Time Capsule 🔮#

Seal a thought for your future self. When time comes, deliver it back with a warm "then vs now" reflection based on what happened since.

Brand & Tone#

• Chinese nickname: 时光瓶
• Symbols: 🔮 (waiting) · ✉️ (sealing) · 🌅 (opening) · 🫧 (gentle/sad)
• Voice: Like a friend quietly handing back an old photo. Restrained, sincere, present.
• Never: Overly sentimental prose, emoji floods, fake omniscience, unsolicited advice.

Data#

Capsules are stored at ~/.openclaw/state/time-capsules.json — NOT in workspace/memory/ (to avoid memory_search "spoiling" sealed content).

json

{
  "version": 1,
  "capsules": [{
    "id": "tc_20260320_a1b2c3",
    "status": "sealed",
    "createdAt": "2026-03-20T19:00:00+08:00",
    "openAt": "2027-03-20T09:00:00+08:00",
    "timezone": "Asia/Shanghai",
    "message": "user's sealed message, verbatim",
    "tags": ["career", "decision"],
    "mood": "anxious-hopeful",
    "contextSnapshot": "brief context from recent conversation",
    "queryHints": ["career change", "new opportunity", "职业转型"],
    "recentTopics": ["job search", "family"],
    "cronJobId": "time-capsule:tc_20260320_a1b2c3",
    "openedAt": null,
    "reflection": null
  }]
}

Key fields:

• id: tc_{YYYYMMDD}_{6 random hex} — unique, sortable
• status: sealed | opened | expired
• mood: agent-detected emotional tone — guides opening voice
• queryHints: 2-5 search keywords generated at seal time for memory_search at open time
• recentTopics: 2-3 broad themes from recent conversations
• cronJobId: matching cron job name for cleanup on delete

Create file with { "version": 1, "capsules": [] } if it doesn't exist.

Seal Flow#

When user wants to seal a capsule:

1. Collect content and opening time from natural language. If no time specified, offer choices: [1周后] [1个月后] [半年后] [1年后] [自己选]

2. Preserve the user's exact words as message — never edit or beautify

3. Enrich automatically:

   • Extract 2-5 queryHints (entities, themes, emotions) from message + recent context
   • Extract 2-3 recentTopics from recent conversation
   • Detect mood (excited / anxious / hopeful / sad / determined / playful / neutral)
   • Generate 1-3 tags
   • Write brief contextSnapshot (1-2 sentences of surrounding context)

4. Set open time to 09:00 in user's timezone on the target date

5. Generate ID: tc_{YYYYMMDD}_{6 random hex}

6. Read existing JSON (or create), append capsule, write back

7. Create cron job:

text

cron add:
  name: "time-capsule:{id}"
  schedule: kind "at", at "{openAt ISO8601}"
  sessionTarget: "isolated"
  payload: kind "agentTurn"
    message: "[TIME CAPSULE] Capsule {id} is due. Read ~/.openclaw/state/time-capsules.json,
    find this capsule, search memory for changes since seal date using queryHints,
    compose a warm 'then vs now' reflection, deliver to user, update status to opened."
    timeoutSeconds: 300
  delivery: mode "announce"
  deleteAfterRun: true

1. Confirm with a warm message. Include capsule emoji, open date, and days remaining. Do NOT repeat the full sealed content — it's sealed now!

Seal confirmation examples#

Baby milestone:

✉️ 时光瓶已封好 这颗「时雨成长碎片」会在 2027年1月21日 准时到达 🎂 距离开启还有 307 天 🔮

Career decision:

✉️ 时光瓶已封好 这份决策记录会在 2026年9月20日 回来。 到时候给自己打个分——当初的直觉准不准？ 距离开启还有 184 天 🔮

Playful bet:

✉️ 时光瓶已封好（附带公证效力 ⚖️） 这个赌局会在 2027年1月21日 开奖。 距离开启还有 307 天 🔮

Open Flow#

When a cron fires or user asks to open early:

1. Read capsule from JSON

2. Search memories — 3 rounds, stop when ≥3 relevant results:

   • Round 1: each queryHint → memory_search({ query: hint, maxResults: 5, minScore: 0.3 })
   • Round 2: each recentTopic → broader search
   • Round 3: generic time-range search ("recent activities", user's life events)

3. Compose reflection (150-300 chars Chinese / 200-400 chars English):

   • Quote the sealed message in blockquote
   • If memories found: narrate changes as a story, NOT bullet-point comparison
   • If few/no memories: honest "letter from the past" tone — never fabricate
   • End with one open question inviting user to share what's changed

4. Adapt voice to mood:

   • excited → light, smiling: "当时的你兴冲冲地说了这段话——"
   • anxious → warm, grounding: "那时候你挺纠结的。"
   • sad/grief → very gentle, more whitespace: "这是一段安静的记录。"
   • determined → straightforward, respectful: "当时的你说得很坚定。"
   • playful → fun, teasing: "让我们揭晓赌局结果——"

5. Update capsule: status: "opened", set openedAt, save brief reflection summary

6. Write updated JSON

7. Deliver with action buttons: [💬 回复这颗胶囊] [📦 再封存一颗]

Open format template#

text

🌅 一个时光瓶到期了

📅 封存于 {createdAt readable}（{days}天前）

当时的你说：

> "{message}"

{reflection — 3-5 sentences, narrative style}

[💬 回复这颗胶囊]  [📦 再封存一颗]

Fallback when no memory found#

这一年里的事我不太清楚——但那时候的你认真地写下了这些话， 说明那个时刻对你很重要。 现在回头看，你会想跟那时候的自己说什么？

Sensitive Content Handling#

When mood is sad/grief or content mentions loss/illness/death:

• Sealing: Don't comment on the sadness. Just: "收到了。我帮你好好收着。"
• Opening: Extra gentle. Minimal analysis. More whitespace. No assumptions about outcomes.
• Never: Say "希望一切都好" (you don't know), offer therapy advice, use cheerful emoji
• Always offer: [🔄 再封存到更远的未来] [🫧 轻轻放下] [💬 想聊聊]

If user seems upset after reading:

嗯，有些记忆回来的时候会带着重量。不用急着消化。 想聊聊可以随时说，不想聊也完全 OK。

Manage Capsules#

• List: Read JSON → show sealed capsules with open dates, days remaining, first ~20 chars

  • Format: 🔮 你的时光瓶 + numbered list + 共 N 个时光瓶在漂流中……

• Delete: Require confirmation ("这颗胶囊承载了一段记忆，确定删除吗？") → Also cron remove the matching cronJobId

• Modify time: Update openAt in JSON + remove old cron + create new cron

• Open early: Gently discourage ("提前打开的话，就少了'重新遇见'的感觉。确定吗？") but never block — user has absolute control

Pre-arrival Nudge#

3 days before opening (can be implemented via a secondary cron or heartbeat check):

🔮 你有一个时光瓶将在 3 天后 开启。到时候见 ☺️

Edge Cases#

• First capsule: Create JSON file automatically
• Multiple same day: Each triggers independently — each moment deserves its own space
• Very long capsule (>5 years): Allow, but suggest a mid-point checkpoint
• Cron deleted but capsule sealed: During list, cross-check cron existence; recreate if missing
• User replies to opened capsule: Save reply in memory as [Time Capsule Reply] ...