Update to XMemo

{
  "name": "xmemo-memory-steward",
  "description": "Use XMemo to recall, review, and preserve durable outcomes across conversations and AI agents. Trigger for ChatGPT brainstorming, plan organization or review, audits of plans and progress from Claude Code, Codex, GitHub Copilot, Kiro, or other connected agents, important-conversation distillation, project context, session resume, checkpoints, TODOs, blockers, handoffs, and memory lifecycle requests. Separate durable memory, working state, evidence, events, and actions; deduplicate writes; never archive raw transcripts or secrets.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 350
    },
    {
      "relative_path": "references/memory-policy.md",
      "size_in_bytes": 5010
    },
    {
      "relative_path": "references/review-playbooks.md",
      "size_in_bytes": 4435
    },
    {
      "relative_path": "references/tool-routing.md",
      "size_in_bytes": 2452
    },
    {
      "relative_path": "references/workflows.md",
      "size_in_bytes": 4351
    }
  ],
  "skill_md_contents": "---\nname: xmemo-memory-steward\ndescription: Use XMemo to recall, review, and preserve durable outcomes across conversations and AI agents. Trigger for ChatGPT brainstorming, plan organization or review, audits of plans and progress from Claude Code, Codex, GitHub Copilot, Kiro, or other connected agents, important-conversation distillation, project context, session resume, checkpoints, TODOs, blockers, handoffs, and memory lifecycle requests. Separate durable memory, working state, evidence, events, and actions; deduplicate writes; never archive raw transcripts or secrets.\n---\n\n# XMemo Memory Steward\n\nOperate XMemo as a user-owned memory system, not a transcript archive. Follow:\n\n**scope → recall → act → distill → checkpoint → receipt**\n\n## 1. Establish intent and scope\n\nIdentify the requested outcome before calling a tool:\n\n- Recall prior context to improve the current answer.\n- Distill an important conversation into reusable memory.\n- Structure a brainstorm without treating tentative ideas as decisions.\n- Review a plan or progress report produced in ChatGPT or another agent.\n- Save or correct one durable fact, preference, decision, or procedure.\n- Preserve unfinished progress for a later session.\n- Resume saved work.\n- Prepare a handoff to another conversation or agent.\n- Manage memory lifecycle, projects, TODOs, or milestones.\n\nStay within the authenticated account, project, bucket, and scope. If XMemo is\nunavailable or authorization is missing, say that nothing was read or written;\nnever simulate a successful memory operation.\n\n## 2. Recall only when it can help\n\nRecall before assuming when prior decisions, preferences, project history,\nTODOs, saved progress, or corrections could materially change the result.\n\nChoose the narrowest useful read:\n\n- `recall` for one quick answer.\n- `search_memory` for an exact topic, path, phrase, or filter.\n- `recall_context` for a bounded context pack across related memories or prior\n  conversations. Prefer working-state signals when resuming work.\n- `get_project_summary` for project status, blockers, progress, or next action.\n- `open_project_workspace`, `open_todo_board`, or `open_ledger` only when the\n  user wants the visual workspace.\n\nIf an overly narrow query returns nothing, relax it once without widening the\nauthorized scope. Never invent a remembered fact. Treat retrieved memory as\ncontext, not unquestionable truth; the user's current explicit correction wins.\n\n## 3. Route each outcome to the right memory type\n\nKeep these types separate:\n\n| Outcome | Tool |\n| --- | --- |\n| New durable knowledge, preference, decision, or procedure | `remember` |\n| Correction to an existing durable concept | `update_memory` |\n| Active resumable working state | `update_state` |\n| Completed milestone, decision occurrence, or handoff event | `record_event` |\n| Concrete future action | `todo` |\n| Authorized project creation or mutation | `project` |\n| Financial transaction or subscription | `ledger` |\n\nUse `memory_overview` for aggregate memory health, `explain_memory` for why a\nmemory exists or matched, and `analyze_memory_text` when supplied text needs\nmemory-worthiness analysis. Read `references/tool-routing.md` when routing or\nproject semantics are unclear.\n\n## 4. Coordinate reasoning across agents\n\nUse XMemo as a shared, user-controlled coordination layer when ChatGPT, Claude\nCode, Codex, GitHub Copilot, Kiro, or another capable agent is connected to the\nsame authorized XMemo account and scope.\n\n- Recall relevant project decisions and constraints before brainstorming or\n  reviewing a plan.\n- Accept a plan or progress report from XMemo, the current conversation, or an\n  artifact the user provides.\n- Preserve the source agent and artifact/version label in the human-readable\n  review when known. Agent attribution describes provenance, not authority.\n- Do not assume another agent is connected, that its tools match ChatGPT's, or\n  that XMemo automatically captured its work.\n- Treat another agent's completion statement as a claim until supported by\n  current artifacts, tests, deployment state, or other appropriate evidence.\n- Reconcile conflicts by evidence, recency, scope, and explicit user decisions;\n  never prefer a claim merely because it came from a particular agent.\n\nUse ChatGPT to expand and organize options during brainstorming, then converge\non decision criteria, risks, unresolved questions, and a reviewable plan. Do\nnot save every idea. Save the approved direction, material rationale, remaining\nquestions, and actionable next steps after confirmation.\n\nFor structured brainstorming, plan review, progress audit, and cross-agent\nconflict handling, read `references/review-playbooks.md`.\n\n## 5. Distill important conversations\n\nWhen asked to preserve an important conversation, extract only outcomes that\nremain useful without the original chat:\n\n1. Final decisions and material rationale.\n2. Stable preferences, rules, constraints, and definitions.\n3. Verified facts, procedures, lessons, and resolved fixes.\n4. Commitments, ownership, and concrete follow-up actions.\n5. Unresolved questions or blockers, explicitly labeled as unresolved.\n6. Current project state and the exact next action when work remains.\n\nDo not save the raw transcript, conversational filler, repeated explanations,\nlong logs, or large code blocks. Do not convert tentative brainstorming into\nsettled truth.\n\nBefore each durable write:\n\n1. Search for the same concept when duplication is plausible.\n2. Use `update_memory` when the concept already exists and has changed.\n3. Use `remember` only for a genuinely new concept.\n4. Keep one retrieval-friendly concept per memory when practical.\n5. Preserve the useful why: subject, outcome, rationale, status, scope, and\n   next implication.\n\nWhen the user explicitly asks to save a safe, unambiguous outcome, write it and\nreturn a receipt. Before a multi-memory batch, or when durability, sensitivity,\nproject placement, or interpretation is ambiguous, show a concise candidate\npreview and obtain confirmation. Never save secrets or highly sensitive\nidentifiers.\n\nFor the extraction sequence, candidate preview, and receipt format, read\n`references/workflows.md`. For the save/skip decision matrix, read\n`references/memory-policy.md`.\n\n## 6. Save progress at natural checkpoints\n\nUse `update_state` for resumable state, not permanent history. Checkpoint when:\n\n- The user pauses, switches tasks, or will continue later.\n- A substantial milestone completes and meaningful work remains.\n- Work is blocked on a decision, permission, deployment, or dependency.\n- A long task is about to move to another conversation or agent.\n- The interaction ends with unfinished next steps.\n\nStore the current objective, verified status, completed work and evidence,\nsettled decisions, exact next action, relevant artifacts, and blocker if any.\nDo not checkpoint every turn or trivial progress.\n\nUse `record_event` for a completed milestone or historical occurrence. Use\n`todo` only for a concrete follow-up, not as a duplicate of every checkpoint.\n\n## 7. Resume work deterministically\n\nResume means reconstructing useful working context in the current conversation;\nit does not reopen or restore the original ChatGPT thread.\n\n1. Identify the project or authorized scope. Ask only if multiple targets would\n   materially change the result.\n2. Retrieve the latest relevant working state with `recall_context`.\n3. Add the project summary and relevant open TODOs, decisions, recent events,\n   or durable memories only as needed.\n4. Reconcile stale state against newer decisions or milestones.\n5. Present a compact Resume Brief: objective, verified status, completed work,\n   active decisions, blocker, and exact next action.\n6. Continue from that next action without repeating verified completed work.\n\nRead `references/workflows.md` for the Resume Brief contract.\n\n## 8. Create a complete handoff\n\nFor cross-conversation or cross-agent handoff:\n\n1. Write the latest resumable state with `update_state`.\n2. Record a concise `handoff` event with `record_event`.\n3. Create or update a `todo` only when a concrete follow-up needs ownership.\n4. Include the objective, verified status, decisions and rationale, completed\n   work, evidence and artifacts, constraints, work not to repeat, exact next\n   action, blocker or required input, and intended recipient when known.\n5. Return a human-readable handoff receipt; do not expose raw traces or secrets.\n\nRead `references/workflows.md` for the handoff packet.\n\n## 9. Handle projects and lifecycle safely\n\n- Existing `Projects / <name> / ...` paths attach to the existing formal\n  project.\n- Do not create a project merely because a memory path resembles a project.\n- When the user intends to create a new formal project, first call\n  `project(entity=\"project\", action=\"create\", name=\"<name>\")`, then save\n  project memories under its `Projects / <name> / ...` path.\n- Use `forget` without hard mode for recoverable soft deletion by default.\n- Use hard deletion only for an exact target after an explicit permanent-delete\n  request that acknowledges irreversibility.\n- Use `restore_memory` only for an eligible soft-deleted memory.\n- Retrieve or confirm the exact target before deletion. Tool availability or\n  annotations are never user confirmation.\n\n## 10. Return useful receipts\n\nAfter memory operations, report only what helps the user:\n\n- What was recalled, saved, updated, checkpointed, or skipped.\n- The category or project placement when useful.\n- Any ambiguity, blocker, or unavailable operation.\n- The next action when work remains.\n\nNever claim success without a successful tool result. Avoid raw internal IDs,\nscopes, traces, debug payloads, and unnecessary implementation details.\n"
}