sanand0/scripts/anand-laptops

Machines

• Prefer graphene (Linux) when it's online. It's the primary machine, and its scheduled jobs keep the snapshots and the context index fresh.
• Use Plutonium (Windows) when graphene is offline, e.g. while Anand travels.
• It has the same ~/... layout, but its ~/Documents and ~/r2 may be missing or older.
• Name the sources you couldn't reach. Unreachable is not the same as absent.
• Prefer LocalMCP2 - which reaches only the graphene folders it was started with. Its ~ is /home/vscode, aliased to /home/sanand. A missing path there may just be unmounted.
• Remote Desktop Commander reaches both machines, so pick the device explicitly.
• Paths:
• When passing a path to a program, write "$HOME/...". That works in bash and in PowerShell; Windows PowerShell 5.1 doesn't expand ~ for programs.
• Desktop Commander's file tools accept ~/... on both machines.
• On Plutonium, avoid Cygwin bash: Windows programs can't open its /home/... paths.
• Sync:
• ~/Dropbox syncs live. On Plutonium only C:\Dropbox is live, so ignore dated or backup copies.
• ~/code holds separate git checkouts on each machine. Check branch and status; push from one machine and fetch on the other.
• ~/Documents, ~/r2, agent sessions and browser state are per machine.

Starting points

Read ~/code/scripts/agents/AGENTS.md for coding, shell, or filesystem-modification tasks: it sets tool preferences and working rules, and remote sessions don't load it.

• Unclear or cross-source: uv run "$HOME/code/scripts/context.py" search '…', if ~/Documents/data/context/ exists.
• Other subcommands: entity, recent, open-loops, thread, style, assets.
• Hits are candidates, so deep-read the ~/... locator.
• status shows each source's latest item. The index can be fresh while a source is weeks old.
• Code, projects: ~/code/README.md (repo index; if absent, list ~/code), then the repo's README.md and AGENTS.md.
• Blog, talks, demos, evals, data stories:
• ~/code/blog/description.md (one line per post)
• ~/code/talks/README.md
• ~/code/llmdemos/config.json
• ~/code/datastories/config.json
• ~/code/llmevals/README.md
• ~/code/til/README.md
• Notes, people, meetings, tasks: ~/Dropbox/notes/, especially about/<person>.md, transcripts/ (date-named), weekly-tasks.md and daily-*.md.
• Mail, chat, calendar, Drive:
• Live: gws. The default account is s.anand@straive.com.
• For root.node@gmail.com or anand@study.iitm.ac.in, set GOOGLE_WORKSPACE_CLI_CONFIG_DIR to ~/.config/gws-<account>.
• Confirm the account before any write. Setup: ~/Dropbox/notes/gws-cli.md.
• History: ~/Documents/data/<account>/ covers mail, chat and calendar. For Drive, use gws.
• Older mail: ~/Documents/Mail/ (its README explains the index).
• WhatsApp, browsing, LinkedIn: ~/Documents/data/README.md lists each archive and the script that builds it.
• ChatGPT and Claude conversations: ~/Documents/chatgpt/, ~/Documents/claude/.
• Coding-agent sessions: uv run "$HOME/code/scripts/agentlog.py" {codex,claude,copilot}, on each machine.
• Daily activity: ~/Documents/activities/.
• Large files, media: ~/r2/{files,media,private}/.
• Setup, personal CLIs: ~/code/scripts/README.md, ~/code/scripts/setup/{linux,windows}.md, ~/Dropbox/scripts/.
• Browser control (graphene): the devtools skill.

Workflows

• Transcribe meeting recordings on graphene
• Convert video to speech-quality Opus using the opus function in ~/code/scripts/setup.fish. Save in ~/Documents/calls/ as YYYY-MM-DD Meeting Title.opus.
• Transcribe via ~/code/scripts/call into ~/Dropbox/notes/transcripts/*.md. If speaker-labeled captions exist (e.g. embedded subtitles), use ## Meeting transcript context in ~/code/blog/pages/prompts/fragments.md to generate speaker context for call --prompt - along with attendee names, meeting notes, ...
• Reuse existing files. Verify completion.

Answering

Label each source as one of:

• live (gws)
• snapshot, with its date
• index hit, not yet deep-read
• inference