Document tea CI-status parsing pitfalls and idle-friendly loop polling #22

Merged

geusebio merged 1 commit from docs/ci-watch-idle-polling-and-status-parsing into main

2026-08-13 20:16:07 +00:00

geusebio commented

2026-08-13 20:00:15 +00:00

Owner

Summary

Two things learned the hard way while babysitting a real PR's CI in a Claude Code session, written up so the next session doesn't re-derive them:

tea pulls <index> --fields ci is not fully authoritative. It has previously dropped a job's row entirely (a PR read all-green while a job had actually failed), and its OSC-8 hyperlink escapes wrap each job's link text — which itself contains a colon (https://…) — so a naive "match after the colon" parser silently grabs the URL instead of the job's status. Documents the escape-stripping needed and recommends cross-checking the commit statuses API (/repos/{owner}/{repo}/commits/<sha>/statuses, grouped by context, newest wins) before calling a PR green.
Watching CI shouldn't hold a session "busy" for the whole wait. A backgrounded shell loop or a persistent Monitor both keep a real process running the entire time, which looks identical to active work from the outside. Documents the alternative: /loop in dynamic mode with no Monitor armed — a single foreground check per iteration, then ScheduleWakeup ends the turn with nothing running until the next firing — plus adapting the delay each iteration to what was just observed instead of a fixed cadence.

Test plan

Docs-only change (docs/Topics/VCS.md) — no code/lint/CI surface in this repo.
Read through for consistency with the surrounding "Pull requests" section's style and terminology.

## Summary Two things learned the hard way while babysitting a real PR's CI in a Claude Code session, written up so the next session doesn't re-derive them: - **`tea pulls <index> --fields ci` is not fully authoritative.** It has previously dropped a job's row entirely (a PR read all-green while a job had actually failed), and its OSC-8 hyperlink escapes wrap each job's link text — which itself contains a colon (`https://…`) — so a naive "match after the colon" parser silently grabs the URL instead of the job's status. Documents the escape-stripping needed and recommends cross-checking the commit statuses API (`/repos/{owner}/{repo}/commits/<sha>/statuses`, grouped by context, newest wins) before calling a PR green. - **Watching CI shouldn't hold a session "busy" for the whole wait.** A backgrounded shell loop or a persistent `Monitor` both keep a real process running the entire time, which looks identical to active work from the outside. Documents the alternative: `/loop` in dynamic mode with no `Monitor` armed — a single foreground check per iteration, then `ScheduleWakeup` ends the turn with nothing running until the next firing — plus adapting the delay each iteration to what was just observed instead of a fixed cadence. ## Test plan - Docs-only change (`docs/Topics/VCS.md`) — no code/lint/CI surface in this repo. - Read through for consistency with the surrounding "Pull requests" section's style and terminology.

geusebio added 1 commit

2026-08-13 20:00:15 +00:00

Document tea CI-status parsing pitfalls and idle-friendly loop polling c17b3b69eb

`tea pulls <index> --fields ci` silently drops job rows and its OSC-8 hyperlink
escapes break naive status-colon parsing; the commit statuses API is the
authoritative source. Separately, a backgrounded shell loop or a persistent
Monitor both hold a session "busy" for the whole CI wait — a plain /loop with
ScheduleWakeup and no Monitor, delay adapted each iteration to what was just
observed, is the way to poll without that.

geusebio merged commit f917d8d068 into main

2026-08-13 20:16:07 +00:00

geusebio referenced this pull request from a commit

2026-08-13 20:16:09 +00:00

Merge pull request 'Document tea CI-status parsing pitfalls and idle-friendly loop polling' (#22) from docs/ci-watch-idle-polling-and-status-parsing into main

geusebio referenced this pull request from a commit

2026-08-15 16:27:45 +00:00

Warn that submodule pins go backwards silently on merge

geusebio referenced this pull request

2026-08-15 16:28:10 +00:00

Warn that submodule pins go backwards silently on merge #25