Save where you were, so you or your agent can pick it up later. Source on GitHub
Step 1 · About 2 minutes
park is one Go binary. It keeps your parked items in a local SQLite database.
$ go install github.com/svandragt/park@latest $ command -v park $ park version
Check that command -v park points into $(go env GOPATH)/bin or your GOBIN. If another park comes earlier on your PATH, that one runs instead.
The database is at ~/.local/share/park/park.db, or under $XDG_DATA_HOME when that is set. Set PARK_DB to use another file.
Unset PARK_SYNC_DIR as well as changing PARK_DB. While PARK_SYNC_DIR is set, every database you point park at reads and writes the same sync log, so test items reach your real items.
$ env -u PARK_SYNC_DIR PARK_DB=/tmp/park-test.db park listStep 2 · About 3 minutes
Park the thing you are in the middle of. Write it for the person who picks it up next, who may be you in a week or an agent with no memory of today.
Run this inside a git repository:
$ park add --name "Fix login redirect loop" \
--desc "Users bounce between /login and /dashboard" \
--why "Blocks the release" \
--how "Start at auth/session.go:42" \
--tags authparked #1: Fix login redirect loop
--name is the one line you see in lists. It is the only required field.--desc says what is wrong or what the work is.--why says why it matters, so you can decide later whether it still does.--how names the exact next step: a file and line, a command, a person to ask.--body holds longer notes. park edit 1 --append-body "..." adds to it later.park records the hostname, the git remote and the current branch for you.
Step 3 · About 3 minutes
Three commands cover most lookups: list what is open, search for a word, then show one item in full.
$ park list # every active item, in every repo $ park list --current # only this repo and branch $ park search redirect # full-text search, active items
#1 [active] Fix login redirect loop
Users bounce between /login and /dashboard
https://github.com/me/my-app (main)$ park show 1 $ park show - # the item you added last
# Fix login redirect loop [#1 · active] **Users bounce between /login and /dashboard** Why: Blocks the release How to apply: Start at auth/session.go:42 Type: task Tags: auth Repo: https://github.com/me/my-app Branch: main Device: my-laptop Parked: 2026-11-20 11:01
Search covers the name, description, body, why, how and tags. Add --status all to include resolved items.
Step 4 · About 1 minute
Resolve an item as soon as the work is done. A short list of open items is one you will read.
$ park done 1 # resolved $ park archive 1 # no longer relevant, kept for the record $ park reopen 1 # back to active
#1 marked as resolved
$ park add --name "Try out park delete" $ park delete - # removes the item you added last $ park prune --days 30 # removes resolved and archived items older than 30 days
Both commands are permanent. Use done or archive when you are unsure.
Step 5 · About 3 minutes
Tags group related items. Types say what kind of work an item is. A milestone collects the items that ship together.
Tag every item for a ticket with the ticket key, so one command finds them all:
$ park edit 1 --tags "auth,ABC-123" $ park list --tag ABC-123
The default type is task. The others are project, bug, feature, chore and docs. Filter with park list --type bug.
A milestone is an item of type project. Point other items at it with --parent, and park show counts how many are done.
$ park add --name "Q3 release" --type project $ park add --name "Write migration notes" --type docs --parent - $ park show 2
# Q3 release [#2 · active]
Type: project
Repo: https://github.com/me/my-app Branch: main
Device: my-laptop Parked: 2026-11-20 11:01
Children: 0/1 done
#3 [active] Write migration notes
https://github.com/me/my-app (main)To file an item under a repository you are not in, pass its remote: park add --name "x" --remote git@github.com:me/other.git.
Step 6 · About 5 minutes
With the hooks installed, the agent sees this repo's parked items when a session starts, and again when it opens a file an item names.
$ park hook # prints the config, changes nothing $ park hook --install # adds it to Claude Code and Codex settings
The installer writes to ~/.claude/settings.json and ~/.codex/hooks.json, for whichever exists, and keeps the hooks already there. Hooks load when a session starts, so open a new session afterwards. In Claude Code, /hooks reloads them.
At session start, the agent gets one line per active item for the current repository, most recently updated first. Claude Code passes on only the first 2 KB, so with many items the older ones may not fit:
Parked items here (id name [tags]). Surface anything relevant; do not act on it unprompted. Before a task: park search <words>; for a ticket, park list --tag <KEY>; park show <id> for detail. 3 Write migration notes 2 Q3 release 1 Fix login redirect loop [auth,ABC-123]
In Claude Code, a second hook runs when the agent reads or edits a file. If an active item names that file, the agent sees up to five matching items and can run park show for detail. Codex gets only the session-start hook.
The repository includes a Claude Code skill. With it installed, you can say "park this", "show parked" or "work on #2".
$ mkdir -p ~/.claude/skills/park $ cp .claude/skills/park/SKILL.md ~/.claude/skills/park/SKILL.md
Step 7 · Optional · About 5 minutes
Skip this step if you use park on one machine. With sync, each machine writes its own log file to a shared folder and reads everyone else's.
Point PARK_SYNC_DIR at a folder your file syncer keeps in step. Leave PARK_DB on its local default.
export PARK_SYNC_DIR="$HOME/sync/park"
$ park sync-seed --i-understand-this-runs-onceA second run, or a run on a second machine, duplicates every item. The command refuses if this machine's log already has entries.
$ park rebuild --yesrebuild replaces the local database with what the logs contain. Run it only when the logs hold everything you want to keep.
SQLite stores one database in several files. A file syncer copies them one at a time, and a half-copied set fails with database disk image is malformed. A log file has one writer and only grows, which syncers handle safely.
Item numbers such as #12 are local to each machine. When two machines change the same item, the later change wins.
Step 8 · About 2 minutes
Open your items in a browser, and check whether parked items come back into your work.
$ park serve # http://127.0.0.1:7654
The web view has search, a repository filter, and clickable tags and types. It is read-only and shows the items as they were when the server started.
$ park stats --days 7Usage in the last 7 days Events surfaced search 1 viewed cli 2 resolved cli 1 Surfaced items: 1 later viewed: 1 (100%) later resolved: 1 (100%) How items came back source surfaced viewed resolved session-hook 0 0 0 file-hook 0 0 0 search 1 1 1 list-tag 0 0 0 direct 0 1 0
Each view or resolve counts towards the way the item last reached you in the 2 hours before: the session hook, the file hook, a search or a tag list. direct means you went to the item without it being shown first. The record stays on this machine and does not sync.
Next time you are interrupted, run park add before you switch. When you come back, run park list --current.
Every command and flag is listed in the README, and for commands that take flags, park <command> -h lists them.