| name | los-quiet-workon |
| description | Use when working or resuming LOS Jira implementation in a fast worktree with minimal chat output, especially after noisy /workon loops, repeated stack setup, test polling, or context/token burn. |
LOS Quiet Workon
Use this workflow when the user wants LOS Jira/worktree implementation work with minimal chat output, especially after context/token burn, noisy /workon runs, repeated fast-stack setup, or test/debug loops.
Output Contract
- Send at most one short opening line saying what will be done.
- Do not send periodic progress updates during normal command execution, polling, file reads, or test waits.
- Speak only for: a real blocker, a risky cleanup decision, a user-visible result, or final handoff.
- Never paste full command output, Docker tables, diffs, Jira descriptions, or test logs into chat unless the user asks. Summarize only the actionable result.
- Prefer low-output tool calls: small
max_output_tokens, narrow rg, targeted sed, and focused test selectors.
- Put durable details in the canonical Agentic OS work item under
/Users/genome/agentic_os/los/02-projects/los_app_los_django/work-items/ instead of chat when the run needs notes, logs, PR bodies, watcher state, or handoff context. Use the packet's artifacts/ directory. Repo .features/<ticket-or-task>/ folders are legacy mirrors only; do not create .features/ artifacts inside an isolated worktree.
- For reopened post-merge work, keep a root
REOPEN.md in the worktree with reason, dates, original PR, follow-up PR/ticket, owner, status, and Docker/OrbStack cleanup note. This file blocks unattended cleanup.
- Commands expected to run longer than two minutes must use
quiet-async-runner,
/Users/genome/agentic_os/harness/bin/agentic-os-quiet-run, a project watcher,
or a scoped verification subagent. Chat is not the polling loop.
Workflow
- Identify the Jira key and create or reuse the canonical Agentic OS work item first. Prefer an existing matching lane object; otherwise run
agentic-os project work-item create los los_app_los_django --root /Users/genome/agentic_os --status building --format packet --title "<JIRA-KEY or task>: <short title>" --summary "<one-line task summary>". Set AGENTIC_OS_ACTIVE_WORK_ITEM=<absolute work-item path> for long-running commands and subagents.
- For implementation, use an isolated LOS worktree. Prefer
~/projects/los/misc/setup-worktree.sh -b feature/<JIRA-KEY>-<slug> --fast; do not use raw git worktree add unless the setup script is unavailable.
- Run
git status --short --branch before edits. Preserve unrelated dirty work and do not stage or revert files outside the task.
- Use the existing skill/router only as needed:
los-fast-workon for stack/worktree setup, workon for Jira implementation flow, and nearest AGENTS.md before editing a subtree.
- Keep investigation narrow. Use Jira comments and prior PRs to form the bug hypothesis, then inspect the exact service, callers, and focused tests.
- Before file edits, make one concise note if needed. Then edit with
apply_patch.
- Verify with the smallest meaningful test target first. Expand only if the change touches shared behavior. Run slow verification through
artifacts/async-runs/<run-id>/ and report only terminal pass, fail, killed, blocked, or timeout with a receipt.
Fast Stack And Resource Handling
- Start with
make fast-status; use make fast-up or the worktree setup script outputs for URLs.
- If setup or tests fail from occupied ports, identify the owning containers first. Stop only clearly stale/unrelated LOS worktree containers, and name that cleanup in the final summary.
- If tests die with exit
137, treat it as likely host/container pressure. Check docker stats --no-stream, stop stale unrelated stacks if safe, then rerun a narrower target.
- Do not keep chat alive with test polling. For any slow setup or test wait, inspect only
state.json; if it still says running, stay silent unless the user explicitly asked for status.
- Use context-mode to summarize large
output.log files from async run artifacts instead of reading full logs into chat.
- When local tests are slow or unavailable and a PR exists, rely on GitHub PR checks as the final readiness signal and capture status notes or check artifacts under the work item's
artifacts/pr-watch/ folder.
Final Handoff
Final response should be short and include:
- ticket/key and branch/worktree when relevant
- Agentic OS work item path
- files changed
- verification run and result
- any cleanup performed
- remaining blocker or next command, only if one remains
- if the work was reopened, whether
REOPEN.md was updated with second-completion and safe-delete notes