# Live demo runbook

## Before the talk

Use the current Skills Workshop checkout containing `recall`, `feedback`, and `insights`.
The presentation snapshot is revision `dc5adeb3a642e5bf1fa03c52c600b1391ac56148`, dated 18 September 2026.
Later memory will naturally have different counts.

Open a terminal in that checkout and keep slide 4 visible on the presentation screen.
Use a readable terminal font.
Check network access in advance.

The commands below inspect existing state.
They do not install skills or alter downstream projects.
Public discovery sends the query to external providers.

## 1. Recall an earlier task

```console
pixi run workshop recall "license compliance"
```

Look for `introduce-reuse-compliance` from `con/skills`.
Point out the matched terms and the prior-use caveats, including the distinction between an agent assertion and a human review.
The useful result is finding a relevant skill through the task rather than remembering its exact name.

Optional second example, only if time allows:

```console
pixi run workshop recall "something I used to verify commit trailers"
```

This query now retrieves `commit-provenance`; the earlier exact-term search did not handle this natural-language wording.

## 2. Compare candidates

```console
pixi run workshop find "license compliance"
```

Show the local memory and registered sources before public results.
Explain that Workshop coordinates existing discovery providers.
Their results can be noisy, unavailable, or incomplete.
A search result is not an endorsement.

For a slow or unavailable provider, switch to the offline search:

```console
pixi run workshop find "license compliance" --offline
```

The registered source pool includes K-Dense, NiPreps, `con/skills`, Anthropic, and OpenAI.
Configured installed skill roots participate in local discovery.
Registered sources are searched at their checked-out revisions, not necessarily their latest upstream versions.

## 3. Inspect actual experience

```console
pixi run workshop show introduce-reuse-compliance
pixi run workshop history introduce-reuse-compliance
```

Open [the skill source](https://github.com/con/skills/tree/master/introduce-reuse-compliance).
Briefly show what it tells the agent to do, then return to the recorded outcome.

The [9 September observation](https://github.com/leej3/skills-workshop/blob/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/memory/events/2026/09/20260909T140926Z-b4c836e1-776d-4b61-b3b9-fdb44d09e68a.json) reports useful licensing, validation, and pre-commit guidance.
It also records a Python compatibility problem that required a project-local `uvx` wrapper.
This is useful preparation for another task, even though the report is unreviewed and does not establish causality.

## 4. Show the feedback loop

```console
pixi run workshop insights
pixi run workshop insights --since 2026-09-18
```

Show the [feedback control skill](https://github.com/leej3/skills-workshop/tree/dc5adeb3a642e5bf1fa03c52c600b1391ac56148/.agents/skills/workshop-feedback).
It asks the agent for one short observation after meaningful participation.
Failures and uncertain benefit belong in memory; a rating is optional.

Inspect a real observation rather than recording a successful use merely for the demonstration.
Global instructions currently trigger the request; there is no shipped lifecycle hook guaranteeing capture.

## Fallback if the terminal or network is unavailable

Use slide 5 and its source link.
The condensed real observation is:

- Task: add REUSE licensing checks to a project.
- Helped: licensing structure, validation, and pre-commit integration.
- Correction: a project-local wrapper was needed for Python compatibility.
- Status: agent-reported and unreviewed.

Explain how another member could use that caveat before applying the skill.
This demonstrates the purpose of memory without pretending a live command ran.

## Discussion after the demo

Compare the group pilot with usage reports in GitHub issues and PRs, and with a shared registry such as SkillNote.
Keep skill development in `con/skills`.
Ask whether a few members would try two recurring tasks over four weeks.

Useful outcomes include an observation influencing another member's choice, a repeated failure being avoided, or a concrete improvement reaching a skill.
Also note recording effort and missed feedback.
More events alone are not the success criterion.