okrdev/checkins/2026-Q3/2026-W32.md

---
cycle: 2026-Q3
week: 2026-W32
attendees: [alex]
---

<!-- Revised 2026-08-07 — outside-adopter references anonymized ahead of open-sourcing;
     substance, dates, and numbers unchanged. -->

## Wins

- alex: Got the outside adopter's repo to okrdev Level 2 on v0.5.0 — an outside project
  didn't just keep the install, it climbed the ladder.

## KR confidence

| KR | DRI | Prev | Now | Δ | Evidence/note |
|----|-----|------|-----|---|---------------|
| KR1.1 | alex | 0.8 | 0.8 | → | 2nd check-in of ≥6 held on cadence; cycle on schedule |
| KR1.2 | alex | 0.6 | 0.75 | ↑ | Outside project upgraded to Level 2 v0.5.0; ≥2-weeks-of-use bar crossed ~2026-08-13 |
| KR1.3 | alex | 0.5 | 0.7 | ↑ | 5 of 8 fixes merged (DRI's count): state-writes bundle ×3 (#2) + check suite as the promoted template-validation CI (#9/#10) + staleness-on-held-check-ins fix (#12) |
| KR2.1 | alex | 0.6 | 0.6 | → | no new datapoint — naive-user timing slipped a second week; carried into focus |
| KR2.2 | alex | 0.5 | 0.7 | ↑ | 0.7 anchor hit exactly: install.sh + README/adoption docs merged (#18) |

## What moved

Covers 2026-07-30 → 2026-08-07. 12 PRs merged, all classified; `check` green on main.

- **KR1.3** — state-writes bundle (#2: issues-as-inbox, batched ledger writes, protect-main
  best practice), then the testing arc: plan (#7), Phase 0 (#9), Phase 1 (#10, #12). 24
  checks now guard the repo, red-first on live defects — a shell injection in a shipped
  template, a silently-dead script, and the staleness tripwire this very file had been
  suppressing.
- **KR2.2** — install.sh merged (#18): marketplace registration + install with no /plugin
  dialog, documented in README + adoption.md. 0.7 anchor hit; 1.0 (clean repo → Level 0,
  <10 min, README only) untouched.
- **KR1.2** — live-coaching session docs (#3, #4, #5); the outside project upgraded to
  Level 2 v0.5.0 (non-code ledger — the work lives in the adopter's own repo).
- **Side-quests** — evidence doctrine (#6), adopter prescription Phase 1.5 (#17). Both
  boxed, both closed under box.
- **Maintenance** — local test loop (#14): pre-push hook + parity advisory. 1 of 12 PRs
  (~8%). No `emergency` work.

## What's blocked

Nothing blocked.

## Health metrics

| Metric | Red line | Now | OK? |
|--------|----------|-----|-----|
| docs/skills cross-file coherence | reviewer-class contradiction on main >1 week | none known; `check` green on main | OK |
| Install footprint | anything beyond `okrdev/` + marked CLAUDE.md block (+ opt-in `.github/` at L2) | outside L2 install this week; opt-in `.github/` permitted at L2, nothing flagged by DRI | OK |
| Coach nag rate | interrupts obviously-maintenance work even once | zero interruptions in any ledger | OK |

## Drift check

Clean. 11 of 12 PRs carry a KR line; #2 (state-writes bundle) has none but matches W31's
focus line word-for-word — classified `KR: 1.3` with DRI approval at this check-in. No
`emergency` work; maintenance share ~8% — no taxonomy flags.

## Judgment calls

- 2026-08-06 — alex — prescribe the engineering methodology (BDD/TDD) to adopters: "okrdev
  is an opinionated framework with OKRs, DDD, PR preview environments, Vercel/Neon" —
  reverses coach's keep-it-agnostic advice; scoped to the stack module + Level 2 rails in
  docs/testing.md — claude/bdd-tdd-plan-d2rzoz / PR #7
- 2026-08-06 — alex — docs/adoption.md is the wrong side of the Level 1 install
  contradiction, not skills/install/SKILL.md: plan owns the cycle file, and an empty
  okrdev/checkins/ is not trackable by git anyway. Unblocked the footprint manifest —
  claude/github-issue-11-78c261 / PR #12
- 2026-08-06 — alex — staleness is measured on held check-ins (a confidence table with rows),
  not on the newest week file: logging an override was disarming the tripwire it should have
  been neutral to, live since this very file was created —
  claude/github-issue-11-78c261 / PR #12
- 2026-08-07 — alex — PR #2 classified `KR: 1.3` post-hoc (no KR line in body; matches W31's
  focus line verbatim) — main
- 2026-08-07 — alex — #15 (local loop as 5th prescription item) sanctioned as an immediate
  side-quest at the check-in, box 1h — spend can now reach the 4h/week line; W32 also opened
  more box-hours (7h) than the weekly budget, first time, flagged as a watch item — main
- 2026-08-07 — alex — outside-adopter and collaborator references anonymized across
  check-ins, testing.md, parking-lot.md, and GitHub threads ahead of open-sourcing;
  substance, dates, and numbers unchanged — claude/open-source-readiness-scan-uepee7 / PR #26
- 2026-08-07 — alex — build the okrdev website now (open-core + DIY/DWY/DFY help paths in
  site/): DRI-directed; serves O2's funnel but no KR measures it, so classified side-quest —
  fourth box of the week (2h), taking opened box-hours to 9h and spend past the 4h line
  (~6h), extending the watch item above — claude/okrdev-site / PR #28
- 2026-08-08 — alex — port okrdev to the OpenAI/Codex plugin directory and submit it, starting
  now: serves no KR and is the largest single item on the board, against a cycle with ~7 weeks
  left, KR2.1 slipped two weeks running, and KR1.3 at 3 of 8. Coach recommended parking it to
  W33 triage or a 3–4h spike; DRI overruled — the skills format is byte-identical
  (`skills/<name>/SKILL.md` + frontmatter), the manifest fields map 1:1 onto the ones PR #5
  just added, and Codex reads `.claude-plugin/marketplace.json` as legacy-compatible, so the
  port is the cheapest it will ever be. Accepted risk: this competes directly with the dogfood
  cycle KR1.1 depends on. Watch at the W33 check-in — claude/codex-port
- 2026-08-08 — alex — the Codex port reclassified from `side-quest` to **KR2.2**, same reasoning
  as the Anthropic directory submission: a second listing makes an external user more likely to
  install quickly and easily, which is KR2.2's 1.0 outcome. Coach's one caveat, recorded rather
  than argued: the Anthropic case listed a thing that already ran, while this builds it to run
  elsewhere first — so KR2.2's 1.0 anchor ("clean repo → Level 0 in under 10 minutes following
  only the README") cannot be claimed on Codex until the install path is verified end-to-end
  against the real CLI. Side-quest budget is unaffected by the port from here — PRs carry
  `KR: 2.2` — but W32's existing ~6h/4h breach stands on its own — claude/codex-port
- 2026-08-09 — alex — draft the feature-flags stack row now (flags-as-code via the Flags SDK
  + opt-in Global Config kill-only brake in docs/stack.md), out of the feature-flagging
  solution debate: serves no active KR, classified side-quest, box 2h — fifth box of W32,
  opening box-hours to 11h with spend already past the 4h line (~7.5h), extending the
  existing watch item. Split-test engine deliberately not wired (traffic floor); recorded in
  the row itself — claude/feature-flagging-solution-comparison-a0dd33 / PR #15

## Parking lot triage

Both inboxes swept to zero — 15 items decided (DRI approved coach recommendations as a
batch, with one amendment: #15 runs now, not next week).

Issues:
- #19 pre-push hook dead in worktrees → **promoted → KR1.3** (chafe: hit in this very worktree)
- #16 Node pins ×5 with no rail keeping them equal → **promoted → KR1.3** (coherence-metric class)
- #13 branch-protection.sh claims code-owner review with no CODEOWNERS → **promoted → KR1.3**
- #15 local loop as 5th adopter-prescription item → **side-quest, box 1h, starting now**
  (decision-sourced, same class as Phase 1.5)
- #8 plugin payload ships the whole repo → **next-cycle candidate** (effort M–L; KR2.2's
  1.0 anchor doesn't need it)

Captured:
- Live-call coaching capability (Zoom RTMS) → **next-cycle candidate**
- Ship the live-coaching rig (templates/live-coach/) → **next-cycle candidate**
- Repo-topology doctrine (seam test, Expo exception, federation note) → **next-cycle
  candidate** (merges into the 2026-07-30 repo-topology candidate)
- Planning throwaway-prototype homework → **next-cycle candidate** (trial at next planning
  session)
- roles.md DRI-inversion paragraph → **next-cycle candidate** (watch: promote only if
  load-bearing in real coaching)
- acme-fitness refresh → **next-cycle candidate** (narrative-floor trigger hasn't fired in a
  live cycle yet)
- Sync-wrinkles maintenance PR → **archived** (delivered by #12/#14; the README repo-map
  sliver folds into the coherence sweep)
- Coherence sweep (~2026-08-12, one week after evidence.md) → **promoted → W33 focus** (KR1.3
  — defends the #1 health metric)
- Vocabulary watch (build/box/buy) → **next-cycle candidate** (no adopter data yet)
- Level 2 "anything to click?" nudge → **next-cycle candidate** (gated on dogfooding showing
  the want)

Side quests: #15 opened at box 1h (see Judgment calls). W32 closed boxes: 4h + 2h boxed,
3h spent, both done.

## Focus for next week

- alex: Time the clean naive-user install end-to-end, clean repo → Level 0 (carried from
  W31 — second naming, so it runs this time) → KR2.1 (and KR2.2's 1.0 anchor)
- alex: Outside-project week 2 — real Level 2 use through 2026-08-13 → KR1.2
- alex: Coherence sweep + the promoted fixes (#19, #16, #13) as capacity allows → KR1.3