Skip to content

AGENT_CRITICAL_GUARDRAILS.md

Latest commit

 

History

History
18 lines (16 loc) · 2.18 KB

AGENT_CRITICAL_GUARDRAILS.md

File metadata and controls

18 lines (16 loc) · 2.18 KB
Raw

Agent Critical Guardrails

These are the hard rules to keep agent work predictable and safe in this repo.

  • Read CLAUDE.md first, then follow docs/FEATURE_ARCHITECTURE_STANDARD.md for new medium and large features.
  • Use pnpm for project commands. Do not switch to npm or yarn.
  • Use the desktop Electron app (pnpm dev) for normal local development and smoke checks unless browser-mode internals are explicitly requested.
  • Do not test agent teams, launch/provisioning, terminal runtime, task assignment, smoke-flow, or agent actions on real user projects. Use only new sandbox/test projects or explicitly test-only existing projects. Real projects such as ~/dev/projects/ai/claude-runtime must not be used even for opening a runtime/terminal without fresh direct user permission.
  • Do not run pnpm lint:fix unless the user explicitly asks for broad formatting changes.
  • Keep new production source files at or below 800 physical lines. Existing oversized files are frozen in scripts/ci/source-file-size-baseline.json; do not add exceptions or raise legacy caps. Run pnpm guard:source-file-size to verify the ratchet.
  • For Team Provisioning, do not add facade-inheritance layers, whole-service ServiceHost casts, or new protected abstract dependency slots. Use explicit composition, follow docs/team-management/team-provisioning-target-architecture.md, and run pnpm guard:team-provisioning-architecture.
  • Keep main, preload, renderer, and shared responsibilities separate.
  • Build interactive UI controls from reusable Radix UI headless primitives under src/renderer/components/ui when a shared primitive exists. Do not add one-off native or hand-rolled controls for selects, dialogs, popovers, tabs, menus, tooltips, switches, or checkboxes.
  • Use wrapAgentBlock(text) instead of manually concatenating agent block markers.
  • Preserve task/subagent filtering, structured task refs, and message parsing semantics.
  • Validate IPC and other main-process inputs defensively and fail gracefully.
  • Treat docs/team-management/debugging-agent-teams.md as the first stop for team launch hangs, bootstrap issues, or missing teammate replies.
  • Do not revert unrelated user changes or other agents' edits.