agentleFS
stevemarksavastevemarksava/superteam/skills/superteam-demo/SKILL.md

superteam-demo

Skill1 starsChanged 2 months agoMITView on GitHub

Hackathon demo runner — pre-flight check, three live demo scenarios, and demo-specific Coach game tape

The file

Show its rules and text

We split this skill into the rules it gives an agent

23 rules. About 0 are common, 16 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈16

    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:

    Do not ad-lib the demo structure — the scenarios are sequenced to build on each other.

    line 12
  2. Distinctive:

    Run before going on stage.

    line 24
  3. Distinctive:

    Do not skip to the demos with a failing pre-flight.

    line 25
  4. Distinctive:

    Verify the required sentinels exist:

    line 29
  5. Very specific:

    If still not running: check mcp-server/dist/index.js exists.

    line 45
  6. Distinctive:

    If not found: the autostart was not registered.

    line 56
  7. Very specific:

    If the model is not loaded, open the Foundry Local app and load phi-4-mini before continuing.

    line 71
  8. Distinctive:

    Run this test for each surface in sequence.

    line 75
  9. Very specific:

    If any surface fails, consult the troubleshooting matrix in docs/using-superteam.md before going on stage.

    line 103
  10. Very specific:

    If not present, record it now using the shot list in docs/hackathon-redmond-stage-script.md.

    line 116
  11. Distinctive:

    Do not go on stage without it.

    line 117
  12. Distinctive:

    When all five checks pass, write the pre-flight sentinel:

    line 121
  13. Distinctive:

    Wait for response. Verify it includes Bob's blueprint and Paulien's task breakdown.

    line 145
  14. Distinctive:

    Wait for response. Verify:

    line 156
  15. Distinctive:

    No agent steps out of lane

    line 159
  16. Distinctive:

    Do not retry more than once on stage.

    line 167
  17. Very specific:

    Wait. Verify the response includes a delegation note (Superman routed heavy reasoning to Claude Code CLI via MCP).

    line 186
  18. Very specific:

    Wait. Verify the response is attributed to [Foundry Local].

    line 199
  19. Distinctive:

    If delegation routing does not appear automatically, run this directly in the CLI:

    line 209
  20. Very specific:

    Wait. Verify the response is attributed to [Foundry Local] — not to Claude cloud.

    line 237
  21. Distinctive:

    If Foundry Local times out:

    line 251
  22. Distinctive:

    Check Window 3 — if the model unloaded, say "cold start" and wait 30 seconds

    line 252
  23. Distinctive:

    If still failing, disable airplane mode, run the same prompt normally, explain that the cloud path gives the same answer — the airplane …

    line 254

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

/superteam.demo — Hackathon Demo Skill

You are running the Superteam hackathon demo sequence. Follow every step in order. Do not ad-lib the demo structure — the scenarios are sequenced to build on each other. Each step has a pass/fail check before proceeding. If a check fails, use the stated recovery action before continuing.

Reference the full stage script at docs/hackathon-redmond-stage-script.md for narration text and fallback procedures. This skill handles the mechanical steps — the stage script handles the words.


Step 1 — Pre-Flight Check

Run before going on stage. If any check fails, fix it before proceeding. Do not skip to the demos with a failing pre-flight.

1a — Sentinel check

Verify the required sentinels exist:

# MCP server running
Import-Module SuperteamControl
$status = Get-SuperteamMcpStatus
if ($status.State -ne "Running") {
    Start-SuperteamMcp
    Start-Sleep -Seconds 5
    $status = Get-SuperteamMcpStatus
}
Write-Host "MCP: $($status.State)"

Expected: MCP: Running

If still not running: check mcp-server/dist/index.js exists. If not: cd mcp-server && npm run build.

1b — Autostart task registered

Get-ScheduledTask -TaskPath "\Superteam\" -TaskName "SuperteamMcp" -ErrorAction SilentlyContinue |
    Select-Object TaskName, State

Expected: TaskName: SuperteamMcp State: Ready or State: Running

If not found: the autostart was not registered. Run:

./scripts/install-superteam.ps1 -Auto claude-code-cli -Yes

1c — Foundry Local loaded

# Verify Foundry Local service is running and model is warm
Get-Service -Name FoundryLocal | Select-Object -ExpandProperty Status

Expected: Running

If stopped: Get-Service -Name FoundryLocal | Start-Service If the model is not loaded, open the Foundry Local app and load phi-4-mini before continuing.

1d — Squad responds on all three demo surfaces

Run this test for each surface in sequence. Do not proceed to demos until all three pass.

Surface 1 — Claude Code CLI:

Superman, status

Expected: squad status response, MCP: connected (host: claude-code-cli)

Surface 2 — VS Code Copilot:

Open VS Code, switch Copilot Chat to Agent mode, type:

@superteam-superman Superman, status

Expected: same squad status response

Surface 3 — Foundry Local (via CLI delegation):

Superman, call superteam_run_local with task "hello"

Expected: response attributed to [Foundry Local] runtime

If any surface fails, consult the troubleshooting matrix in docs/using-superteam.md before going on stage.

1e — Fallback video ready

Confirm the Kryptonite C5 fallback recording is on the local filesystem and playable:

Test-Path "$env:USERPROFILE\Videos\superteam-c5-fallback.mp4"

Expected: True

If not present, record it now using the shot list in docs/hackathon-redmond-stage-script.md. Do not go on stage without it.

Pre-flight complete

When all five checks pass, write the pre-flight sentinel:

"DEMO PRE-FLIGHT PASSED — $(Get-Date -Format 'yyyy-MM-dd HH:mm')" |
    Set-Content ".squad/demo-preflight.md"

Step 2 — Demo 1: Cross-Surface Consistency

Target time: 90 seconds
Message: same squad, every tool

Run

Window 1 (Claude Code CLI):

Type at the claude prompt:

Superman, plan a small feature: add a --dry-run flag to the install script

Wait for response. Verify it includes Bob's blueprint and Paulien's task breakdown. Note the total response time — it should be under 10 seconds.

Switch to Window 2 (VS Code Copilot Chat, Agent mode):

Type:

@superteam-superman Superman, continue the --dry-run plan. Eric, implement task 1.

Wait for response. Verify:

  • Superman references the same task structure (context continuity)
  • Eric acknowledges the brief and begins implementation
  • No agent steps out of lane

Pass criteria

  • CLI response includes "Bob:" and "Paulien:" attributions or equivalent routing note
  • VS Code response references the same feature brief without being re-explained
  • Response times: CLI under 10 seconds, VS Code under 15 seconds

On failure

Do not retry more than once on stage. Switch to the prepared terminal with the pre-run output visible and narrate: "Here is what the response looks like — the demo surface is unresponsive but the squad itself is live."


Step 3 — Demo 2: Cross-Tool Delegation

Target time: 120 seconds
Message: one tool triggers another automatically

Run

Window 2 (VS Code Copilot):

@superteam-superman Do a full architectural review of the install script and flag any risks

Wait. Verify the response includes a delegation note (Superman routed heavy reasoning to Claude Code CLI via MCP). The response should reference superteam_delegate or show routing attribution.

Show Window 1 briefly — scroll up to show any MCP tool call output visible in the CLI panel. Narrate the handoff.

Window 2 again:

@superteam-superman Route the risk summary to Foundry Local and ask it to produce a one-paragraph plain-English version

Wait. Verify the response is attributed to [Foundry Local].

Switch to Window 3 (Foundry Local app) — show inference activity on screen.

Pass criteria

  • Delegation note appears in VS Code response (or MCP call visible in CLI)
  • Foundry Local response includes [Foundry Local] or [foundry] attribution
  • Foundry Local app shows inference activity during the third prompt

On failure

If delegation routing does not appear automatically, run this directly in the CLI:

Superman, call superteam_delegate with runtime foundry and task "one paragraph risk summary of install script"

Show the Foundry output. Narrate that in production, Superman decides routing automatically based on task type and credit headroom.


Step 4 — Demo 3: Airplane Mode

Target time: 90 seconds
Message: works everywhere, including nowhere
Action required: enable airplane mode before this step

Run

Enable airplane mode on the demo laptop (taskbar → network icon → Airplane mode). Show the disconnected status to the audience.

Window 1 (Claude Code CLI):

Superman, summarise the last three game tapes

Wait. Verify the response is attributed to [Foundry Local] — not to Claude cloud. The content should be a real summary (Coach game tapes must exist — check that .squad/coach/game-tape/ has at least three entries).

Switch to Window 3 (Foundry Local app) — show inference activity.

Disable airplane mode after the response arrives.

Pass criteria

  • Response arrives within 60 seconds of the prompt (Foundry cold-start budget)
  • Response is attributed to Foundry Local runtime
  • Content is a coherent summary of the actual game tape files (not a hallucination)

On failure

If Foundry Local times out:

  1. Check Window 3 — if the model unloaded, say "cold start" and wait 30 seconds
  2. Re-run the prompt once
  3. If still failing, disable airplane mode, run the same prompt normally, explain that the cloud path gives the same answer — the airplane mode path requires Foundry Local to have been pre-warmed (which it was — the demo surface had a problem, not the architecture)

After Demo 3

Mark the demo sequence complete:

"DEMO COMPLETE — $(Get-Date -Format 'yyyy-MM-dd HH:mm')" |
    Set-Content ".squad/demo-complete.md"

Step 5 — Coach Writes Demo Game Tape

Immediately after coming off stage, trigger Coach to write a demo-specific game tape. This is separate from the normal post-push tape — it captures what worked, what did not, and what to drill before the next demo event.

Run from CLI

Coach, write a demo game tape. Cover: what held during the demo, what failed or
needed recovery, any audience questions that revealed gaps, and what to drill
before the next demo.

Coach writes to:

.squad/coach/game-tape/YYYY-MM-DD-demo.md

(Demo tape uses the -demo suffix to distinguish it from the regular delivery tape.)

What Coach should cover in the demo tape

  • Demo 1 result: did cross-surface consistency land? What was the audience reaction?
  • Demo 2 result: did the delegation routing appear automatically, or did we fall back to manual? If fallback, what caused it?
  • Demo 3 result: did Foundry Local respond within the window? Cold start issue?
  • Fallback used: yes or no; which scenario triggered it; what the audience saw
  • Judge questions: any question that revealed a gap in the demo or the product
  • Top drill: the one thing to practice before the next demo event

Pass criteria for this step

File .squad/coach/game-tape/YYYY-MM-DD-demo.md exists and contains at minimum:

  • A line covering each of the three demos
  • A "Top drill" recommendation
  • The date and event name in the frontmatter or opening line

Skill Complete

The demo is done and the game tape is written.

Report to Steve:

DEMO DELIVERED
Pre-flight: all 5 checks passed
Demo 1: [pass / recovery used]
Demo 2: [pass / recovery used]
Demo 3: [pass / recovery used]
Fallback: [not needed / triggered at Demo N]
Coach tape: .squad/coach/game-tape/YYYY-MM-DD-demo.md

Clean. The team is on record.

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.