Lane 4 · Salvage & finish

Orphaned, discarded & unfinished work worth shipping

A full sweep of 727 remote branches, the followups ledger, spec index, and Program-3 status surfaces. The large majority of “open” branches are already squash-merged (pure git hygiene); the genuinely orphaned items below are few, specific, and several are cheap wins that directly affect UAT.

727

remote branches swept

branches with real unmerged value

high-value followup bugs, ownerless

landmine PR to close (deletes CLAUDE.md)

Top salvage targets (ordered by value ÷ effort)

#	Item	Where	Why it matters	Effort
1	PR #519 draft-autosave fix — autosave PUT clobbers `canvas_state`/suggestions; committed-prompt replay miss	`origin/codex/workflow-draft-suggest-race-20260629` (commits `be6c54dc`, `8e856793`, with tests)	Both bugs verified still live on main (`workflow-session-store.ts:1330–1348`, `useWorkflowCanvasPage.ts:509–521`). Users lose suggester output to an autosave tick.	S — cherry-pick 2 commits
2	Employee blank-page routing gap	`docs/followups/2026-06-13-my-dashboard-routing-gap.md`	Non-admin login hits a blank page — directly blocks UAT of the employee role, i.e. the actual audience.	S–M
3	Workflow-chat persistence P0 — configured-but-unlaunched sessions silently lost on reload	`docs/followups/2026-05-22-workflow-chat-persistence-no-job-spawned.md`	Data-loss-feeling bug in the flagship flow.	M
4	Suggester clobbers manual composer edits (3-track fix plan already written)	`docs/followups/2026-05-22-workflow-part-ui-overhaul-spec.md`	Trust-killer for the compose flow; zero regression coverage today.	M
5	Cross-tenant read risk — owner UI reads cross-employee admin endpoints	`docs/followups/2026-06-13-api-surface-audit.md`	Security/scoping issue you don’t want discovered during UAT.	M
6	Program-3 master plan docs (GOAL.md, INITIAL.md, QUEUE-1..5)	`origin/codex/pre-uat-and-prod-ready-plan` — never merged	STATUS/HANDOFF/implementation notes on main cite `GOAL.md` by path dozens of times; unresolvable for anyone auditing main.	S — docs-only merge
7	Spec 104 windx ADR — see Channel Spine tab (SPN-7)	Uncommitted in the dirty coordinator checkout: `/workspace/win/docs/specs/104-surface-unification-windx.md` (285 lines) + `slice-4a-ratification-package-20260703.md`	Blocks launch decision D10 / PROD-4 — and the finished draft is sitting on disk, one `git add` from safety.	S — commit + index
8	Slice 1e Slack copy/cards — see Slack tab (SLK-3)	storyboard only	Half-day of visible UAT polish once beta is deployed.	S once unblocked

Landmine — close, don’t merge PR #102 / track/20260604-json-persistence-spec looks like a legitimate persistence refactor but deletes CLAUDE.md, AGENTS.md, two skills, and the very followup doc describing this risk (docs/followups/2026-06-04-102-json-persistence-reconciliation.md). Close it explicitly so a future branch-cleanup pass can’t merge it by momentum.

RESTORE-DECISION is probably already done The CPO’s Runs/Workflows UX regression “restore” ask maps to PRs #552/#553/#628/#646 — verified code-complete on main (live duration ticker, thinking toggle, running-detail clarity in JobStatusSection.tsx, EventBlocks.tsx, JobDetailUnified.tsx). What’s missing is deploy + hosted proof, not engineering. Answer the RESTORE-DECISION by deploying and demoing, not by rebuilding.

Also ready-to-build (researched, never started)

From the repo’s own roadmap (CLAUDE.md “Research Completed, Ready to Build”) — not Program-3 orphans, but high-leverage candidates for this push, especially for Slack UX and ops:

Slack interactive elements — Block Kit question buttons / accept–revise / retry-cancel cards; needs the /api/slack/interactive route that now exists. Pairs naturally with SLK-2/SLK-3.
Failure Review meta-agent — on job failure, categorize root cause from the event log (read-only). Directly serves UAT triage velocity.
Job recovery completion — resumeJob/recovery.ts exist; missing POST /api/jobs/:id/recover endpoint + Slack retry wiring. Overlaps SLK-2.
Self-healing cron — proxy/DB/token/zombie-job checks with safe remediation; becomes your first production watch tool (see Deploy tab INS-9).

Discard / defer explicitly (so they stop haunting the backlog)

1f-P2 daemon per-user workspace isolation — correctly post-UAT gated by its own spec.
Embeddings suggester retrieval — no doc, no owner, no scope; wait for miss-rate telemetry from 1f-P1 (#555) to justify it.
~15 already-squash-merged codex/slice-* branches — content verified on main; delete the branches, recover nothing.
PRs #528/#536 — intentionally superseded; working as designed.
Claude-Tag cluster (#454–457) + OKF (#382) — held pending Scott’s business-mandate call; don’t un-draft without it.
June salvage/preservation branches (pre-uat-preservation-20260614, pr-138-conflicts, 057-salvage-main) — need a dedicated audit before deletion; park behind launch.

Plan of attack

Salvage worksheet0/7

ORP-1Cherry-pick #519 fix (be6c54dc + 8e856793) onto main; rerun useComposerSuggestion + workflow-session-store suites

ORP-2Fix employee blank-page routing gap (followup 2026-06-13) — pre-UAT hard requirement

ORP-3Fix workflow-chat persistence P0 + suggester-clobber (followups 2026-05-22), with regression tests

ORP-4Close the api-surface cross-tenant read gap (followup 2026-06-13)

ORP-5Merge the Program-3 plan docs branch (docs-only) so GOAL.md citations resolve on main

ORP-6Close landmine PR #102 with an explanatory comment; delete the ~15 verified-merged codex/slice-* branches

ORP-7Ownership pass over docs/followups: assign or explicitly defer each of the 69 baseline-grandfathered orphans

The ownership-baseline gate keeps new orphans out but the legacy 69 need one triage sweep — an hour with the followups index, not a program.

ORP-8Rescue the dirty coordinator checkout’s untracked strays: Spec 104 draft + ratification package, Program-3 STATUS/HANDOFF/completion-audit, pre-uat+prod-prep-2 packet — commit (or archive) each, then prune the stale .wrkts worktrees

git status in /workspace/win shows the whole Program-3 coordination history exists only on one machine’s disk. The session lost artifacts to stale worktrees twice; don’t let the disk be the only copy of the program record. Fixes the gh-pr-merge-breaking main-checkout worktree while you’re there.

Agent handoff prompt

prompt · ORP-1 cherry-pick #519

Salvage the unmerged PR #519 fix in the WIN repo (bigwinai/win). Branch
origin/codex/workflow-draft-suggest-race-20260629 holds commits be6c54dc and 8e856793
(fix + tests) for two bugs verified still live on main: (a) the workflow-session draft
autosave PUT merge-and-overwrites canvas_state/title on every tick, silently reverting
applied composer suggestions (packages/server/src/services/workflow-session-store.ts:1330-1348);
(b) committedPromptReplay() misses the case where the next prompt equals the committed
prompt (packages/ui/src/components/workflows/hooks/useWorkflowCanvasPage.ts:509-521).

Do: create a fresh worktree from origin/main; cherry-pick both commits, resolving drift
against the current files (the surrounding code has moved since 06-29 — preserve the fix
semantics, not the exact hunks); run bun test on useComposerSuggestion.test.ts,
useWorkflowCanvasPage.test.ts, and the server workflow-session-store suite plus full
cross-package typecheck; red-first proof: confirm each test fails when its fix is reverted.
Open a PR per repo rails (PR record file, proven-vs-deferred in body), then close PR #519
as superseded with a link.

WIN launch workbook · generated 2026-07-06 by Claude (Fable) from session mining + repo audit of origin/main@734eff41 · worksheet state persists in your browser (localStorage, key prefix winwb:orphans)