Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions specs/epics/FIX-1649/BUSINESS-RULES.md
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ checks. Each says who owns it and where it's checked.
| ER-6 | No child restyles an FSD component or hardcodes paint on it inside App Lab. Registry components arrive by `fsdev ui add` and stay unedited copies of their source; a component that can't take the skin is fixed at its source by FIX-1655 and re-synced, and FIX-1655's re-sync check fails on any App Lab copy that differs from its source | [D2](DECISIONS.md#d2). A restyled copy drifts from every later FSD fix, and a stale one makes leg c test the copy, not the skin |
| ER-7 | No child adds an L1 noun (Agent, Team, Channel, MessageBoard, Project-as-required) or puts Workforce concepts into `core` or `engine`; the design-system package imports nothing from `@flow-state-dev/workforce` | Jake's layer rule. L1 offers generic hooks only |
| ER-8 | No child builds anything below [the box's fence](SPEC.md#whats-in-the-box), and none adds `CHANNELS.md` or `kind:` frontmatter | The PRD's and the Architect's invent-kills, and D-12; the fence is their one list |
| ER-15 | No child writes into a worker's session except through that session's shipped operations: a turn from the task composer, a workstream's `@worker` or a reply from Inbox, Interrupt, Hand off. The shell writes no session item of its own and shows a message as delivered only when the session says so. Owned by FIX-1664; FIX-1662's composers (a workstream's, Inbox's reply) consume it. Two writes go to the person's own sessions, never a worker's:<br>· a project room's `read`, `post` and `join`, on the room kind (FIX-1718)<br>· opening the person's room session, on the room kind (FIX-1718, FIX-1752) | A second write path into a harness session is a model the shell invented; the stream's "sent into PAY-14" line must be the session's answer, not the shell's hope |
| ER-15 | No child writes into a worker's session except through that session's shipped operations: a turn from the task composer, a workstream's `@worker` or a reply from Inbox, Interrupt, Hand off. The shell writes no session item of its own and shows a message as delivered only when the session says so. Owned by FIX-1664; FIX-1662's composers (a workstream's, Inbox's reply) consume it. Two writes go to the person's own sessions, never a worker's:<br>· a project room's `read`, `post` and `join`, on the room kind (FIX-1718)<br>· opening the person's room session, on the room kind (FIX-1718, FIX-1752)<br>Both decided at closure, 2026-10-04 ([EVOLUTION](EVOLUTION.md#amendment--2026-10-04--wrap)) | A second write path into a harness session is a model the shell invented; the stream's "sent into PAY-14" line must be the session's answer, not the shell's hope |

## How the set is run

Expand All @@ -38,5 +38,5 @@ checks. Each says who owns it and where it's checked.

| # | The epic is done when | Proved by |
|---|---|---|
| ER-13 | [The goal](SPEC.md#the-goal-and-how-well-know-its-met) is met: legs a, b, c and d pass, leg c fails under its control, and leg d under each of FIX-1737's three controls (leg d added 2026-10-02) | FIX-1663's goal check, in a browser, real model |
| ER-13 | [The goal](SPEC.md#the-goal-and-how-well-know-its-met) is met: legs a to d pass and each control fails where it must, per [FIX-1663's goal check](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) (leg d added 2026-10-02) | FIX-1663's goal check, in a browser, real model |
| ER-14 | The docs teach skinning FSD components from one token set, and App Lab's README says how to open a Lab in it, what each level and destination shows, and how to switch day and night | FIX-1655's, FIX-1662's, FIX-1664's, FIX-1722's, FIX-1723's and FIX-1725's docs, per [DOCS.md](DOCS.md) |
21 changes: 21 additions & 0 deletions specs/epics/FIX-1649/EVOLUTION.md
Original file line number Diff line number Diff line change
Expand Up @@ -58,3 +58,24 @@ Out of this amendment, and staying with their owners: rows waiting on FIX-1650,
data (54 rows, some only in part), FIX-1675's watches and FIX-1474's also-post, and the v1
structure the 2026-10-01 amendment kept (X1 to X7). FIX-1736 and FIX-1737 block FIX-1663 in
Linear.

<a name="amendment--2026-10-04--wrap"></a>
## Amendment · 2026-10-04 · wrap

The closure, FIX-1663, merged on [#2709](https://github.com/fixpoint-labs/flow-state-dev/pull/2709)
(`894c418d2`) after a clean run on `01444863c`: every leg and part passed, and every control was
right. Its goal check,
[`goal.md`](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) with
`legs.mts` beside it, is the one home for how legs a to c are checked, as `it-draws-v2s-look/` is for leg d. Two calls were made during the
closure; they are recorded here so no later Lab reopens them. The review records stay
[#2421](https://github.com/fixpoint-labs/flow-state-dev/pull/2421) and the two amendments above.

| Prior intent in this set | Treatment | Reason / evidence | Replacement | Compatibility |
|---|---|---|---|---|
| *"Leg c's sweep, pinned once"*: the react chrome App Lab mounts (navigator, roster, board panels, seat detail), every registry component App Lab copies, the devtool page ([SPEC](SPEC.md#the-goal-and-how-well-know-its-met)) | **Amended**, narrowed to what Shift Manager draws | Shift Manager mounts no `@flow-state-dev/react` chrome; it imports providers, hooks and the item renderer from it, no chrome component. FIX-1663 narrowed its own plan after closure run 1 at `899e059` (EM call with FIX-1688 and FIX-1689, [#2532](https://github.com/fixpoint-labs/flow-state-dev/pull/2532)) and left this text to the wrap ([FIX-1663's evolution](../../issues/FIX-1663/EVOLUTION.md)) | Leg c sweeps the registry cards Shift Manager draws and the devtool page, not react chrome: [SPEC goal §](SPEC.md#the-goal-and-how-well-know-its-met) + [`goal.md`](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) | ER-2, ER-3 and the theme's value list are unchanged. [D2](DECISIONS.md#d2)'s two shelves stay the contract a host may use; Shift Manager uses the registry shelf only |
| Leg c's control: *"one reused component given a hardcoded App Lab colour: leg c must FAIL naming it"* ([SPEC](SPEC.md#the-goal-and-how-well-know-its-met), ER-13) | **Amended**, named | `hardcoded-accent` patches the tool card's copy; on `01444863c` it failed at `c:tool` only. FIX-1664's `worker-session` points the run-lab's Session, where leg c draws four of its parts, elsewhere, so goal.md lets it turn `c:sweep` red beside its `a2` | Each control fails at its own signal only, as the closure check grades it: [SPEC goal §](SPEC.md#the-goal-and-how-well-know-its-met) + [`goal.md`](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) | The control's patch and the theme values are unchanged |
| Leg a's Inbox journey: the ask *"in Inbox with its kind and workstream"*, its card *"in the workstream's stream"*, read as the asking seat's first channel ([SPEC](SPEC.md#the-goal-and-how-well-know-its-met)) | **Amended**, the reading only | Once FIX-1718 put `eng.em` in four channels, Inbox named all four for an ask with no workstream session, and the specs left that case open (the run on `dd9054581` was red at a1 alone). What shipped, and was decided for the closure on 2026-10-04 as right: FIX-1662's [BR-18](../../issues/FIX-1662/BUSINESS-RULES.md) draws an ask's card on the Stream of every channel that has its seat as a member (`asksFor()`), whether or not a post started it; Inbox names the channel whose post started the ask, and otherwise every channel the seat is in (`workstreamsOf()`) ([`6c6030939`](https://github.com/fixpoint-labs/flow-state-dev/commit/6c6030939)) | a1 reads its expected workstreams from the store as Inbox does, holds Inbox to exactly that set, and checks the card on each Stream in it. For an ask a post started, that set is the parent channel only, so a1 proves the parent Stream and not the seat's other channels' Streams. Checking every member Stream is an open follow-up, rolled into the goals/lib helper fold: [SPEC goal §](SPEC.md#the-goal-and-how-well-know-its-met) + [`goal.md`](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) | BR-18, `asksFor()` and Inbox are unchanged; the closure's ask had no parent, so a1 checked all four Streams |
| ER-15: no child writes into a worker's session except through its shipped operations ([BUSINESS-RULES](BUSINESS-RULES.md#what-no-child-may-do)) | **Amended**, narrowly | FIX-1718 and FIX-1752 added two writes part 4 did not name: `sendAction(` in `labs/shift-manager/src/lib/talk.ts` and `createSession(` in `lib/reads.ts`. Both write the person's own sessions on `ROOM_KIND`, never a worker's. Decided for the closure on 2026-10-04 ([`c59c57491`](https://github.com/fixpoint-labs/flow-state-dev/commit/c59c57491)) | The two room-kind writes, as [ER-15](BUSINESS-RULES.md#what-no-child-may-do) states them; part 4 grades them: [SPEC goal §](SPEC.md#the-goal-and-how-well-know-its-met) + [`goal.md`](../../../goals/shift-manager/a-lab-is-worked-through-one-skinned-shell/goal.md) | No new write into a worker's session; ER-15's other doors and its owner are unchanged |

[The path](PLAN.md#the-path) is redrawn as it shipped. The set table stays the 30 September
snapshot; live state is Linear and the implementation PRs.
58 changes: 31 additions & 27 deletions specs/epics/FIX-1649/PLAN.md
Original file line number Diff line number Diff line change
Expand Up @@ -8,19 +8,23 @@ and [BUSINESS-RULES.md](BUSINESS-RULES.md) (ER-n).

## The path

![The path, as of 30 September 2026, drawn in phases rather than dates because Cycle PM has not scheduled the build. An input lane for the design: hand-back v1 came back during review; design pass 2 runs from the gate to the final hand-back, a vertical line that gates final visuals only. An input lane for the sibling epics, which nothing waits on. This epic spec in review, with the now line on it. FIX-1655: spec, then the token contract, neutral defaults, skin fixes and the re-sync check, then the light and dark App Lab values after the final hand-back. FIX-1662: spec, then the frame and the project and workstream levels, then its final visual pass. FIX-1664: spec beside FIX-1662's, then the task level once FIX-1662's frame lands, then its final visual pass. FIX-1663: the QA plan written beside them, and its run after all three merge. The critical path runs the gate, design pass 2, the final passes and the closure run](figures/path.svg)

Three lanes run from the gate, and all split at the same wait: the final hand-back. Everything
structural starts at the gate, except the task view's build, which waits for FIX-1662's frame;
only the final look waits on design ([ER-9](BUSINESS-RULES.md#how-the-set-is-run)). So design
pass 2, not any child, is the critical path's long pole. The dependency graph is in
[the spec](SPEC.md#how-the-issues-flow-into-each-other); this adds time.

The figure is the 30 September path. The 2026-10-01 amendment adds three lanes after it: FIX-1722
and FIX-1723 (spec, then build in the shipped frame) and FIX-1725 (built directly). They start
after the final hand-back, and the closure run now waits for them too. The 2026-10-02 amendment
adds two more: FIX-1736 (the fonts, built directly), then FIX-1737 (v2's look: spec, then four
PRs), and the closure's leg d runs FIX-1737's check.
![The as-shipped path, 30 September to 4 October 2026; the detail is in the prose below](figures/path.svg)

Each bar runs from filed (or the gate) to merged. FIX-1655, FIX-1662 and FIX-1664, with the DevForce
lab prerequisites FIX-1666 to 1668, specced and built on 30 September. Closure run 1's findings
(FIX-1684, FIX-1688 to 1696) merged overnight, and FIX-1690's door into a running session on 1
October. Design v2 landed that evening, then FIX-1725, FIX-1722 and FIX-1723. The late fixes,
FIX-1730 to 1735, FIX-1747 and FIX-1749, landed on 2 and 3 October. The v2-look audit filed
FIX-1736, merged on 2 October, and FIX-1737, whose four slices landed early on 4 October beside
FIX-1650's FIX-1718 and FIX-1752. The closure ran until one run was clean, on `01444863c`, and
[#2709](https://github.com/fixpoint-labs/flow-state-dev/pull/2709) merged that day.

The long pole was v2's look, not the shell: the structure the 30 September plan named merged on
the first day, and the last three went to design v2, the look it set
([ER-16](BUSINESS-RULES.md#what-a-team-gets-and-what-it-doesnt)) and the closure's findings. The
plan's critical path ran through design pass 2 and the final passes; that held, with FIX-1736 and
FIX-1737 as the final pass. Four children filed under the
epic (FIX-1671, 1673, 1675, 1705) are not on the path and did not block the closure.
Comment thread
jhoffner marked this conversation as resolved.
Resolved

## The design hand-back is an input, not a child

Expand Down Expand Up @@ -54,23 +58,23 @@ v1 fixes is an amendment to this epic ([ER-10](BUSINESS-RULES.md#how-the-set-is-
FIX-1655 carries no Kind label in Linear, so its route defaults to spec. FIX-1662 and FIX-1664
are Features; FIX-1663 an Improvement.

## Where it is
## Where it ended

[The set table](SPEC.md#the-set--as-of-2026-09-30) is the review-time snapshot; follow its
Linear links for live state. The build is a Cycle 2 candidate: Cycle PM decides when it starts.
[The set table](SPEC.md#the-set--as-of-2026-09-30) is the review-time snapshot. The build was
planned as a Cycle 2 candidate; it ran from the gate on 30 September to the closure's merge on
4 October, as [the path](#the-path) shows. Linear and the implementation PRs hold the final state.

## What unblocks what, from here
## What unblocked what, as it happened

1. **This spec is approved and merged** → design pass 2 goes to Claude Design; the FIX-1655,
FIX-1662 and FIX-1664 specs start in parallel; FIX-1663's QA plan is written beside them.
2. **FIX-1662 merges** → FIX-1664's build merges into its frame.
3. **The final hand-back is linked on FIX-1649** → FIX-1655 sets the light and dark values;
FIX-1662 and FIX-1664 do their final visual passes. A structural change goes to an
amendment first.
4. **Every other child merges** (FIX-1655, FIX-1662, FIX-1664, and since the 2026-10-01
amendment FIX-1722, FIX-1723 and FIX-1725, and since 2026-10-02 FIX-1736 and FIX-1737) →
the closure run, on one `main` commit, legs a to d. A finding is filed as a child
of FIX-1649 that blocks FIX-1663; the whole plan runs again after the last fix merges.
1. **The spec merged** (30 September): the FIX-1655, FIX-1662 and FIX-1664 specs ran in parallel,
with FIX-1663's QA plan beside them, and design pass 2 went to Claude Design.
2. **FIX-1662 merged**, and FIX-1664's build merged into its frame the same day.
3. **The final hand-back, v2, landed** (1 October): FIX-1655's light and dark values came with it.
Its structural changes went to the 2026-10-01 amendment first (FIX-1722, FIX-1723, FIX-1725),
and its look to the 2026-10-02 amendment (FIX-1736, FIX-1737).
4. **Every other child merged**, and the closure ran on one `main` commit, legs a to d. Each
finding was filed as a child of FIX-1649 blocking FIX-1663, and the whole plan ran again after
the last fix merged, until it was clean on `01444863c`.

The plan is written on [D3](DECISIONS.md#d3): one app, so no child builds a chrome package.

Expand Down
Loading
Loading