NICHOLAS WHITE
Back to projects and tools

DOCUMENTATION TOOLS / SOURCE REVIEW

Learn from Source

I built Learn from Source to have AI review articles, documents, and repositories against the tools and instructions I was using. It read the source, assessed the useful ideas, checked them against my environment, and prepared specific changes. After review, it applied the accepted changes and recorded their source.

MY CONTRIBUTION

I designed the source-review workflow, its evaluation criteria and review steps, and adapted it through use. AI performed the reading, analysis, proposed edits, and recordkeeping; I also used AI coding assistance to develop it.

TOOL TYPE

An automated source-review skill in Prompts and Agents, using the assistant’s reading, project inspection, editing, and logging tools.

THE WORK

What I wanted from a source review

I used this workflow when I found an article, document, or repository with ideas that might help my projects. I supplied the source and the project context; the assistant gathered and analyzed the material and worked out which changes were useful.

AI compared the source with the project’s existing instructions and tools. It checked for duplication, conflicting rules, and assumptions that did not fit the working environment before producing concrete edits for review.

WORKFLOW

How AI turned a source into a proposed change

The assistant retrieved the supplied source, filtered it for relevant findings, and assessed credibility and applicability. It scored findings for applicability, novelty, and confidence, then located the actual instructions, skills, or notes that would benefit.

AI source review with a decision before changes
  1. 01AIRead the sourceRetrieve the material and identify ideas relevant to the project.
  2. 02AICheck and scoreAssess credibility and fit, then inspect existing project instructions.
  3. 03AI + NickPropose and reviewPrepare exact edits and their expected effects. I review the proposed changes.
  4. 04AIApply and recordApply accepted changes and log their source and the decision.

The assistant handled the research, comparison, edits, and recordkeeping. Findings that did not fit the project were recorded without changing the files.

The outputs included a source analysis, a proposed-edit table, an environment and impact review, approved file changes, and a knowledge log recording what was accepted or rejected. The review covered effects on other workflows as well as the file being edited.

RECORDED EXAMPLE

A change to Daily Reflection

In one recorded use, the assistant reviewed Tamir Dresher’s Squad Skills repository and identified a correction-capture pattern that could improve my existing reflection workflow: save explicit corrections during a session so they are available when the session is reviewed.

Simplified explanation based on a recorded change · not a transcript or original output

  1. Source idea: Capture an explicit correction when it occurs.
  2. Project check: Compare that behavior with the current reflection instructions.
  3. Proposed change: Add an early-capture step and preserve the source credit.
  4. After review: Update the reflection instructions and record the accepted idea.

After my review, the assistant updated Daily Reflection and recorded the source credit. That addition let the reflection workflow retain explicit corrections as they happened during later sessions.

DEVELOPMENT & USE

What supports that example

  • A preserved working sessionThe saved session records the source review and my approval of the resulting changes.
  • A knowledge-log entryThe log records the review and the accepted correction-capture idea.
  • A matching implementation changeThe development history adds the new section to Daily Reflection and credits Dresher’s Reflect plugin for that pattern.

These are related development records for the same change. The private conversation and working files are not embedded in this page.

VERSIONS & CREDIT

The working skill and simpler variants

The working toolkit version proposed changes to instructions, skills, and memory. A generic adaptation replaced environment-specific tools and paths. I also developed a simpler standalone prompt that asked about a person’s focus and note preferences, then read, filtered, scored, and saved useful material.

Those versions had different scopes. The simpler notes prompt did not contain every operational step in the full toolkit workflow, and the presence of a generic copy does not establish a reviewed public release.

I developed the workflow with AI coding assistance. The broader skill-learning toolkit credited Karpathy’s autoresearch as a design influence. The correction-capture idea in the example is credited to Tamir Dresher’s Reflect plugin.