Update to Kumiho Memory

{
  "description": "Review Kumiho memories with the current app model, apply authorized corrections, and schedule bounded maintenance through the app's own scheduler when requested. Use for Dream State or memory-cleanup requests.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 456
    }
  ],
  "name": "dream-state",
  "skill_md_contents": "---\r\nname: dream-state\r\ndescription: Review Kumiho memories with the current app model, apply authorized corrections, and schedule bounded maintenance through the app's own scheduler when requested. Use for Dream State or memory-cleanup requests.\r\n---\r\n\r\n# Dream State — hosted memory review\r\n\r\nUse the current app model for assessment, MCP tools for data operations and the\r\napp's own scheduler for requested recurring runs. No external LLM API key, separate\r\nmodel service or operating-system cron is needed. Runs consume the host app's\r\nnormal usage. This workflow does not call the native server-side Dream State\r\nengine, retag identity documents or merge graph nodes. Report the operations\r\nactually performed instead of claiming the native engine ran.\r\n\r\n## Preview\r\n\r\n1. Establish the project/topic/space the user wants reviewed. An ambiguous\r\n   cleanup request means preview. Do not fan out across all projects.\r\n2. Use `kumiho_memory_retrieve` for at most five relevant items initially, with the\r\n   specified project and space filters. Inspect returned published revisions with\r\n   `kumiho_get_revision_by_tag` before judging duplicates or contradictions.\r\n3. Report candidate pairs, their krefs, evidence and uncertainty. Age alone is not\r\n   a reason to retire a memory; distinguish proposals, accepted decisions and\r\n   later corrections. Keep useful differences and provenance visible.\r\n4. Preview performs reads only. Do not call space profiling merely to preview: that\r\n   tool can persist results. No consolidation, decomposition, tagging or retirement\r\n   is authorized by a preview request.\r\n\r\n## Apply approved changes\r\n\r\nShow the exact changes and obtain authorization for each affected item unless the\r\nuser already supplied those exact changes. Native published-item safeguards are\r\nnot automatically supplied by this manual path, so do not use a broad \"clean up\"\r\nrequest to authorize bulk retirement. Never retire all copies of information.\r\n\r\n- Use `kumiho_memory_reflect` for a user-approved correction with relevant\r\n  `source_krefs`, `revises` set to the corrected memory's `kref`, its memory type\r\n  and language, the subject restated, and `discover_edges: false`, so it becomes a\r\n  new revision of that memory. If you cannot tell which memory to revise, save it\r\n  normally; a result that is a separate item leaves the old memory recalled until\r\n  its approved retirement. Omit session ID initially and follow a returned\r\n  `session_required` ID for this conversation only.\r\n- Use `kumiho_deprecate_item` only for the specifically approved item kref and\r\n  report the result. Retirement excludes it from normal recall; it is not permanent\r\n  erasure. Do not clear chat buffers or run unexposed tools as a substitute.\r\n\r\nStop after the bounded reviewed set. Report reviewed/proposed/applied/failed counts\r\nseparately and only from observed results. User instructions take precedence;\r\ninstallation or testing does not authorize maintenance of real memories.\r\n\r\n## Recurring maintenance in the app\r\n\r\nWhen the user asks for recurring maintenance, use the host's supported task or\r\nautomation tool after checking its current schema. Reuse an existing matching task\r\nwhen possible. Do not create an OS scheduled task, server cron, subprocess daemon\r\nor external model subscription. A request to add this skill is not a request to\r\ncreate an actual schedule.\r\n\r\nPut a self-contained instruction in the scheduled task: the connected Kumiho\r\nplugin, exact project/space/topic, requested timezone/cadence, at most five items\r\nper run unless otherwise authorized, review steps above, and allowed mutations.\r\nDefault to read-only review and proposed changes. If the user authorizes automatic\r\nchanges, encode the precise policy and bounds; do not re-ask on each run for the\r\nsame authorized operations, but stop for ambiguous targets or operations outside\r\nthat policy. An automatic retirement policy must identify the evidence required\r\nand which retained item preserves the information. \"Clean up everything\" alone\r\ndoes not establish that policy.\r\n\r\nKeep credentials and transcript content out of the schedule prompt. Connect through\r\nthe app's existing OAuth integration. Suppress routine notifications when nothing\r\nactionable changed; report actual changes, failures or required user decisions,\r\nunless the user requests regular reports. Verify the scheduler returned an active\r\ntask before saying it is scheduled. If this host or plan lacks scheduling or cannot\r\ncall the plugin from scheduled runs, say so; do not claim a task will run elsewhere.\r\n"
}