Skip to main content

Knowledge > Runbooks > Agent Operations

Agent Operations Runbooks

Runbooks for Claude Code agents working across the ChurchWiseAI portfolio. These cover how to start tasks correctly, maintain knowledge, work safely in a multi-agent git environment, and meet quality standards before shipping.

Runbooks

RunbookDescription
how-claude-agents-work.mdOverview of how Claude Code agents operate in this system
starting-new-task.mdHow to start a new agent task correctly
knowledge-maintenance.mdHow to keep knowledge docs current when code changes
multi-agent-git.mdGit safety rules for concurrent multi-agent workflows
deploy-checklist.mdPre-deployment checklist for all code changes
qa-checklist-runbook.mdHow to run the QA checklist like a senior engineer
decision-logging.mdHow and when to update the decision log
feature-registry-maintenance.mdHow to use and maintain the feature registry

The Non-Negotiables

Every agent task, without exception, must:

  1. Read CLAUDE.md firstC:\dev\CLAUDE.md (cross-project) + the relevant codebase's own CLAUDE.md
  2. Read AGENT_QUALITY_PRINCIPLES.md before writing any code — C:\dev\AGENT_QUALITY_PRINCIPLES.md
  3. Run through QA_CHECKLIST.md before declaring work complete — C:\dev\QA_CHECKLIST.md
  4. Update the decision log after any significant decision — C:\dev\DECISION_LOG.md
  5. Update knowledge docs if code changes reference them — C:\dev\knowledge\

Trust Hierarchy for Information

  1. C:\dev\knowledge\ — canonical, always trust
  2. Per-codebase CLAUDE.md — trust for code-level context
  3. C:\dev\knowledge\drafts\ — proposals only, verify before using
  4. Other .md files — check if a canonical version exists in knowledge/ first
  5. C:\dev\knowledge\archive\ — do NOT use for current decisions