An assistant specification
Write down who the assistant serves, how it should respond, what sources it uses, and where it should stop. Export the instructions and supporting material for a separately chosen platform.
Inside the AskJamie workshop
AskJamie Found‑Ry
Give useful thinking a home.
A clear explanation is a good beginning. Found-Ry is where that explanation becomes an assistant brief, a decision guide, or a workflow someone can return to.
Working local alpha · Private workbench
This page opens the workshop door a little. The application runs locally for its maintainer; a public, hosted version is not available.
Why it exists
Keep the reasoning with the result.
A useful answer deserves more than a place in an old conversation.
When a question comes around again, it helps to know what the answer was built on: the source, the audience, the choices, and the limits. Found-Ry gives those details a place beside the capability itself.
It grows out of AskJamie’s interpretive work: making a complicated situation understandable enough that someone can choose a next step. The workbench turns that approach into something you can save, test, revise, and hand over.
Three starting points
Choose the shape that fits the question.
A decision tool
Turn a repeatable choice into questions with Yes and No paths. Preview the outcomes, test expected results, and export a self-contained HTML decision guide.
A guided workflow
Put a known sequence into steps someone can follow. Keep the purpose, instructions, and source alongside the checklist so the next person can understand the process.
Inside the workbench
From the first brief to a useful handoff.
The work stays editable. A new revision gives the next decision a traceable starting point.
01 · Frame
Name the job
Define the purpose and audience. Keep the original source and a note about where it came from.
02 · Shape
Make the behavior clear
Write instructions, constraints, and the form a useful response should take. Add decision branches or ordered steps where they help.
03 · Reference
Bring in a useful pattern
Search Skillz metadata and attach relevant references. Each keeps its source, maturity, and evidence status.
04 · Save
Save the revision
Reopen saved work and inspect its history. Evaluation results stay associated with the revision that was checked.
05 · Check
Try the important cases
Run decision paths against expected results. For assistant drafts, check a response you supply against required and forbidden text.
06 · Package
Prepare the handoff
Validate the draft and export a private ZIP with its specification, sources, instructions, tests, and proposed registration record.
A small example
Is the request clear enough to proceed?
A decision guide can make even this modest choice explicit.
Yes: proceed with the agreed scope.
The result names what happens next. A test checks that answering Yes reaches this outcome.
No: ask for the missing context.
The alternate path has its own outcome and test. Uncertainty becomes a useful next question.
Illustrative example. In the workbench, these branches execute as a deterministic decision graph. No AI model is needed to choose the path.
Where it belongs
A regional workshop that learns in both directions.
OverKill Hill is the centroid and baseline pattern for the universe. Its Found-Ry mentors the AskJamie and Glee-fully Found-Rys. Each regional workshop can adapt that pattern, offer improvements back, or help its peer.
AskJamie’s contribution is interpretation: understandable behavior, visible decision paths, and enough context for the next person to pick up the work.
Current scope
What the working alpha can promise.
Local, saved, and reviewable
- Drafts and revision history persist in a local database.
- Decision previews and recorded evaluations are implemented.
- Packages preserve source, lineage, and privacy controls.
- The workbench uses Python, SQLite, and a browser interface.
Further steps remain deliberate
- Assistant specifications still need configuration and testing on their destination platform.
- Workflow checklists do not execute external actions.
- Text checks on supplied responses do not prove model quality.
- Exporting a package does not publish it or create a remote repository.
The workbench makes no model-provider calls. Its privacy controls belong to the local application; this public feature page follows the site privacy notice.
Maturation
Prove a useful capability, then grow from there.
The next steps are proposed work, not a release schedule.
Implemented
The local building cycle
Authoring, persistence, decision previews, evaluation records, Skillz references, and private package generation form the current alpha.
Next
Real work and a dependable handoff
Pilot useful AskJamie capabilities. Complete accessibility, narrow-screen, exported-runner, and backup-restoration checks.
After the pilot
A destination that earns its place
Choose a platform adapter for a proven use case. Consider optional AI assistance or authenticated hosting when the work calls for it.
A few practical questions
Before you pull up a chair.
Can I open the workbench here?
The current application is a private local alpha. This page describes the work; it does not provide public access. Contact Jamie to discuss a capability or walkthrough.
How does it relate to the Lens System?
The Lens System presents focused AskJamie experiences and case studies. Found-Ry provides a place to develop and package capabilities; it does not automatically connect the lenses or their data.
Does it build the assistant for me?
You supply the brief, sources, instructions, and example responses. The application organizes and checks that material. It does not silently generate an assistant or configure a vendor account.
Bring the question
What would be useful enough to use again?
A recurring question, a difficult handoff, or a decision that needs a clearer path is a good place to begin.