Adopt the coordinating-epics skill for #24 #52

Activity

1. gregoryfoster commented on Oct 15, 2026

   

   gregoryfoster

   on Oct 15, 2026

   ContributorAuthor

   More actions

   Hand-off from the burn-down epic #24, 2026-10-15. Step 1 (vendoring) is next, in its own thread. Step 2, the adoption, stays with #24's coordinator: it is that thread's next write, at the check-in after step 1 merges.

   Pre-flight (coordinator, 2026-10-15, measured)

   • The vendored pointer on main (244ca3f) is 5ff7a82, 100 commits behind fef0611. The daily hook last bumped it on 2026-10-13. skills/coordinating-epics doesn't exist yet.

   • The dry run of v1 against Burn-down: open issues after v1 go-live (done 2026-10-11; next: #37; parked: #54) #24's live body (the skill's scripts at fef0611, run from a scratch copy; nothing written):

     • --lint-all: 55 problems on 33 lines (48 short refs, plus 7 bare numbers from other repos). That confirms this issue's count.
     • This issue's dry run missed one thing. A header op keeps the line's existing lead. Burn-down: open issues after v1 go-live (done 2026-10-11; next: #37; parked: #54) #24's cursors were in a paragraph, so the call applied without errors but Last check-in stayed in the paragraph, and Last probe, Last issue seen and Last verified SHA were appended after the Milestones list. Reading the diff showed it; exit 0 didn't.

   • Operator decision (2026-10-15): pre-flight now. At 15:05Z, Burn-down: open issues after v1 go-live (done 2026-10-11; next: #37; parked: #54) #24's last write by the old protocol:

     • put the header into the template's list (Last check-in, Last probe, Last issue seen Adopt the coordinating-epics skill for #24 #52, Last verified SHA 244ca3f, Next);
     • wrote every cross-repo ref as owner/repo#N.

     The skill's own lint() now finds 0 problems in the body.

   • The adoption call, dry-run against the new body: scan.sh, then epic-edit.sh --adopt-live --lint-all --scan … --ops (header ops with {checked} and {last_seen}, plus a log line). It exits 0, and the header op rewrites the cursor in place. So step 2 no longer needs this issue's explicit --watermark 51 --from-sha …: the header supplies both.

   • A v1 limit, new here: scan.sh's cross_repo report reads with one token. With this repo's GH_TOKEN, it read 23 of 28 refs; the 5 CannObserv/cannobserv refs came back 404. The coordinator keeps checking cannobserv by hand with GH_TOKEN_CANNOBSERV until that changes. Reported on New skill: burning-down-an-epic - a lightweight, interactive, operator-paced counterpart to orchestrating-issue-backlog gregoryfoster/skills#348.

   Step 1 brief

   Touchpoints (checked against main at 244ca3f):

   • the submodule skills-vendor/gregoryfoster-skills;
   • two new symlinks, skills/coordinating-epics → ../skills-vendor/gregoryfoster-skills/skills/coordinating-epics and .claude/skills/coordinating-epics → ../../skills/coordinating-epics (docs/SKILLS.md's layout table; managing-skills Step 2/2b);
   • docs/SKILLS.md's inventory line.

   Traps:

   1. tests/test_skills.py::test_skills_md_inventory_matches_the_links holds docs/SKILLS.md's **gregoryfoster-skills (N …):** line to the links. Add coordinating-epics to it and bump the count. The other tests (both discovery paths list the same skills; .claude/skills routes through skills/) fail if either link is missing.
   2. The bump spans 100 commits, so every skill already linked from that repo moves too. Run bash .skills/doctor.sh and the skills tests, and check that no linked skill was renamed or removed upstream (a dangling link). List the skills this repo uses whose content changed: reviewing-code-python-fastapi, shipping-work-python-fastapi, using-git-worktrees, writing-plans, init-socraticode, using-mayfly-chat, managing-skills. Don't re-derive what changed in them; just name them.
   3. The refresh hook pushes submodule bumps to main daily. If it lands during the PR, the pointer conflicts. Rebase, and keep whichever commit is newer, as long as it is at or after fef0611. After the merge, run git pull --ff-only in the main checkout, or the hook's next push is rejected.
   4. No runtime path changes, so there's no deploy, and the drift check stays ok. .socraticodeignore already excludes skills/ and skills-vendor/.

   Done when:

   • the pointer is at fef0611 or later, and both links resolve;
   • .claude/skills/coordinating-epics/scripts/epic-edit.sh --help runs (SKILL.md's script-path resolution finds it);
   • the skills tests and doctor.sh pass, and CI is green on the PR and on main;
   • the PR says Refs #52, not Closes, because Adopt the coordinating-epics skill for #24 #52 closes only after step 2's read-back.

   Out of scope: any edit to #24's body (step 2, the coordinator's). Fixing upstream skills (change those upstream).

   Ship: branch in a worktree → PR → FF merge → git pull --ff-only. The closing note goes on #52 as a comment: each box with its evidence, plus the list from trap 2.

   Kickoff prompt:

   Pick up processor#52, step 1 only (vendor coordinating-epics). Read its hand-off comment from epic #24 first. Don't edit #24.
2. gregoryfoster commented on Oct 15, 2026

   

   gregoryfoster

   on Oct 15, 2026

   ContributorAuthor

   More actions

   Step 1 shipped in #53, FF-merged as 4b3a500. #52 stays open for step 2, the coordinator's adoption call on #24.

   • Pointer at fef0611 or later. skills-vendor/gregoryfoster-skills is at fef0611 (upstream HEAD). The daily refresh hook moved it on main in 98c5032 at 2026-10-15T15:40Z, before this thread branched, so #52 chore: vendor coordinating-epics #53 carries no bump and trap 3's conflict never came up.
   • Both links resolve. skills/coordinating-epics → ../skills-vendor/gregoryfoster-skills/skills/coordinating-epics, and .claude/skills/coordinating-epics → ../../skills/coordinating-epics. readlink -e resolves both in the main checkout, and Claude Code lists the skill.
   • .claude/skills/coordinating-epics/scripts/epic-edit.sh --help exits 0. SKILL.md's skill-scripts block resolves both epic-edit.sh and scan.sh under .claude/skills/coordinating-epics/scripts/. This repo's own scripts/ is checked first and has neither.
   • Tests and doctor.
     • tests/test_skills.py failed on the inventory line alone, then passed with both links (8 passed).
     • bash .skills/doctor.sh exits 0. Its advisory that the brainstorming override lags obra-superpowers predates this change: that pointer didn't move.
     • Full suite (-m 'not integration'): 594 passed, 14 skipped. Ruff is clean.
   • CI is green. On the PR: run 38762837707. On main at 4b3a500: run 38763356964. Both lint and test passed.
   • docs/SKILLS.md: gregoryfoster-skills goes from 12 to 13, with coordinating-epics (Burn-down: open issues after v1 go-live (done 2026-10-11; next: #37; parked: #54) #24 since Adopt the coordinating-epics skill for #24 #52).

   Trap 2 (5ff7a82..fef0611, 100 commits). None of the linked skills changed: reviewing-code-python-fastapi, shipping-work-python-fastapi, using-git-worktrees, writing-plans, init-socraticode, using-mayfly-chat, managing-skills, and the other five. The span touched only skills/coordinating-epics/ (new), skills/patching-hosts/, and upstream's .skills/ and docs/plans/. Nothing linked here was renamed or removed.

   Aside, out of scope: patching-hosts (upstream, predates this span) is neither linked here nor on docs/SKILLS.md's "Not linked" list.

   No runtime path changed, so there was no deploy, and the drift check is unaffected.
3. gregoryfoster commented on Oct 15, 2026

   

   gregoryfoster

   on Oct 15, 2026

   ContributorAuthor

   More actions

   Done, both steps (2026-10-15).

   • Step 1, vendoring: #52 chore: vendor coordinating-epics #53 (4b3a500, FF). CI green by job (lint, test) on the PR run and the main push run. The coordinator checked independently:

     • both links resolve;
     • the scripts resolve through SKILL.md's block from .claude/skills/coordinating-epics/scripts;
     • of the linked skills, only coordinating-epics changed across 5ff7a82..fef0611.

     The pointer arrived through the refresh hook's daily bump (98c5032), not through the PR. No deploy (no runtime path changed).

   • Step 2, the adoption: Burn-down: open issues after v1 go-live (done 2026-10-11; next: #37; parked: #54) #24 was written by epic-edit.sh --adopt-live --lint-all --scan in one call. Exit 0, "readback": "matches", and --lint-all passed on the whole body. The cursors came from scan.sh ({checked}, {last_seen} = #52 chore: vendor coordinating-epics #53). From here every body edit goes through epic-edit.sh, and every check-in starts with scan.sh.

   Known limits (v1, reported on gregoryfoster/skills#348):

   • scan.sh's cross_repo reads with one token, so CannObserv/cannobserv refs come back 404. The coordinator checks those by hand.
   • No op renames a heading or edits the intro paragraph. A header op takes only cursor fields. So Measured state's heading keeps its 15:01Z time, and the protocol note lives in the Parking-lot section.

   Follow-ups: none here.
4. gregoryfoster

   closed this as completedon Oct 15, 2026