Skip to content

agents: obs-worker-loop skill for the self-directed worker on a timer #2075

Closed

Description

@zackees

The maintainer runs workers on a timer too (/loop 10m take on the role of a worker that is looking for issues, figuring out the most important next thing, and doing it). obs-worker covers one issue from claim to merge; obs-coordinator covers a timer that moves other people's work and never claims. Nothing covers the timed worker: what to check each tick, how to rank the next item, and the traps hit along the way. Requested by the maintainer on 2026-10-24.

Lessons from a day of running that loop (2026-10-23/24) to put in a new obs-worker-loop skill:

  1. Own PRs first, every tick. Queue green ones and confirm with the merge-queue GraphQL query (gh pr merge can print nothing). After a stacked base merges: git rebase --onto origin/master HEAD~N, retarget, then close and reopen, since CI does not start after a retarget.
  2. Master health next. Scheduled runs (Sanitizers, the nightly full run) are not run on PRs, so they can stay red for days unseen. Before filing, search open and recently closed issues and merged PRs for the error text (ci: nightly ASan build fails: null format string (-Werror=format-truncation) in libobs.rust-variadic+.h.c #2014 duplicated bug: Sanitizers asan build fails: null format string in os_process_args_add_argf (pipe-variadic.c.h) #2005, already fixed by libobs: fix the Sanitizers asan build (os_process_args_add_argf via dstr_vprintf, #2005) #2007).
  3. Ranking unclaimed work: unblocks others > bug with clear RED > port order. Skip proposals the maintainer has not approved (ask with Decision needed:), trackers whose children are all closed (close them), and issues filed in the last minutes by an active session (they are usually claimed within a tick).
  4. Stale claims: past the window, ask the claimant before taking over; check for PRs in other repos, which the claim bot does not see (obs-websocket: RemoveInput keeps the scene items of the input in scenes that are not rendered #1613: the fix was requesthandler: Remove the scene items of a removed input obs-websocket#20).
  5. Staging: never git add -A / -u: local builds leave AppDir/ and squashfs-root/ in the tree (a 103 MB file got pushed and rejected) and submodule pointers move (obs-websocket). Name the files.
  6. Environment: use extern clones (one per concurrent branch); git submodule update --init before native builds; bosn run --task appimage; when a shared cache (zccache cool-down) breaks local builds, stop and let PR CI, or one workflow dispatch on the branch, validate.
  7. Delegation: hand multi-file work to one background agent with a tight brief (clone, branch, RED/GREEN, never -A); don't touch its files meanwhile.
  8. Idle ticks: a short report, no make-work.

Plan: one docs PR adding .claude/skills/obs-worker-loop/SKILL.md, linking it from AGENTS.md (worker role), plus a ci check that every skill in .claude/skills/ is linked from AGENTS.md and that worker skills link SECURITY.md.

Activity

  1. zackees commented on Oct 24, 2026

    @zackees
    MemberAuthor

    Claiming: #2075. One PR today: the skill, the AGENTS.md link, and the ci check (RED first).

  2. added
    claimedAn agent or person has claimed this; check the claim comment before starting
    on Oct 24, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    claimedAn agent or person has claimed this; check the claim comment before startinginfraBuild, gate, CI and tooling

    Type

    No type

    Projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions