You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
10. **Never use the deprecated `lab job logs`** — see the "Job logs: three real commands" section below.
227
229
11. **After queuing a task, ASK the user if they'd like you to watch the logs.** Don't start streaming or polling automatically — jobs can take minutes to hours, and `--follow` blocks. Report the Job ID and ask: "Want me to watch the logs and report back?"
228
230
12. **Never create API keys programmatically** — if auth fails, ask the user to provide an API key from the web UI
231
+
13. **Always pass `--description/-m` when queuing a task. Generate it yourself — never ask the user.** See "Always write a run description" below.
232
+
233
+
### Always write a run description
234
+
235
+
Every `lab task queue` call MUST include `--description/-m "..."`. The description is markdown stored on the job and shown in `lab job info`. Its audience is a future researcher reading `lab job list` weeks later — they have git and the task code, but NOT this chat. The description is the only bridge.
236
+
237
+
**Treat it like a short PR description for this run.** Draft 1–5 lines (bullets for multi-point notes) covering:
238
+
239
+
1. **What changed vs the prior run / baseline** — the concrete diff (hyperparameters, code, model, data, infra). If nothing changed, say so and link the prior job.
240
+
2. **What hypothesis you're testing** — why this run is worth doing.
241
+
3. **What a future reader should remember** — gotchas, prior surprises, things to check in the output.
242
+
243
+
Pull these from the conversation and recent git diff / edited files. If the note has newlines or shell-awkward characters, pipe it: `printf '%s' "$DESC" | lab task queue abc123 -m -`.
- Testing whether higher lr clears the eval/loss=2.1 plateau seen in job 7f21 around step 2k.
249
+
- Watch: earlier runs with lr≥5e-5 diverged by step 500." | lab task queue abc123 --no-interactive -m -
250
+
251
+
# Good: small change — one line is enough
252
+
lab task queue abc123 --no-interactive -m "Rerun on H100 (was A100) to confirm throughput regression from #1850."
253
+
254
+
# Good: nothing changed
255
+
lab task queue abc123 --no-interactive -m "Rerun of job 7f21, no code or config changes (network flake on first attempt)."
256
+
257
+
# Bad: generic filler that tells the reader nothing
258
+
lab task queue abc123 -m "train model"
259
+
```
260
+
261
+
Don't restate the task name, full hyperparameter dict, or file paths — those are already on the job record. Don't copy the user's last message verbatim — synthesize. If the conversation is truly empty of signal, fall back to `"Rerun of <id>, no changes"`.
229
262
230
263
### Selecting a provider when queuing a task
231
264
@@ -308,7 +341,7 @@ This applies to launching jobs, fetching logs, checking cluster status, and ever
308
341
| `lab task init` | Scaffold `task.yaml` + `main.py` in the current directory (`--interactive` to prompt) | No |
309
342
| `lab task add [dir]` | Add task from directory or `--from-git` URL (`--no-interactive`, `--dry-run`) | Yes |
310
343
| `lab task delete <id>` | Delete a task (`--no-interactive` to skip confirmation) | Yes |
| `lab task queue <id>` | Queue task on compute provider (`-m/--description` for a markdown run note; required for agents, see "Always write a run description") | Yes |
Copy file name to clipboardExpand all lines: .agents/skills/transformerlab-cli/references/commands.md
+1Lines changed: 1 addition & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -131,6 +131,7 @@ Queue a task on a compute provider.
131
131
| Option | Description |
132
132
|---|---|
133
133
|`--no-interactive`| Skip prompts. Uses the task's configured provider or first available. Parameters use defaults. **Always use this in automated workflows.**|
134
+
|`-m`, `--description <text>`| Markdown note describing what this run is trying to accomplish (stored on the job, shown in `lab job info`). Pass `-` to read from stdin. **Agents: required per SKILL.md rule 13.**|
134
135
135
136
**JSON output:** Returns the created job object with `id` and `status`.
0 commit comments