Session Feedback

Evaluate AI session interaction logs and generate actionable feedback to help users improve their AI-assisted development workflow.

Overview

This skill reads the session reports and log files produced by session-interaction-logger and produces a structured feedback report. The report highlights what the user did well, identifies missed opportunities, and provides concrete suggestions for future sessions.

Read Session Logs → Analyze Patterns → Evaluate Against Rubric → Generate Feedback Report

When to Use This Skill

• •After completing a coding session to get retrospective feedback
• •When asked to "review session", "evaluate logs", or "give feedback"
• •Before starting a new session to review lessons from the last one
• •When preparing for an AI-usage evaluation or audit

Input Sources

The skill reads from the following files (produced by session-interaction-logger):

If structured JSONL logs are not available, the skill falls back to analyzing the Markdown session reports.

Output

A feedback report written to:

logs/feedback/report-{YYYY-MM-DD}.md

Evaluation Rubric

The feedback report evaluates the session across these dimensions:

1. Prompt Quality (Weight: 25%)

What to look for:

• •Did the user reference design docs, files, or specific requirements?
• •Were prompts self-contained or did they require many follow-up clarifications?
• •Did the user specify constraints (language version, patterns, testing expectations)?

2. Planning & Specification (Weight: 20%)

What to look for:

• •Was a plan/spec document created before coding?
• •Did the user review the plan and provide feedback?
• •Were steps broken down into manageable pieces?

3. Iterative Refinement (Weight: 20%)

What to look for:

• •Did the user request changes or improvements after initial output?
• •Were tests run and failures addressed?
• •Did the user verify behavior matches expectations?

4. Decision Documentation (Weight: 15%)

What to look for:

• •Were architectural choices explained?
• •Were alternatives considered and trade-offs discussed?
• •Can someone reading the logs understand why decisions were made?

5. Testing & Verification (Weight: 20%)

What to look for:

• •Were tests requested as part of implementation?
• •Were build/test commands run to verify correctness?
• •Were test failures analyzed and resolved?

Step-by-Step Workflow

Step 1: Gather Session Logs

1. •List all files under logs/copilot/
2. •Identify session report files (session-report*.md)
3. •Check for structured JSONL logs (interactions.log, decisions.log, etc.)
4. •If a specific date is requested, filter to that session; otherwise analyze the most recent

Step 2: Analyze Session Content

Read each log source and extract:

• •Prompts: User requests — count, specificity, context provided
• •Iterations: How many rounds of refinement occurred per task
• •Decisions: Architectural choices, rationale, alternatives
• •File changes: Volume, organization, whether tests accompanied code
• •Commands: Build/test execution, success/failure patterns
• •Clarifications: How often the AI needed to ask for more info

Step 3: Score Against Rubric

For each rubric dimension:

1. •Review the relevant evidence from the logs
2. •Assign a rating: Excellent, Good, or Needs Work
3. •Provide specific examples from the session to justify the rating
4. •Calculate a weighted overall score

Step 4: Generate Suggestions

Based on the evaluation, produce:

• •Strengths: 2-4 things the user did well (with examples)
• •Improvement Areas: 2-4 areas to focus on (with examples)
• •Actionable Tips: 3-5 concrete things to try in the next session
• •Prompt Templates: Example improved prompts based on patterns observed

Step 5: Write Feedback Report

Generate the report at logs/feedback/report-{YYYY-MM-DD}.md using the structure below.

Report Template

The generated feedback report follows this structure:

# Session Feedback Report

**Date**: {date}
**Session(s) Analyzed**: {session IDs or report file names}
**Generated By**: session-feedback skill

---

## Overall Score: {X}/100

| Dimension | Weight | Rating | Score |
|-----------|--------|--------|-------|
| Prompt Quality | 25% | {rating} | {score}/25 |
| Planning & Specification | 20% | {rating} | {score}/20 |
| Iterative Refinement | 20% | {rating} | {score}/20 |
| Decision Documentation | 15% | {rating} | {score}/15 |
| Testing & Verification | 20% | {rating} | {score}/20 |

---

## Strengths

{Bulleted list with specific examples from the session}

## Areas for Improvement

{Bulleted list with specific examples and why they matter}

## Actionable Suggestions for Next Session

{Numbered list of concrete things to try}

## Prompt Improvement Examples

{Before/after prompt examples based on patterns observed}

---

_Generated on {timestamp} by session-feedback skill_
_Source logs: logs/copilot/_

Script Usage

Generate a feedback report from the command line:

.github/skills/session-feedback/scripts/generate-feedback-report.sh [YYYY-MM-DD]

• •If a date is provided, it scans for session-report-{date}.md
• •If no date is provided, it analyzes all available session reports
• •Output is always written to logs/feedback/report-{date}.md

Best Practices

References

• •Evaluation rubric details: references/evaluation-rubric.md
• •Session interaction logger skill: ../.github/skills/session-interaction-logger/SKILL.md