AI Feedback Loop Rules

Turn feedback into versioned, reversible prompt and rule improvements.

v1StandaardTerug naar overzicht

Inhoud bewerken

Pas de instructie aan. Opslaan maakt een nieuwe versie en is terug te draaien.

Gelijk aan standaard

Voorbeeld

De huidige opgeslagen inhoud.

Toon inhoud (1.218 tekens)

# AI Feedback Loop Rules

How Assembly Hangar improves AI behavior from feedback. This is not model
training. It is a structured feedback loop that improves prompts, rules,
templates and agent behavior over time.

## Process
- Store feedback on AI-generated output where relevant.
- Analyze what the user corrected.
- Identify whether the issue was: tone, structure, missing context, wrong assumptions, wrong formatting, incomplete task, bad UX decision, technical mistake, unsafe change, or bad prioritization.
- Convert repeated feedback into prompt or rule improvements.
- Version prompt and rule changes; keep rollback possible.
- Do not automatically rewrite core rules without review.
- High-impact rule changes require human approval.
- Keep a changelog of prompt and rule changes.

## Feedback loop output
- Original output issue
- User correction
- Root cause
- Proposed rule or prompt improvement
- Risk level
- Version update
- Rollback note

## How this connects to this module
The AI Instructions admin module is the editable home for these rules. Every
edit is versioned and reversible, so improvements identified through feedback can
be applied here, reviewed, and rolled back if they cause regressions.

# AI Feedback Loop Rules How Assembly Hangar improves AI behavior from feedback. This is not model training. It is a structured feedback loop that improves prompts, rules, templates and agent behavior over time. ## Process - Store feedback on AI-generated output where relevant. - Analyze what the user corrected. - Identify whether the issue was: tone, structure, missing context, wrong assumptions, wrong formatting, incomplete task, bad UX decision, technical mistake, unsafe change, or bad prioritization. - Convert repeated feedback into prompt or rule improvements. - Version prompt and rule changes; keep rollback possible. - Do not automatically rewrite core rules without review. - High-impact rule changes require human approval. - Keep a changelog of prompt and rule changes. ## Feedback loop output - Original output issue - User correction - Root cause - Proposed rule or prompt improvement - Risk level - Version update - Rollback note ## How this connects to this module The AI Instructions admin module is the editable home for these rules. Every edit is versioned and reversible, so improvements identified through feedback can be applied here, reviewed, and rolled back if they cause regressions.