Create a compact handoff for a new Hermes chat.

This is a mini-brief, not a full continuation brief and not a deep save. Optimize for speed and usefulness when switching models, starting a new Slack thread, or moving an active task to a fresh Hermes session.

## Instructions

1. Identify the current Hermes session and conversation origin from available session metadata.
2. Use the current conversation as the primary source.
3. Retrieve prior Hermes context only when needed to resolve a reference, decision, unfinished item, or current state. Do not reread or reproduce the entire conversation.
4. Check Open Brain, project files, and other available sources only when they materially improve accuracy.
5. Prefer current verified facts over memory, old summaries, or confident claims.
6. Do not execute the user's underlying task.
7. Do not execute `/deep-save`, create a full continuation brief, create a project artifact, or make broad memory writes unless explicitly requested.
8. Do not include secrets, passwords, API keys, bearer tokens, OAuth tokens, session cookies, tokenized URLs, or private credential-bearing command output.
9. If the current conversation contains no meaningful resumable work, say so and produce a minimal handoff rather than inventing context.

## Produce exactly two parts

### Part 1: Mini-brief

Use these headings and keep the whole brief concise:

- `Objective`
- `Verified state`
- `Decisions and constraints`
- `Completed work`
- `Open work`
- `Immediate next action`
- `Unverified or blocked`

Rules:

- State the desired outcome, not merely the last request.
- Include only facts needed to continue.
- Mark claims as `VERIFIED`, `INFERRED`, `UNVERIFIED`, `BLOCKED`, or `SUPERSEDED` when their status matters.
- Include exact paths, URLs, IDs, versions, branches, timestamps, or test results only when they affect the next action.
- For each open item, include its owner and acceptance test when known.
- Give exactly one immediate next action.
- If something is blocked, state the exact failed attempt, the real boundary if one has been established, the owner, and the unblock action.
- Name stale or superseded guidance when the next chat could otherwise follow it.

### Part 2: Fresh-chat prompt

Write a copy-ready prompt that I can paste into a new Hermes chat. It must:

- Tell the new chat to use the target model if one was specified in this conversation.
- Identify the prior Hermes session using the exact session ID from the current session metadata.
- Include the Slack channel, channel ID, thread ID, or other origin identifiers when available.
- Tell the new chat that the new thread does not automatically inherit the prior transcript.
- Tell it to retrieve the prior Hermes session and use the mini-brief as the fast starting point.
- Tell it to verify the current state before editing, sending, deploying, publishing, deleting, or making other consequential changes.
- State the one immediate next action.
- Preserve approval, credential, privacy, production, financial, destructive, and external-communication boundaries.
- Require execution when the next action is safe and in scope, rather than another planning summary.
- Require verification of the underlying outcome.
- Require a final report separating complete, unverified, and blocked work.

Do not ask the new chat to reread the entire prior conversation unless a specific fact cannot be recovered from the session, mini-brief, Open Brain, or current source-of-truth files.

## Output format

Return only:

```markdown
## Mini-brief

[concise handoff]

## Fresh-chat prompt

[copy-ready prompt]
```

Do not add a preamble, postscript, generic advice, or a vague promise. Do not claim that the new chat has inherited context. State that it must retrieve the identified prior session.

## Final quality check

Before responding, verify silently that:

- The session ID and origin identifiers came from current metadata, not guesswork.
- The objective and immediate next action are unambiguous.
- The mini-brief contains no unnecessary transcript replay.
- Every completion claim is supported by the conversation or a checked source.
- Stale and unverified claims are labeled.
- The fresh-chat prompt is executable by a new Hermes session.
- No secret or credential-bearing URL appears.
- `/deep-save` was not executed unless explicitly requested.