGENTS.md

File metadata and controls

39 lines (27 loc) · 1.23 KB
Raw

Agent Instructions

This repo is code-airlock: a small Bash wrapper around Docker Sandboxes for running coding agents in disposable microVMs.

Working Rules

  • Keep the tool dependency-light. Do not add package managers, build systems, or new runtime dependencies unless explicitly requested.
  • Prefer small, reviewable Bash changes in code-airlock.
  • Keep docs in README.md aligned with script behavior.
  • Do not change security defaults casually. Explain any change to network policy, credential handling, clone mode, or sandbox deletion behavior.
  • Do not push to GitHub or publish releases unless explicitly asked.

Verification

After editing the wrapper, run:

bash -n code-airlock
shellcheck code-airlock
./code-airlock help

For docs-only changes, at least check for stale command names:

rg -n "claude-sbx|code-airlock|sbx " README.md code-airlock sandbox.conf.example AGENTS.md

Sandbox Workflow

The agent works in Docker Sandboxes clone mode. Commit any finished edits inside the sandbox so the host can retrieve them with:

code-airlock fetch
code-airlock diff
code-airlock merge

If work is not committed in the sandbox, it will not come back through the normal fetch/diff/merge workflow.