The state management server is not running. Verify installation and restart:

If the server is missing entirely, reinstall with npx pokayokay.

The work command requires an existing plan. Run /pokayokay:plan first with a PRD path. If you already planned, check that the state file exists:

Chaining requires the ohno MCP server to persist state. Verify it is running and that the .ohno/ directory is not in your .gitignore. Also check that you are using a work mode that supports chaining (semi-auto, auto, or unattended).

Individual agents have their own context windows. If a single task is too large for an agent's context, try breaking it into smaller tasks with /pokayokay:revise. For very large codebases, the yokay-explorer agent (running on Haiku) is optimized for fast, focused searches rather than full-context loading.

If the two-stage review pipeline consistently rejects implementations, check your task descriptions for clarity. Ambiguous tasks often produce implementations that do not match the spec. Use /pokayokay:revise --direct to add more detail to task descriptions. If using Kaizen, check .kaizen/rules.json for rules that may be too strict.

pokayokay dispatches multiple agents per task (implementer + reviewers). If performance is a concern:

• Use -n N to limit tasks per session
• Switch agents that do not need Sonnet-level reasoning to Haiku in your .claude/agents/ config
• Reduce the scope of audit checks for development iterations