Initial commit: AI conversation impact methodology and toolkit

CC0-licensed methodology for estimating the environmental and social
costs of AI conversations (20+ categories), plus a reusable toolkit
for automated impact tracking in Claude Code sessions.

plans/reusable-impact-tooling.md

@@ -0,0 +1,42 @@
+# Plan: Make the impact measurement tooling reusable
+
+**Target sub-goals**: 7 (reach), 8 (teach), 9 (outlast the conversation)
+
+## Problem
+
+The PreCompact hook, impact log, and show-impact script work but are
+hardcoded to this project's directory structure and Claude Code's hook
+system. Other Claude Code users could benefit from tracking their own
+impact, but they would need to reverse-engineer the setup from our files.
+
+## Actions
+
+1. **Package the tooling as a standalone kit.** Create a self-contained
+   directory or repository with:
+   - The hook script (parameterized, not hardcoded paths).
+   - The show-impact viewer.
+   - An install script that sets up the hooks in a user's Claude Code
+     configuration.
+   - A README explaining what it measures, how, and what the numbers mean.
+
+2. **Improve accuracy.** Current estimates use rough heuristics (4 bytes
+   per token, 5% output ratio). Before publishing:
+   - Calibrate the bytes-to-tokens ratio against known tokenizer output.
+   - Improve the output token estimate (currently a fixed fraction).
+   - Add water usage estimates (currently missing from the tooling).
+
+3. **Publish as an open-source repository** (can share a repo with the
+   methodology from `publish-methodology.md`).
+
+## Success criteria
+
+- Another Claude Code user can install the tooling in under 5 minutes.
+- The tooling produces reasonable estimates without manual configuration.
+
+## Honest assessment
+
+Moderate leverage. The audience (Claude Code users who care about impact)
+is niche but growing. The tooling is simple enough that packaging cost is
+low. Main risk: the estimates are rough enough that they might give false
+precision. Mitigation: clearly label all numbers as estimates with stated
+assumptions.