fix(initiative-driver): don't release a sub-issue that is itself an epic (nested-epic-driven-as-story) #976

Activity

1. don-petry

   added

   dev-leadFor dev-lead agent pickupFor dev-lead agent pickup

   on Jun 28, 2026
2. github-actions

   added

   bugBug reportsBug reports

   on Jun 28, 2026
3. github-actions commented on Jun 28, 2026

   

   github-actionsBot

   on Jun 28, 2026 – with GitHub Actions

   Contributor

   More actions

   Thanks for the detailed write-up — the incident reference to #934/#938 makes the failure mode crystal clear and will be a big help when writing the bats coverage. One clarifying question before the fix lands: the issue describes the skip condition as 'carries the initiative label and/or has ≥1 native sub-issue of its own' — should the implementation treat these as equivalent alternatives (either check is enough to skip), or should one be preferred as the canonical signal (e.g. always use the initiative label, with the sub-issue count only as a fallback)? Knowing which signal is authoritative will keep the guard condition tight and avoid false positives.
4. don-petry commented on Jun 28, 2026

   

   don-petry

   on Jun 28, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead Implementation Plan

   Issue: #976 — fix(initiative-driver): don't release a sub-issue that is itself an epic (nested-epic-driven-as-story)

   Scope

   In drive_epic's ready-sub-issue selection, skip any sub-issue that is itself an epic so the driver never labels a nested epic dev-lead (the #934/#938 incident); the nested epic continues to be driven independently via the sweep.

   Detecting "is itself an epic"

   The planner labels both epics and stories with the plain initiative label (apply-plan.sh:134 for the epic, :170 for each story), so initiative alone cannot distinguish an epic from a story — keying on it would skip every story and release nothing. The reliable epic signals are:

   • carries GATE_LABEL (initiative:auto) — an armed nested epic, which is exactly what makes it appear in the sweep and get driven on its own, OR
   • has ≥1 native sub-issue of its own (structural marker; stories have none).

   Implementation Checklist

   • Add an is_epic helper (gate-label check + native sub-issue count) to scripts/initiative-driver.sh
   • Call it in the release loop in drive_epic, before labeling; log #N — skip: is itself an epic (drive it independently).

   Tests to Write (tests/test_initiative_driver.bats)

   • Nested epic detected by initiative:auto gate label is skipped (no gh issue edit)
   • Nested epic detected by having its own native sub-issue is skipped
   • A normal story (plain initiative, no sub-issues, no gate) is still released — guards against over-skipping

   Test Command

   bats tests/test_initiative_driver.bats (full suite via lint.yml bats job); lint via bash .dev-lead/scripts/dev-lead-lint.sh
5. don-petry commented on Jun 28, 2026

   

   don-petry

   on Jun 28, 2026

   CollaboratorAuthor

   More actions

   Dev-Lead: Implementation Complete

   Plan Execution

   • Added sub_issue_numbers helper and a nested-epic skip in drive_epic's release loop
   • Skip logs #N — skip: is itself an epic (drive it independently).
   • Added bats coverage for the nested-epic skip (gate-label signal, native-sub-issue signal, and a non-regression "plain story still released" case)

   How "is itself an epic" is detected

   The skip fires when a candidate sub-issue carries initiative:auto (the gate label) OR has ≥1 native sub-issue of its own.

   The plain initiative label is deliberately not used: initiative-planner/apply-plan.sh labels both epics (:134) and their story sub-issues (:170) with initiative, so it cannot distinguish the two — keying on it would skip every story and release nothing. The gate label is the precise signal for the incident class (epic #934 was an initiative:auto epic nested under armed #872) and is exactly what makes the nested epic appear in the sweep to be driven on its own (AC #2). The native-sub-issue count is a structural backstop. The children fetch is a direct assignment (not inside the if condition) so a transient API failure trips errexit instead of being masked into a false "not an epic" — matching the fail-closed behavior of the adjacent blocked_by check.

   Test Results

   $ bats tests/test_initiative_driver.bats
   1..15
   ok 1 has_label matches an exact label name
   ok 2 has_label with -F does not match regex dot against a different label
   ok 3 has_label returns non-zero when label is absent
   ok 4 gate: exits 0 with no-op when epic lacks initiative:auto
   ok 5 gate: proceeds past gate when epic carries initiative:auto
   ok 6 cap: does not label new issues when in-flight equals MAX_IN_FLIGHT
   ok 7 blocked_by: does not label an issue that has an open blocker
   ok 8 blocked_by: labels a sub-issue when all blockers are closed
   ok 9 nested-epic: skips a sub-issue carrying the gate label (drive it independently)
   ok 10 nested-epic: skips a sub-issue that has its own native sub-issues
   ok 11 nested-epic: a plain story (initiative label, no children, no gate) is still released
   ok 12 sweep: drives every open epic carrying initiative:auto when EPIC is empty
   ok 13 sweep: a gh failure on one epic surfaces as an error (not masked) and the sweep continues
   ok 14 sweep: no-op when no open epic carries initiative:auto
   ok 15 cross-repo: a non-self target_repo sweep targets that repo on every API and label call
   
   $ bash .dev-lead/scripts/dev-lead-lint.sh
   [lint] shellcheck on 81 file(s)...
     [lint] validating 4 agent profile(s)...
     [lint] validating .github/CODEOWNERS...
     [lint] all checks passed
   

   Full repo bats suite: 911 passing. One unrelated pre-existing failure (tests/test_batch_fallback.bats — "Gemini trust error (55) falls back to Copilot") reproduces with my changes stashed, so it is independent of this change.

   Files Changed

   • scripts/initiative-driver.sh: added sub_issue_numbers helper; in drive_epic's release loop, skip any candidate that is itself an epic (gate label or own native sub-issues) before labeling dev-lead.
   • tests/test_initiative_driver.bats: added 3 nested-epic tests; updated the blocked_by and cross-repo mocks to model that a plain story has no native sub-issues of its own (their assertions are unchanged — the per-child sub_issues call is new to this code path).

   Notes

   • AC test issue from agent #1 (nested epic not labeled dev-lead) and AC Add @claude delegation, auto-merge, and rebase handling #3 (bats coverage) are covered directly. AC Go-live improvements for PR review agent #2 (nested epic still driven independently) is unchanged — top-level gated-epic sweep behavior is untouched; the skip only prevents double-driving the nested epic as a story of its parent.
   • Deviation from the literal "carries the initiative label" wording is intentional and documented above (plain initiative is on every story, so it cannot identify epics).
6. don-petry

   added a commit that references this issue on Jun 28, 2026

   feat: implement issue #976 — fix(initiative-driver): don't release a …

   ...

   549daa1
7. don-petry

   closed this as completedin #979on Jun 28, 2026