Bring your AI history into ORCA

You are helping me migrate my useful AI history into a knowledge platform called ORCA.

ORCA has these building blocks:
- PEARL: a reusable skill/prompt for a repeatable task
- KNOWLEDGE PACK: durable facts, reference material and decisions used for retrieval
- CREWMATE: an AI agent with a persona, instructions and skills
- WORKSPACE: a container for a whole area of work (files, memory, agents, recurring tasks)
- DOCUMENT/SOP: a finished artefact or a step-by-step procedure

Read everything available to you in THIS conversation/project (all messages, any project instructions, and any attached files). Do NOT summarise it yet. Instead, produce an inventory and triage:

1. ONE-PARAGRAPH OVERVIEW: what is this history actually about, and what was I trying to achieve across it?

2. EXTRACTABLE ASSETS: a table with columns:
   | # | Asset (short name) | What it is | Best ORCA target (Pearl / Knowledge Pack / Crewmate / Workspace / Document) | Why | Reusability (High/Med/Low) |
   List every distinct reusable thing you can find: repeated tasks, hard-won facts, personas I leaned on, procedures, templates, decisions with rationale, glossaries.

3. THE INSIGHTS I WOULD LOSE: bullet the non-obvious lessons, gotchas, decisions and reasoning embedded in this history that a plain summary would throw away. Be specific: quote or paraphrase the actual insight, not the topic.

4. RECOMMENDED NEXT EXTRACTIONS: for the top 3 to 5 assets, tell me which follow-up extraction to run and in what order.

Be honest about low-value or duplicate material: flag what is NOT worth migrating.

Turn this conversation into a reusable "Pearl": a self-contained skill I can run again on new inputs inside ORCA. Study how the task was actually done here, including the parts that went wrong before they went right, and capture the working method, not just the final answer.

Return this exact structure in Markdown:

# Pearl: <clear, action-oriented name>

**Purpose (one sentence):** what this Pearl does.

**When to use it / trigger:** the situations where I would reach for this. Include the phrases a user might say that should activate it.

**Inputs it needs:** what the user or ORCA must provide each time (documents, values, context).

**Instructions (the reusable prompt body):**
Write this as clean, generalised instructions the AI should follow every time, NOT tied to this one example. Convert the specific case into variables/placeholders like {topic}, {document}, {audience}. Bake in the good judgement that emerged during the conversation: the checks, the order of steps, the format that worked, the mistakes to avoid.

**Method / steps:** the ordered steps this skill follows.

**Guardrails & gotchas:** what tripped us up in the original conversation and how to avoid it. Include the "we tried X, it failed, so do Y instead" lessons explicitly.

**Tools / data it may need:** any external systems, files, or lookups involved (so ORCA can attach the right tools).

**Worked example:** one condensed input-to-output example drawn from this conversation, clearly labelled as an example so it is not mistaken for the instructions.

Keep the instructions model-agnostic and paste-ready.

Extract the durable, reusable KNOWLEDGE from this conversation into a "Knowledge Pack" that ORCA can store and retrieve later. I care about the facts, the reasoning and the decisions, not the small talk or the dead ends. Preserve nuance; do not oversimplify.

Return this structure in Markdown:

# Knowledge Pack: <topic title>

**Summary (3 to 5 sentences):** what this body of knowledge covers and why it matters.

**Source & date context:** where this knowledge came from and roughly when it was true (flag anything time-sensitive that may go stale).

**Key facts & findings:** a well-organised bulleted list of the concrete, reusable facts. Group them under sub-headings. Keep each fact standalone and retrievable (a reader should be able to understand it without the surrounding chat).

**Decisions made & the reasoning behind them:** list each decision, the options considered, and WHY we chose what we chose. This rationale is the most valuable part: capture it fully.

**Insights & non-obvious lessons:** the moments of realisation, the trade-offs, and the things that surprised us.

**Glossary / key entities:** important terms, names, systems, people or acronyms and what they mean in this context.

**Open questions & unknowns:** what was left unresolved.

**Do-not-lose quotes:** any 1 to 3 verbatim snippets worth keeping exactly as written.

Deduplicate. If something was corrected later in the conversation, keep only the corrected version and note that it superseded an earlier take.

I want to recreate the assistant persona I have been using in this conversation as a reusable "Crewmate" (an AI agent) in ORCA. Reverse-engineer the persona from how it actually behaved here: its voice, its expertise, its standards. Then write a durable configuration for it.

Return this structure in Markdown:

# Crewmate: <name/role>

**Role & purpose:** who this agent is and what it is for.

**Persona & voice:** tone, style, and any consistent quirks that made it useful (for example: blunt, Socratic, uses Australian English, always asks for the goal first).

**Expertise & scope:** what it knows and what it should stay away from.

**System instructions (paste-ready):** write the full standing instructions for this agent in the second person ("You are..."). Encode the behaviours I clearly valued in this thread: how it structures answers, what it always checks, what it refuses, its defaults.

**Skills it needs (candidate Pearls):** list the repeatable tasks this agent performs. Each of these could become a Pearl attached to it.

**Knowledge it relies on:** the reference material or context it assumes (candidate Knowledge Packs).

**Example interaction:** one short representative exchange that shows the persona in action.

Base everything on evidence from the conversation, not a generic template.

This Project represents a whole area of my work. I am migrating it into an ORCA "Workspace": a container that holds the mission, the standing context/memory, the files, the agents and skills, and the recurring work for this area. Read the project instructions, the attached files, and the conversation history, then reconstruct it as a Workspace blueprint.

Return this structure in Markdown:

# Workspace: <project name>

**Mission / purpose:** what this whole area of work is for, in 2 to 3 sentences.

**Business domain:** the domain this belongs to (for example: Finance, Compliance, Marketing, Ops) so ORCA can tag and route work correctly.

**Standing context & memory:** the durable facts, preferences, constraints and background that should ALWAYS be in context for anyone working here. This is the project's long-term memory: pull it out of the instructions and the recurring themes in the threads. Include key people, systems, conventions, do's and don'ts, and any "we always / we never" rules.

**Custom / standing instructions:** the operating instructions that governed this project, cleaned up and generalised so they read as durable policy, not one-off requests.

**Files & artefacts:** list every file or artefact referenced or produced, with a one-line note on what each is and whether it is reference knowledge, a template, or a finished output.

**Recurring tasks & workflows:** the repeatable jobs done in this project. For each, note how often, the trigger, and the steps. These become scheduled tasks or Pearls.

**Agents/skills needed:** the personas and repeatable skills this workspace should contain (candidate Crewmates and Pearls), each with a one-line purpose.

**Key decisions & their rationale:** the important calls made in this project and WHY, so the reasoning travels with the work.

**Open threads:** what is in flight or unresolved.

Prioritise capturing insight and context over completeness of trivia. If the project is large, tell me which parts you compressed.

Take the artefact/output produced in this conversation and turn it into a clean, standalone document I can store in ORCA. Separate the finished artefact from the chat around it, and add just enough front-matter that it stands on its own.

Return this in Markdown:

# Document: <title>

**Type:** (SOP / Policy / Template / Checklist / Spec / Reference: pick the best fit)

**Purpose & audience:** what it is for and who uses it.

**Status & version note:** how finished it is, and anything that would need updating later.

**The document itself:**
Reproduce the finished artefact in clean, well-structured form. If it is a procedure, format it as numbered, actionable steps with clear owners/conditions. Remove chat cruft, keep the substance.

**Assumptions & context baked in:** the context this document assumes (so a future reader is not missing the "why").

**Change/decision notes:** anything we changed our minds about while producing it, and why.

If this is a step-by-step procedure, format it so it could drive an automated SOP: each step as "When <condition>, do <action>, producing <result>."