agentleFS
sindre-aisindre-ai/maskin/.agents/skills/workspace-observer-onboarding/SKILL.md

workspace-observer-onboarding

Skill8 starsChanged 3 months agoApache-2.0View on GitHub

Guides the Workspace Coach to run onboarding for a new workspace — detecting an empty workspace, creating an onboarding session, identifying the owner, and posting context prompts in sequence to get the workspace to its first bet.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 7 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run this skill when your observation detects a workspace that:

    line 10
  2. Very specific:

    If an onboarding_session object already exists for this workspace, exit silently — onboarding is already underway.

    line 15
  3. Very specific:

    Call create_objects to create a single object:

    line 21
  4. Very specific:

    After each reply, call create_objects once with both the knowledge node and the about edge in the same batch.

    line 56
  5. Very specific:

    Keep the existing relates_to → onboarding_session edge in the same batch so the session view still lists its captures.

    line 101
  6. Very specific:

    Do NOT write anything to the actor's memory field.

    line 103
  7. Distinctive:

    After all five prompts are answered (or if the owner stops responding after 24h), update the onboarding session:

    line 107
  8. Distinctive:

    Add a closing comment: "Done — agents now have the context they need.

    line 109
  9. Distinctive:

    Do not run this for workspaces older than 24h, even if they have zero bets

    line 113
  10. Distinctive:

    Do not post all five prompts at once — sequence matters.

    line 114
  11. Distinctive:

    Do not create the onboarding session more than once per workspace

    line 115
  12. Distinctive:

    Do not capture knowledge objects if the owner did not reply — only record actual answers

    line 116
  13. Very specific:

    Do not create the knowledge row and the about edge in separate calls — they must commit in the same create_objects transaction

    line 117
  14. Very specific:

    Do not write to actors.memory — that field is Config, not Facts

    line 118

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Workspace Coach Onboarding

When to run

Run this skill when your observation detects a workspace that:

  • Has onboarding_enabled = true (read from list_workspaces — exit silently if false)
  • Was created within the last 24 hours (check createdAt on the workspace)
  • Has zero bets (no objects of type bet exist in the workspace)

If an onboarding_session object already exists for this workspace, exit silently — onboarding is already underway.

What to do

1. Create the onboarding session

Call create_objects to create a single object:

  • type: onboarding_session
  • title: "Getting your workspace ready"
  • status: active
  • content: brief description of what this session is — "A guided conversation to capture the context agents need to run quality bets. Takes 5–10 minutes."

Save the returned object ID — all prompts in the next step are comments posted on this object.

2. Identify the workspace owner

List workspace members and identify the human actor (type != "agent") who created the workspace or is listed as owner. Keep the owner's actor ID and the workspace ID handy — every knowledge write in step 4 needs them.

3. Post prompts in sequence

Post the five prompts below as comments on the onboarding session object, in order. Wait for a reply to each prompt before posting the next one. Capture each reply as a knowledge object (see step 4).

The tone is conversational — you are an assistant asking questions, not a form. Write each prompt as a short message.

Prompt 1 — Product vision (prompt_key: product_vision)

What does your product do and who is it for? A sentence or two is enough — just enough for agents to understand what you're building and what outcome you're going for.

Prompt 2 — ICP (prompt_key: icp)

Who is your ideal customer? The sharper the better — role, company type, the specific pain they have. If you have real customers already, describe one of them.

Prompt 3 — First-bet hypothesis (prompt_key: first_bet_hypothesis)

What's the single most important thing to figure out or build right now? This becomes your first bet — what would move the needle most if it worked?

Prompt 4 — North Star metric (prompt_key: north_star_metric)

How will you know the product is working? Name one number — the metric that, if it goes up, you're succeeding.

Prompt 5 — Customer evidence (prompt_key: customer_evidence)

What have you already heard from customers or potential customers? Even a single quote or observation is useful — agents use this to calibrate bet quality and avoid building the wrong thing.

4. Capture each reply as a knowledge object AND an about edge — one atomic call

After each reply, call create_objects once with both the knowledge node and the about edge in the same batch. The knowledge row and its edge must commit together — no bare-knowledge row about the workspace owner may survive.

The edge target depends on the prompt:

  • product_vision, icp, first_bet_hypothesis, customer_evidence → owner-targeted: edge target = the workspace owner's actor id, metadata.subject_kind = "workspace_owner".
  • north_star_metric → workspace-targeted: edge target = the workspace id, metadata.subject_kind = "workspace".

Payload shape (owner-targeted example):

{
  "nodes": [
    {
      "$id": "k1",
      "type": "knowledge",
      "title": "Product vision",
      "status": "validated",
      "content": "<owner's reply verbatim, or a clean restatement if conversational>",
      "metadata": {
        "source": "workspace_onboarding",
        "prompt_key": "product_vision",
        "subject_kind": "workspace_owner",
        "subject_id": "<owner-actor-uuid>",
        "claim": "<one-sentence restatement of what the reply asserts>",
        "confidence": "medium",
        "valid_from": "<ISO timestamp of the reply>",
        "valid_to": null,
        "tags": ["onboarding", "provenance:workspace_onboarding"]
      }
    }
  ],
  "edges": [
    { "source": "k1", "target": "<owner-actor-uuid>", "type": "about" },
    { "source": "k1", "target": "<onboarding-session-object-id>", "type": "relates_to" }
  ]
}

For north_star_metric, swap the about target to the workspace id and set subject_kind / subject_id to "workspace" / <workspace-uuid>.

Field notes:

  • claim is a one-sentence normalized restatement so downstream agents can reason without re-reading the raw reply.
  • source is the fixed string "workspace_onboarding" — this identifies the row as coming from this skill.
  • confidence defaults to "medium" for owner self-report; bump to "high" if the reply cites a concrete customer or metric.
  • valid_from is the reply's timestamp; valid_to stays null until superseded.
  • Keep the existing relates_to → onboarding_session edge in the same batch so the session view still lists its captures.

Do NOT write anything to the actor's memory field. That field is reserved for operating Config (approval gates, Slack id, escalation rules) per the ratified user-info model — Facts live only as knowledge objects with an about edge.

5. Close the session

After all five prompts are answered (or if the owner stops responding after 24h), update the onboarding session:

  • status: done
  • Add a closing comment: "Done — agents now have the context they need. Your first bet can start anytime."

What NOT to do

  • Do not run this for workspaces older than 24h, even if they have zero bets
  • Do not post all five prompts at once — sequence matters; each answer informs the next question
  • Do not create the onboarding session more than once per workspace
  • Do not capture knowledge objects if the owner did not reply — only record actual answers
  • Do not create the knowledge row and the about edge in separate calls — they must commit in the same create_objects transaction
  • Do not write to actors.memory — that field is Config, not Facts

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.