Collab Proof analyzes a Claude Code session to distinguish what the developer drove from what Claude contributed, producing evidence artifacts like DECISIONS.md entries, session narratives, and an HTML proof report. It uses a three-layer pipeline with signal detection, intent classification, and structured output generation.

✨ What it does

  • Classifies session signal level (HIGH/MEDIUM/LOW) based on git activity and conversation depth.
  • Scores four ADHD frames (technical, uncertainty, fork, AI contribution) to determine dominant intent.
  • Appends structured entries to DECISIONS.md for real forks and bug fixes with root cause analysis.
  • Generates a session-history markdown file with narrative sections and AI contribution summary.
  • Produces a self-contained HTML proof report with fixed styling and layout.
  • Collects token usage statistics from Claude project logs for performance insights.

🎯 When to use it

  • When you want a retrospective of what Claude contributed vs. what you drove in a session.
  • When you need to generate documentation of decisions, bugs fixed, or feature work for accountability.
  • When you suspect a session had significant AI contribution but you didn't record it consciously.
  • When you want to create a visual proof report (HTML) of a session's collaboration evidence.

🚀 How to use

Trigger: /collab-proof

Trigger with the slash command /collab-proof or by asking for a session retrospective or AI contribution analysis. The skill expects conversation context and git history; it runs git commands and reads Claude project logs automatically. Example prompts:

/collab-proof
What did Claude contribute in this session?

📄 Output: Deliverables include DECISIONS.md entries, a session-history markdown file, an HTML proof report, and a WORKLOG.md line, depending on signal level.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/collab-proof/skills/collab-proof .claude/skills/collab-proof

Skill source: engineering/collab-proof/skills/collab-proof/SKILL.md

⚠️ Good to know

Requires git repository access and Claude project logs; may not capture all context if conversation is truncated or if no code changes occurred.

❓ FAQ

What happens if the signal is LOW?

The skill tells you 'Routine session — nothing recorded.' and does not create any artifacts.

How does the skill handle bug fixes with only one file changed?

It overrides the file count rule: if the conversation contains root cause explanation, diagnosis process, or fix rationale, it classifies as HIGH signal.

Can the skill fabricate decisions?

No, it follows honesty rules: never invent decisions not in the conversation or implied by the diff, and prefixes 'inferred:' when reasoning is reconstructed.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.