Session distiller

Session distiller captures prior art from an external Claude session — a widget, a multi-step workflow, ad-hoc tool use — and turns it into a draft skill in your organization's review queue. No Harriet Desktop, no device enrollment, no connector deploy required.

When to use it

Use Session distiller when someone has already built something useful in Claude and you want it reviewed and stored in Harriet before rolling it out to the team. It covers exactly the gap Skilify does not: Skilify runs on a provisioned desktop, while Session distiller works inside any Claude session, before the author has installed anything.

The order of operations is capture, then review, then rollout. The distiller only helps with capture; approval and provisioning happen in Harriet afterwards.

What it captures

The output is an Anthropic-compatible skill package: a SKILL.md at the root (YAML front matter plus Markdown instructions), optional references/*.md files for checklists and examples, and small UTF-8 scripts only when the workflow truly needs them.

The distiller walks the session through a deliberate workflow: confirm the user wants a repeatable team skill rather than a frozen copy of the chat, inventory what made the experiment work, separate the repeatable procedure from disposable UI, draft SKILL.md, strip PII and credentials, and mentally test whether a fresh session could reproduce the outcome from the package alone.

Some things are deliberately excluded from the package: HTML pages, dashboards, or widget UIs that Harriet would have to host; artifact galleries; and one-off demo data tied to a single customer. Interactive behavior becomes numbered written steps instead, and the package names the connectors the organization should attach after approval.

Running the flow

1. Download the distiller. In Endpoint AI, open My AI or Help → Claude Desktop with your Claude subscription, and under Import session prior art select Download Session distiller (zip). Add the zip to your Claude session.
2. Distill in the same session. Load Session distiller in the session where the experiment lives and follow its instructions to produce SKILL.md and any reference files.
3. Generate an upload key. Back in Endpoint AI, select Generate upload key and copy the server, the hss_… token, and the expiry into Claude. The token is shown once and expires after about an hour; you can revoke keys from the same screen.
4. Submit for review. The distiller submits through Harriet's Skilify submit API using the upload key. The package appears in the review queue as a draft skill with status Submitted; it is not published to teammates until an approver acts.

After approval

Once a reviewer approves the skill (see Review workflow), admins roll it out through normal Endpoint AI provisioning: assign it via profiles, teams, or direct assignment (How skills reach people), and attach any connectors the skill depends on.