feat(framework/tools): orchestration helpers — lane-brief.sh + ci-wait.sh #547

jason.woltje · 2026-06-18T18:30:35Z

jason.woltje commented

2026-06-18 18:30:35 +00:00

Closes #546.

Two additive orchestration tools for the framework suite, both already adopted by the live U-Connect orchestrator (mos-claude) before this canonical contribution. Distilled from forensic analysis of a live delivery session under a wall-clock-first optimization mandate.

lane-brief.sh (`git/`)

Live dispatch brief for a repo lane — one call returns the CURRENT open issue set from Gitea, classified for dispatch. Kills the observed failure mode: workers self-report issue state from stale static briefs and name already-CLOSED issues as 'todo', forcing per-issue re-verification before dispatch.

Closed excluded by definition (the whole point).
Partitions open issues by PR-linkage (reliable signal) — NOT assignee/dependency, which this fleet leaves empty in the provider, so it deliberately does not claim 'unassigned == available'.
Login resolution: -L > $GITEA_LOGIN > owner inference > detect-platform.sh fallback (the shared default-login resolver is not owner-aware; relevant to the #538 login-resolution line of work).

ci-wait.sh (`woodpecker/`)

Blocks until one or more Woodpecker pipelines reach terminal state, wrapping the existing pipeline-status.sh (resolves repo->id, instance-aware). Replaces hand-authored while true; curl .../repos/1/pipelines/\$n; sleep loops that HARDCODE Woodpecker repo id 1 and re-implement URL building with raw curl. Intended as the COMMAND of a Monitor / event-driven re-invoke (primary) + one long (>=1500s) timed fallback — never a tight <300s poll. Exit: 0=all success / 1=terminal non-success / 2=usage / 3=timeout.

Notes

Both resolve their dependencies as BASH_SOURCE siblings (works installed and in-repo).
New scripts are mode 100755 to match the sibling tools; no cosmetic mode-flips on existing files.
No version bump — release/publish is a separate PR per the release(mosaic): bump ... convention.
Tested live against usc/uconnect (lane-brief vs the open board; ci-wait vs pipeline 3943, exit 0).

Suggested follow-up (not in this PR): git config core.fileMode false is advisable in working checkouts to avoid 0755 mode-flip noise.

Fixes #546

Closes #546. Two additive orchestration tools for the framework suite, both already adopted by the live U-Connect orchestrator (`mos-claude`) before this canonical contribution. Distilled from forensic analysis of a live delivery session under a wall-clock-first optimization mandate. ### lane-brief.sh (`git/`) Live dispatch brief for a repo lane — one call returns the CURRENT open issue set from Gitea, classified for dispatch. Kills the observed failure mode: workers self-report issue state from stale static briefs and name already-CLOSED issues as 'todo', forcing per-issue re-verification before dispatch. - Closed excluded by definition (the whole point). - Partitions open issues by PR-linkage (reliable signal) — NOT assignee/dependency, which this fleet leaves empty in the provider, so it deliberately does not claim 'unassigned == available'. - Login resolution: `-L` > `$GITEA_LOGIN` > owner inference > `detect-platform.sh` fallback (the shared default-login resolver is not owner-aware; relevant to the #538 login-resolution line of work). ### ci-wait.sh (`woodpecker/`) Blocks until one or more Woodpecker pipelines reach terminal state, wrapping the existing `pipeline-status.sh` (resolves repo->id, instance-aware). Replaces hand-authored `while true; curl .../repos/1/pipelines/\$n; sleep` loops that HARDCODE Woodpecker repo id 1 and re-implement URL building with raw curl. Intended as the COMMAND of a Monitor / event-driven re-invoke (primary) + one long (>=1500s) timed fallback — never a tight <300s poll. Exit: 0=all success / 1=terminal non-success / 2=usage / 3=timeout. ### Notes - Both resolve their dependencies as BASH_SOURCE siblings (works installed and in-repo). - New scripts are mode 100755 to match the sibling tools; no cosmetic mode-flips on existing files. - No version bump — release/publish is a separate PR per the `release(mosaic): bump ...` convention. - Tested live against `usc/uconnect` (lane-brief vs the open board; ci-wait vs pipeline 3943, exit 0). Suggested follow-up (not in this PR): `git config core.fileMode false` is advisable in working checkouts to avoid 0755 mode-flip noise. Fixes #546

jason.woltje added 1 commit 2026-06-18 18:30:36 +00:00

feat(framework/tools): orchestration helpers — lane-brief.sh + ci-wait.sh (#546 )

ci/woodpecker/push/ci Pipeline was canceled

Details

ci/woodpecker/pr/ci Pipeline was canceled

Details

d7028f959b

Two additive orchestration tools distilled from forensic analysis of a live
U-Connect delivery session, both adopted by the live orchestrator before this
contribution.

lane-brief.sh (git/): one call returns the CURRENT open issue set for a repo
lane (milestone/label) from Gitea, classified for dispatch. Defeats stale
worker self-report (workers brief from static notes and report already-CLOSED
issues as "todo"). Closed excluded by definition; partitions by PR-linkage
(reliable) not assignee/dependency (empty in this fleet). Login resolution:
-L > $GITEA_LOGIN > owner inference > detect-platform.sh fallback.

ci-wait.sh (woodpecker/): blocks until pipeline(s) reach terminal state,
wrapping pipeline-status.sh (resolves repo->id, instance-aware). Replaces
hand-rolled `curl .../repos/1/pipelines/$n` loops that hardcode repo id 1.
Intended as a Monitor command + long (>=1500s) timed fallback, not a tight
poll. Exit 0=all success / 1=terminal non-success / 2=usage / 3=timeout.

Tested live vs usc/uconnect. README updated. No version bump (separate
release PR per convention).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kt2D8TsnDwhtzEAPijsNmR

jason.woltje added 1 commit 2026-06-18 18:45:10 +00:00

fix(framework/tools): self-review fixes from defect survey (#546 )

ci/woodpecker/push/ci Pipeline failed

Details

ci/woodpecker/pr/ci Pipeline failed

Details

46d828f166

Two defects an independent survey of the tooling surface found in the new
helpers, fixed pre-gate:

- lane-brief.sh: label filter used jq contains() (substring) — `-l security`
  wrongly matched label `domain/6-security`. Now exact-token match against
  tea's space-separated labels string. Verified live: `-l security` -> 0,
  `-l domain/6-security` -> the real holders.
- ci-wait.sh: unknown owner silently defaulted to the `usc` Woodpecker
  instance (wrong credentials, wrong pipelines). Now fails hard requiring
  `-a <instance>`, matching lane-brief's FATAL-on-unresolved behavior.

Verified: usc owner still infers and exits 0; unknown owner exits 2 with
guidance.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Kt2D8TsnDwhtzEAPijsNmR

jason.woltje added 1 commit 2026-06-18 21:49:25 +00:00

fix(framework/tools): lane-brief.sh — classify PR-body-linked issues as work-underway (#546 )

ci/woodpecker/push/ci Pipeline failed

Details

ci/woodpecker/pr/ci Pipeline failed

Details

1d87b76e1f

Remediation of coder3 independent-validation blocker on PR #547.

lane-brief.sh inspected only the open-PR index/title/head fields, never the PR
BODY or Gitea issue linkage. A body-only "Closes #546" was therefore invisible,
so issue #546 (open, with PR #547 'Closes #546' in its body) was placed under
DISPATCH CANDIDATES with work-underway count 0 — re-dispatchable in-flight work,
unacceptable for a dispatch-truth tool.

Fix:
- Fetch open PRs as JSON including `body`; resolve PR->issue links via Gitea's
  closing-keyword set (close/closes/closed, fix/fixes/fixed, resolve/resolves/
  resolved), case-insensitive, word-boundary anchored, `#` directly following the
  keyword. Any issue so linked from an OPEN PR is classified WORK UNDERWAY.
- Preserve the prior title/head bare-ref heuristic and per-repo behavior; require
  `#` immediately after the keyword so cross-repo `owner/repo#N` forms don't leak.
- Bare `#N` prose mentions in a body are intentionally NOT links (e.g. "#538 line
  of work") to avoid marking live, dispatchable issues as in-flight.

Tests (committed, RED-on-revert non-vacuity):
- test-lane-brief-pr-linkage.sh: open-PR-with-'Closes #546'-in-body excludes #546
  from candidates (and a reverted copy with the body-scan removed regresses #546
  to a candidate — RED proof); bare #777 and substring 'hotfix #999' stay
  candidates (word-boundary + closing-keyword-only guards).
- test-ci-wait-exit-matrix.sh: ci-wait.sh exit matrix 0 (all-success) / 1
  (terminal-not-success: failure + error/killed) / 2 (usage) / 3 (timeout).

shellcheck -x + bash -n clean on all four files; no secret values. ci-wait.sh
unchanged (coder3 PASS preserved). Closed-issue exclusion unchanged.

Refs #546, PR #547

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>