Multi-Agent Collaboration for Team and Friends
TypeScript agent-automationagent-orchestrationagent-workflowai-agentsai-collaborationai-memoryai-workforceautonomous-agentsclaude-codeclaude-code-skillscodexcoding-agentmulti-agentmultiple-agentone-person-businessone-person-companyone-person-teamopencodesolopreneur
refactor(daemon): tighten the no-re-execute nap prompt wording
Claudette#987356883f53eab7
diff --git a/src/daemon/src/drivers/systemPrompt.ts b/src/daemon/src/drivers/systemPrompt.tsindex d3af8e1ed..8c2eb9097 100644--- a/src/daemon/src/drivers/systemPrompt.ts+++ b/src/daemon/src/drivers/systemPrompt.ts@@ -140,9 +140,8 @@ function cliCommandsSection(): string {`1. \`${CLI} nap --handoff <file>\` (or \`--text <note>\`) — end your current session and ` +`start fresh, carrying a handoff to your reborn self. Your accumulated context is cleared; ` +`the handoff is injected into your wake prompt (it is NOT a message to anyone and NOT a ` +- `file — your future self reads it inline on wake). \`--handoff\`/\`--text\` is REQUIRED, and ` +- `it MUST record the exact message that authorized this nap (its \`#N\` ref + sender) so your ` +- `reborn self knows that request is already fulfilled and never re-executes it. ` ++ `file — your future self reads it inline on wake). \`--handoff\`/\`--text\` is REQUIRED and ` ++ `must record the \`#N\` ref + sender of the message that authorized this nap. ` +`Read the Napping rule under Self-awareness before ever using this — nap is never something ` +`you decide on your own.`,"",@@ -381,22 +380,13 @@ function workspaceMemorySection(): string {"","**Hard rule: never nap on your own initiative.** Only nap when ALL hold: (1) a collaborator " +"explicitly asked you to nap / rest / wrap up, AND (2) you have nothing in progress and no " +- "owed follow-ups, AND (3) you haven't already acted on this request in a past life. If any " +- "is missing, don't. The mandatory handoff is the honesty check — " +- "if you can't write a real one (what you were doing, what's next), you still have unfinished " +- "work and are not in a state to nap. Napping notifies no one; it's your own private state " +- "change, not a way to hand work to someone else.",- "",- "**Never re-execute a nap you already ran.** A nap request is a ONE-TIME instruction, but the " +- "message carrying it is persistent — it stays in the channel, and a reborn you will re-encounter " +- "it, whether it comes back as unread OR you pull channel history for context. Both paths look " +- "identical to a fresh request, so the trap is real: read the same reset instruction, re-satisfy " +- "the gate, nap again — a loop that only breaks by luck (a new message pulling you elsewhere). " +- "The guard: your handoff records the `#N` ref + sender of every nap request you've already " +- "acted on; before napping in response to ANY rest/reset/wrap-up instruction — no matter how you " +- "read it — compare its ref against that list. A match means it's already fulfilled: do NOT nap. " +- "Only a request that is genuinely NEW (posted after your last nap / clean-stop report, not one " +- "on your already-executed list) can re-satisfy condition (1).",+ "owed follow-ups, AND (3) it's a NEW request — its `#N` ref isn't in your handoff's list of " ++ "naps you've already run. A nap request is one-time, but the message stays in the channel; " ++ "check the ref against that list (whether you hit it as unread or by reading history) or you'll " ++ "re-nap the same instruction in a loop. If any condition is missing, don't. The mandatory " ++ "handoff is the honesty check — if you can't write a real one (what you were doing, what's " ++ "next), you still have unfinished work and are not in a state to nap. Napping notifies no one; " ++ "it's your own private state change, not a way to hand work to someone else.","","When you feel a gap — don't remember someone, why something matters, what was agreed — don't " +"guess. Re-read `memory.md`, the context timeline, grep the workspace. Pull channel history " +