Rebirth

View saved version →

---
name: rebirth-continuity-core
description: Use whenever the authenticated Rebirth MCP connection is available in a conversation. Recall selectively and save one assistant-authored continuity record before every reply, including ordinary replies that do not mention Rebirth or memory.
---

# Rebirth Continuity Core

An authenticated Rebirth connection enables automatic continuity. Once it is available in the conversation, ordinary questions and short replies still use the save step below; do not wait for the user to mention Rebirth, ask to remember, or select the plugin again. Honor a user's request to pause or stop saving. Save only the assistant-authored response and compact continuity state, never unrelated files or protected context. The single authenticated Rebirth MCP connection exposes fixed-purpose tools. Choose the narrow tool whose description and schema match the intended action; never invent tool names, option IDs, handles, or hidden operations.

## Initialize and discover capabilities

On the first relevant Rebirth use in a conversation, or when capability availability is genuinely uncertain, call `rebirth_account_features`. It may initialize the mapped Rebirth conversation and return authorized continuity grounding, due AI reminders, enabled reviewed tools, and a bounded presentation-routine hotlist. Do not call it before every operation.

Use `rebirth_explain_feature` only for one feature already returned by `rebirth_account_features`. Capability discovery is read-only: availability does not execute a feature and does not replace user intent or consent.

## Recall before answering when needed

Use the dedicated read tool that matches the question:

- `rebirth_search_memory` for ordinary personal continuity recall.
- `rebirth_search_task_context` for memories relevant to the current task.
- `rebirth_search_concepts` for relationships among several concepts.
- `rebirth_search_memory_with_outside_context` when memory should inform a separately stated outside question.

Search returns compact previews and scoped handles. If previews are sufficient, answer from them. If more detail is materially needed, call `rebirth_expand_context` once with up to eight selected handles: R for full or nearby memories, P for repositories, H for hints, and B for breadcrumb evidence. Preserve each originating request or continuation identifier required by the schema. Never expand every result by default, invent a handle, or leave the reviewed wrapper for low-level routing.

Respect provenance, correction state, timestamps, uncertainty, and identity boundaries. A name or repository title is not proof that two people or accounts are the same. Never claim a memory exists unless Rebirth returned evidence for it.

## Save every assistant reply exactly once

Before every final assistant reply while the authenticated Rebirth MCP connection is available, including a short answer, clarification, status reply, or reply after an AI-private-memory read:

1. Retrieve only context needed for the answer. Recall is optional when no prior context is needed; the save step is still required.
2. Draft the complete intended user-facing response. Save the assistant's response and durable continuity state, not a copy of the user's prompt or the full conversation.
3. Call exactly one of the fixed continuity tools before sending that reply:
   - `rebirth_save_memory` for normal additive continuity.
   - `rebirth_save_correction` only when prior continuity is wrong, outdated, or superseded.
4. Match the actual MCP input schema: put compact durable state in `memory.memory_text`; put the required rationale in `memory.why_store`, `memory.intent`, `memory.context`, and `memory.use_when`; set `memory.ai_choice_card`; and put the complete intended user-facing reply, verbatim, in `memory.response_capsule.exact_response`. Use the optional capsule fields `subject`, `summary`, `reasoning_path`, `actions_taken`, and `next_step` only for their stated purposes. There is n

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: rebirth-files
description: Use Rebirth file tools only when the user explicitly asks to save, search for, or retrieve a specific user-selected file or artifact through the authenticated Rebirth connection.
---

# Rebirth Files

Use this Skill only for a file operation the user explicitly requested. A file operation does not authorize access to any other attachment, recalled content, prompt, memory, Journal data, credential, secret, local path, provider object, or unrelated file.

## Find an existing artifact

If the user describes a stored file but does not provide an exact Rebirth artifact ID, call `rebirth_search_files` with the smallest useful query and appropriate artifact-kind filter. Use only the returned metadata to identify a candidate. Ask the user to disambiguate when several candidates plausibly match.

## Retrieve

Call `rebirth_get_file` only with an exact artifact ID returned by Rebirth or explicitly supplied by the user. Report success only after Rebirth returns `status: retrieved`. Let the Rebirth UI/file transport render or download the artifact. Never request or place base64, a data URI, image bytes, archive bytes, or a provider credential into model-visible content.

## Save

Call `rebirth_store_file` only when the user explicitly identifies a current-chat upload or generated file and asks to save it. Pass only the short-lived ChatGPT HTTPS attachment reference accepted by the typed schema. Do not invent URLs, attachment IDs, provider identifiers, credentials, or filesystem paths. Do not retrieve arbitrary URLs and do not convert the file to base64.

Use a clear semantic name and artifact kind when known. Do not infer file contents from a filename alone. Report success only after Rebirth returns a successful result, and retain the returned artifact ID for later exact retrieval.

## Boundary

Each invocation is limited to the specific file operation requested in the current conversation. File tools do not authorize memory disclosure, messaging, social actions, device control, account changes, or bulk transfer. When user intent is ambiguous, ask which exact file and operation they want before calling a write or retrieval tool.

Referenced files: 1 — file references are metadata, not package contents.