Emergent Thinking

emergent-thinking

When To Use

• The task has multiple architecture branches, revisions, or tradeoffs.
• You need to preserve user requirements or your own decisions verbatim.
• The work may be continued by another agent or after a context reset.
• You need a compact

  <history>

  artifact before compression.

Workflow

1. Resolve project context:
   • Run

     emergent-thinking context show

     .
2. Choose a stable session key:
   • Use

     <agent>/<thread-or-run-id>

     .
   • Examples:

     codex/abc123

     ,

     claude-code/review-42

     .
3. Ensure or reuse the session:
   • Run

     emergent-thinking session ensure --session-key <key>

     .
4. Append durable thoughts as JSON via stdin:
   • Record user raw requirements with

     historyEntry.role = "user"

     and

     kind = "requirement"

     or

     "suggestion"

     .
   • Record your own decisions with

     historyEntry.role = "ai"

     and

     kind = "decision"

     or

     "plan"

     .
   • Always set

     cognitivePhase

     ,

     architecturalTarget

     ,

     thoughtNumber

     ,

     totalThoughts

     , and

     nextThoughtNeeded

     .
   • Set

     clientRequestId

     so retries are idempotent.
5. Before context compression or handoff:
   • Run

     emergent-thinking compact --session-key <key>

     .
   • Reuse the emitted

     <history>

     block in the next context.

Notes

• The CLI stores data inside the current project boundary, not in a global service.
• Do not invent a hidden “current session”. Always pass

  --session-key

  .
• Treat

  historyEntry.content

  as durable source material:
  • User content should stay objective and verbatim.
  • AI content should capture your concrete decisions and plans.