Skip to content

Track plan mode, auto-approve and Auto model; list skills; read Copilot plugins and Cursor's database - #8

Merged
Expecho merged 1 commit into
mainfrom
adoption-modes-skills-cursor-db
Nov 4, 2026
Merged

Expecho merged 1 commit into
mainfrom
adoption-modes-skills-cursor-db

Conversation

@Expecho

@Expecho Expecho commented Nov 4, 2026

Copy link
Copy Markdown
Owner

Summary

Adoption report

  • Plan mode and Auto-approve columns in the adoption cohorts and the weekly phase mix, plus rows in Agent adoption.
    • Auto-approve means running without approving each step: Claude Code's auto mode, or Copilot's autopilot.
    • Cursor reports leave Auto-approve out, because Cursor doesn't log auto-run.
  • Claude Code sessions now record their permission modes: permissionMode on prompts, permission-mode records, and EnterPlanMode/ExitPlanMode calls.
  • Auto model row for Cursor, from the model picked per chat ("default" is Cursor's Auto). It's a share of the sessions that log a model choice.
  • The "Hand over whole tasks" recommendation now gives plan mode and auto mode counts for Claude Code too.

Skills

  • Every period report (session, daily, weekly, monthly, project) ends with the full Skills table, as in the summary.
  • --top limits only the breakdown sub-rows, not this list.

Plugins

  • Installed plugins now include GitHub Copilot's:
    • The list comes from installedPlugins in ~/.copilot/config.json, which can contain // comment lines.
    • The enabled state comes from settings.json.
  • A plugin's skills count whether Copilot logs them as plugin:skill or by the bare skill name.
  • A Product column appears when the plugins come from more than one product.

Cursor's database

  • New src/cursor-state.mjs reads the editor's state.vscdb read-only, with node:sqlite.
    • That needs Node 22.13 or later. Older Node skips the database and prints a notice.
    • Only metadata is read: mode, chosen model, title, workspace, message type and time, tool name and command. Chat text and tool results are never read.
  • Chats that have no transcript (older editor chats) are now counted. Token counts are left out, because Cursor fills them in for very few messages.

README

  • Documents all of the above.
  • Fixes the Cursor note on subagents: launches are counted, but a subagent's own work isn't.

Test plan

  • npm test: 53 passing. New tests cover permission modes, the plan and auto columns, the Skills table in period reports, Copilot plugins, and a Cursor database fixture.
  • Ran adoption, session -b -a -r --source all and --source cursor against real local logs.

🤖 Generated with Claude Code

…ot plugins and Cursor's database

- Adoption: Plan mode and Auto-approve (Claude Code auto mode, Copilot
  autopilot) as cohort columns, weekly phase-mix columns and capability rows.
  Claude Code sessions now record their permission modes. Cursor reports
  leave auto-approve out, since Cursor doesn't log auto-run.
- Period reports (session, daily, ...) end with the full Skills table, as in
  the summary; --top limits only the breakdown sub-rows.
- Installed plugins include GitHub Copilot's (config.json, enabled state from
  settings.json), with a Product column when plugins span products.
- Cursor: read the editor's state.vscdb (node:sqlite, Node 22.13+; skipped
  on older Node) for the model picked ("default" = Auto, shown as an Auto
  model row), the mode per chat, and chats that have no transcript. Only
  metadata is read.
- README: document the above and fix the Cursor subagents note.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Expecho
Expecho merged commit ad45c4c into main Nov 4, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant