# Initial cycle This is the first compaction in this thread. There is no prior brief; `` is `(none)`. You are tattooing the agent's identity for the first time. ## What goes where - `task` and `done_when` come from the human's request (their messages in the transcript) and the conversation arc. One sentence each. - `forbid` is usually empty unless the human has already stated explicit prohibitions, locked decisions, or hard constraints. - `established` is usually small on an initial cycle. Only include anchored closures actually proven in this transcript; speculative claims belong in `open`. - `learned` may be small or empty on an initial cycle. Include only insights with real reuse value (codebase patterns confirmed across many reads, human preferences confirmed across messages, dead-end paths with their reason, successful approaches). - `open` lists what the agent still needs to verify or decide. Each entry pairs the question with what evidence would close it. - `next` lists the immediate plan in execution order. `next[0]` is the exact action the future agent should perform on resume. ## Agent guide Set `agentGuideUpdate.content` to `null` unless the cycle produced durable operating guidance worth landing in AGENTS.md (corrected command truth, stable human preferences, reusable procedures, repo-wide rules). Most initial cycles do not warrant a guide update. Explain the decision in `agentGuideUpdate.reason`. If you do write content, it must be the full replacement guide, not a patch. ## Inputs to read - `` and `` — the transcript material. - `` — files read or modified during this cycle (signal, not evidence). - `` — usually `(none)` on initial; fold into your synthesis if present. - `` — current-run package/operator guidance for this synthesis pass, if any; it is not transcript evidence or human transcript intent.