dui learning

Capture Doing–Using–Interacting (DUI) learning to turn fleeting project experiences into durable, evidence-backed judgment records. Most engineering decisions are lost in Slack threads, PR comments, or unrecorded experiments, forcing teams to solve the same problems twice. This skill identifies meaningful milestones and reconstructs the rationale behind tradeoffs, failed attempts, and tool discoveries.

What it does

• Identify milestones by scanning experiment notes, diffs, and chat logs for consequential changes in approach.
• Reconstruct rationale to capture the "why" behind decisions, citing specific test results, stable links, or message dates.
• Formulate conditional judgments that define when to consider an action based on specific observed evidence.
• Maintain versioned records that preserve prior reasoning while updating judgments as new data emerges.
• Distinguish evidence types by labeling agent interpretations, participant explanations, and raw observations separately.

How it works

1. Context Analysis: The skill scans your current task, diffs, or conversation history for surprising outcomes or revised decisions.
2. Knowledge Retrieval: It searches existing judgment-records/ to see if the current situation aligns with or contradicts prior findings.
3. Record Drafting: It generates a Markdown record documenting objective, constraints, attempts, outcomes, and a conditional judgment.
4. Validation: It labels the record as "provisional" or "supported" based on the completeness and source of the evidence.

Frameworks & tools

This skill produces standard Markdown files. It is designed to work within any git-based repository, documentation folder, or local judgment-records/ directory without requiring external databases or proprietary connectors.

Why this beats prompting it yourself

Generic prompts often result in vague "lessons learned" that lack the specific evidence needed for reuse. This skill enforces a strict separation between raw observation and inferred rationale, ensuring records are grounded in verifiable artifacts like commit hashes and test logs rather than LLM hallucinations.

Use cases

• Documenting why a specific library was rejected after a 2-day spike.
• Recording the specific conditions that cause a flakey integration test to fail.
• Capturing architectural tradeoffs made during a high-pressure refactor.
• Preserving coordination feedback from senior reviewers to prevent future regressions.

Known limitations

Does not support background monitoring or automatic chat retrieval outside the provided context. Requires a project structure that allows Markdown file storage for persistence.