| name | quick |
| description | Use for errands BETWEEN "answer inline" and "file a ticket" — a web lookup on a live site, a small one-off script, a repo you need summarized. Dispatches a short-lived specialist agent via `beckett quick` and hands you its report; no ticket, no worker, no worktree. |
quick — the no-ticket lane
Some asks are too heavy to answer from your own head but far too light for a tracker ticket:
checking a price on a live site, writing a 40-line conversion script, telling someone what's
in a GitHub repo. Filing a ticket for these wastes a worker and makes a 2-minute answer take
ten. Doing them yourself bloats your context with browsing or a whole codebase. Instead,
dispatch a quick agent and relay its report.
The agents
| agent | give it | it returns |
|---|
quick-code | a small coding errand ("script that converts this CSV to JSON", "why does this snippet throw") | the answer + absolute paths of any files it made (in a scratch dir) |
repo-explorer | a repo + a question ("what does owner/name do, how do I run it") | a ~250-word brief answering the question, with file paths |
beckett quick list prints this menu. Browser / computer-use work is NOT in this lane — it
runs in the dedicated background browser agent (beckett browser "<task>", see the browser
skill), which can pause for a human answer mid-run and resume; quick agents cannot.
How to dispatch
beckett quick quick-code "write a script that dedupes the attached wordlist; input at /path/x.txt" --channel <id>
beckett quick repo-explorer "clone anthropics/claude-code and tell me how its hook system works" --channel <id>
- Always pass
--channel <id> (from the turn stamp). If the run outlives the sync
window, the result routes back to that channel through you — without it, a slow result
has nowhere to go.
- Write the task like a good ticket one-liner: everything the agent needs is IN the task
text (URLs, paths, the actual question, any email/name it should use). The agent knows
nothing about the conversation and can't ask follow-ups.
- The command blocks up to ~4 minutes, so ack first (
beckett discord reply) like any
task. Two outcomes:
- A report — relay it in your voice with a second
beckett discord reply (after a
CLI ack, plain turn text won't auto-post).
- "still working (run … detached)" — your ack already covers it; end the turn. The
report arrives later as a
quick-agent result update turn; relay it then with
beckett discord reply --channel <id>.
When NOT to use this lane
- Real work — anything you'd want reviewed, anything multi-file, anything building on a
project repo (
~/Projects/*) → file a ticket as usual. Quick agents are forbidden from
project repos by their own doctrine.
- Things you already know or can read in one file — just answer inline; a quick agent
is still a whole model spin-up.
- Anything owner-gated or destructive — approvals, account deletions, payments. Bring
those to the owner instead of dispatching an agent at them.
- Anything touching a live website or browser — that's the background browser agent
(
beckett browser), never this lane.
- The lane holds a few concurrent runs; if it answers "quick lane is full", either wait or
file a ticket — don't spam retries.
Treat an agent's report like a worker's output: trustworthy about what it did, but you own
the relay — paraphrase, keep names/links exact, and say plainly when a run failed or timed
out.