AI Session Contribution Attribution Report
Evidence-backed breakdown of what you drove vs what Claude contributed in a session: decisions, code, catches, and corrections.
Before & after
Same prompt, same agent — the only difference is the skill.
That's an interesting question. I can try to look back at what happened in our conversation and break it down. Typically, I might have helped with brainstorming, offering suggestions, or refining ideas you already had. You probably came up with the core concept and direction. I might have filled in details or asked clarifying questions. It's hard to say exactly without looking at the specific conversation, but usually it's a mix of your ideas being shaped by my input and suggestions. Different people value collaboration differently, so what counts as a contribution might vary.
COLLABORATION BREAKDOWN YOU DROVE (core direction) → Chose design system color palette (brand primary + accent) → Decided on 5-column grid vs 6-column (you tested both mockups) → Selected navigation pattern (top bar vs sidebar) → Reviewed competitive set and rejected 3 approaches → Final sign-off on design: 100% your decision CLAUDE CONTRIBUTED (refinement + iteration) → Generated 8 design variations given your constraints (30 min) → Named components following your naming convention → Audited accessibility (contrast, spacing, touch targets) → flagged 2 issues, you fixed → Suggested typography pairing (you approved without change) → Wrote CSS-in-JS template (you adapted to your framework) → Documented 12 component specs → you used 11, modified 1 TO-TALS Decisions: 12 | You made: 8 (67%) | Claude made: 4 (33%) File outputs: 8 | You authored: 2 | Claude generated: 6 Time saved (estimated): 2.5 hours of iteration + documentation BOTTOM LINE → You controlled direction, vision, final approvals → Claude handled speed (faster variation) and polish (specs, a11y) → High collaboration ratio (33% Claude input is typical for design-led projects) FILE: collab-proof.json logged in project memory
About this skill
You finished a long session with Claude and now someone is asking how much of it was you. That could be your manager, a PR template, a course policy, or you. A vague "we worked on it together" is useless, and a single "70% AI" number is worse because it hides who actually made the calls. This skill walks the session and produces a decision-by-decision attribution, with each claim tied to the moment it happened.
What it does
- Splits the work into five separate axes (direction, design decisions, execution, discovery, verification) instead of one misleading percentage
- Builds a decision ledger where every significant call is tagged with its origin: you specified it, you picked from options Claude offered, Claude proposed and you accepted, Claude decided alone, or you overrode Claude
- Anchors every attribution to a turn, a short quote, a file, or a commit, and labels anything it cannot anchor as unattributable instead of guessing
- Flags the judgment calls Claude made without being asked, so you can review the ones that touched behavior, security, data, or dependencies
- Names its own blind spots: compacted context, edits you made outside the session, CLAUDE.md rules you wrote earlier, and tool output that neither of you authored
- Shapes the result for its purpose: a team retrospective, a PR or assignment AI-use disclosure, portfolio evidence, or a lesson on how you delegate
What you get
A short summary of who drove each axis, a decision ledger with evidence, a list of unreviewed agent judgment calls, the stated limits of the analysis, and an optional ready-to-paste disclosure statement.
Best for
Developers who need to disclose AI assistance on PRs or client work, students and researchers writing AI-use statements, engineering leads running retrospectives on AI-assisted work, and anyone who wants an honest read on how they actually use Claude.
SKILL.md preview
---
name: ai-session-contribution-attribution
… (sign up to view the full skill)More development skills
View all Development skills →Performance Bottleneck Profiling: Node, Python, Go
Classifies a slowdown, gives exact profiler commands, reads the output, and proves the fix with same-conditions p95/p99 numbers.
Feature Scaffolding with Working Tests
Turn a short feature spec into code, runnable tests, and integration wiring that match your existing project.
Feature Flag Rollouts and Kill Switches
Design staged flag rollouts, verify rollback paths, and audit flag debt with explicit defaults, gates, and retirement plans.
Microcontroller Firmware and RTOS Engineering
Build and debug MCU firmware with bounded ISR/RTOS timing, safe DMA ownership, and measured power budgets.