Operations

Overnight Agent Instructions#

Use this when asking agents to build overnight.

Mission#

Build the Co-Vibe customer-readiness slice described in:

  • docs/product/problem.md
  • docs/ops/build-loop-target.md
  • docs/ops/build-loop-tests.md
  • docs/ops/build-loop-done.md
  • docs/engineering/repo-rules.md
  • docs/engineering/architecture.md
  • docs/engineering/data-model.md
  • docs/engineering/mcp-contract.md
  • docs/engineering/auth-plan.md
  • docs/engineering/security-checks.md
  • docs/engineering/stack-decision.md

Non-Negotiables#

  • no file over 300 lines
  • no loose Markdown as source of truth
  • no deferred feature code
  • no secrets in git
  • docs stay current
  • auth must be real or clearly called out as incomplete
  • MCP tools must return warnings and blocks to agents
  • tests must prove agent logging and feedback
  • security checks must run
  • browser QA must run if UI exists
  • Python may be used where it helps, but scripts need docs and tests

Build Goal#

A user can:

  1. start the app
  2. log in
  3. create an MCP token
  4. connect an agent
  5. log a task through MCP
  6. see the task in Workstreams
  7. trigger duplicate-work warning
  8. see blocked start without confirmation
  9. continue with confirmation
  10. see events in Activity
  11. generate a weekly summary

Before connecting agents, run npm run setup on a fresh clone, or rerun npm run doctor later to verify local readiness without reinitializing setup.

During long runs, agents should periodically call covibe_team with operation: "audit_parallel_work", their token, and session_id. Treat a warning as a stop-and-review signal before continuing parallel implementation.

Agents should also call covibe_team with operation: "state" before resuming work. If it returns parallel_audit_conflicts, review those active conflicts before continuing.

Required Output#

Write reports/morning-report.md.

Include:

  • stack chosen
  • whether Python was used and why
  • auth provider chosen
  • what was built
  • what tests passed
  • what tests failed
  • security check results
  • browser QA result
  • doctor check result
  • files changed
  • known limitations
  • first thing a human should try

Stop Conditions#

Stop and report clearly if:

  • auth cannot be wired
  • MCP SDK setup blocks progress
  • app cannot start
  • tests cannot run
  • security checks fail badly
  • file-size rule cannot be satisfied

Do not pretend done.

View as .md