resume agent work — a tiny CLI to label and resume Cursor agent sessions.

Cursor agent sessions are identified by UUIDs. Once you've got a dozen of them going you lose track of which is which. raw lets you pin a human-readable label to a cursor-agent --resume=<id> command and pick it back later with an fzf picker.

# after an agent run, stash it with a label
raw save "Work on skip events" agent --resume=b281f6e3-917c-441f-a5dc-dacadd52d059

# later, from anywhere — pick from a fuzzy list, hit Enter, you're back in the agent
raw continue    # or: raw c

resume ›
> [api-mobile] Work on skip events   —   2026-04-20T10:13:48
  [api-mobile] audio-capacity2       —   2026-04-20T10:13:27
  [web]        Fix login bug         —   2026-04-19T18:02:11

 label:   [api-mobile] Work on skip events
 saved:   2026-04-20T10:13:48
 agent:   b281f6e3-917c-441f-a5dc-dacadd52d059
 cwd:     /Users/you/code/api-mobile

 command:
   agent --resume=b281f6e3-917c-441f-a5dc-dacadd52d059

You probably have this problem if:

• You run several Cursor agents in parallel across different repos.
• Half-finished sessions pile up and you can't remember what each one was doing.
• You want a zero-ceremony way to "pick up where I left off" without scrolling back through terminal history.

• Fuzzy picker via fzf with a live preview pane showing label, agent id, cwd and the full command.
• Restores the working directory you were in when you saved — so resuming drops you back into the right repo.
• Plain-text storage at ~/.resumer/history (TSV). Grep it, edit it, sync it across machines, version it if you're that kind of person.
• Dedupe on agent id — re-saving an existing session updates the label and bumps it to the top instead of duplicating.
• Short aliases: raw s, raw c, raw l.
• fish completions out of the box, including label suggestions for raw rm.
• No runtime deps beyond bash + coreutils + optional fzf. Numbered-menu fallback when fzf is missing.

• bash (tested with 3.2 / 5.x)
• awk, column, cut (any POSIX system has these)
• fzf — strongly recommended for the picker. Install with brew install fzf / apt install fzf. Without it, raw continue falls back to a numbered menu.
• fish — only if you want the shipped shell completions. The raw command itself is a plain bash script and runs anywhere.

git clone https://github.com/glava/raw.git ~/src/raw
cd ~/src/raw
./install.sh

The installer creates symlinks (not copies), so git pull instantly updates your installed command:

• ~/.local/bin/raw → bin/raw
• ~/.config/fish/completions/raw.fish → completions/raw.fish

If ~/.local/bin isn't on your PATH, add it in ~/.config/fish/config.fish:

fish_add_path ~/.local/bin

Uninstall any time with:

rm ~/.local/bin/raw ~/.config/fish/completions/raw.fish
rm -rf ~/.resumer   # optionally nuke the history

Everything lives in one human-readable TSV file:

~/.resumer/history

Format — one session per line, tab-separated:

<iso_timestamp>	<label>	<agent_id>	<cwd>	<command>

• Newest entries are kept at the top.
• Saving a command with an agent id that already exists replaces the old row.
• Override the location with RAW_HOME=/some/dir (useful for testing or sandboxing).

• raw continue cds back to the directory you were in when you ran raw save, then execs the command. Your shell becomes the agent.
• If the original directory no longer exists, raw warns and runs in the current cwd — the agent session itself is cwd-independent.
• Labels may contain spaces and punctuation, but not tab characters (they'd corrupt the TSV).

raw/
├── bin/raw                 # the script itself
├── completions/raw.fish    # fish completions
├── install.sh              # symlink installer
├── LICENSE                 # MIT
└── README.md

Small focused PRs welcome. The whole tool is a single bash script — keep it that way unless there's a very good reason not to.

MIT © 2026 glava

Built for use with Cursor. Not affiliated with Anysphere.