readwise-methods-reviewlisted

# Readwise methods review ## What this is for One question, asked of a week of reading: **does anything here change how the user works with models?** Not "does it discuss AI." Most reading corpora are mostly domain material with a minority of AI commentary, and most of that commentary is forecasting rather than method. Judge each item against what you would actually do differently on Monday. The baseline to beat is high. You already use structured task prompts, standing instructions, investigation loops, adversarial subagents, project notes, and persistent vault knowledge. "Be specific in your prompts" is not a finding. ## Access — decided, don't re-derive **Use the Readwise MCP server** (`mcp2.readwise.io`, OAuth). This was settled 2026-07-25: - `@readwise/cli` is **not installed** and there is **no `READWISE_TOKEN`** in the environment, shell profile, or `~/.config/`. The `readwise-cli` skill documents a path that does not work on this machine. - Browser automation is strictly worse — MCP returns complete document bodies. - Re-check the environment each run in case a token appears, but expect MCP. **Never print or store credential values.** ## Workflow ### Phase 0 — window and fetch ```bash python3 ~/.claude/skills/readwise-methods-review/triage.py window ``` Prints the local window, the `updated_after` value to pass to the API, and the output filename. Then fetch both locations, saving each raw response: - `reader_list_documents(location="new", updated_after=<f