From c0dfaff7daf3e8e6ef364d5a712680e84acf47a1 Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 17 Sep 2026 00:51:56 +0000 Subject: [PATCH 01/19] spec(project): Workforce: Layer 2 Abstraction MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The standing set for the Linear project: the outcome and the territory, the calls that bind more than one epic, the rules every epic obeys with owners, and the arc. Absorbed from the project's existing Linear content (6,060 chars) rather than written fresh — the substrate/convention test, the Layer 1/Layer 2 split, the agents-as-humans framing and the settled renames were all already there with nowhere above one epic to live. Nothing was dropped as stale; the vocabulary manifest stays in Linear as the reference list it is. Two things the set records rather than tidies: FIX-1359 is marked done with 7 of 11 issues closed, and 32 of 65 work issues sit under no epic at all — the vocabulary era the project began in. Both are the useful signal. PR-4 carries the W3 correction in its general form — every derived key is refused, not the one key found first — because the original seven-rule table rotted exactly once already. One open question, verified not assumed: rooms have no declaration surface (`RoomManifest|createRoom|roomRegistry|defineRoom` → 0 matches across packages/), while FIX-1352 describes them as written in code. It binds W3 and W4, so it is recorded as an ask rather than answered here. Figures rendered in both themes before commit; the arc lost its dashed not-started bars, which read as duration. Verify: fences even, no risky mermaid labels, both SVGs self-contained with aria-labels and under 70 lines, budgets 572/799/502/444 against 700/900/600/700. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01MpacHaJN9Rba13WrsPJosZ --- .../BUSINESS-RULES.md | 43 +++++++++ .../DECISIONS.md | 95 +++++++++++++++++++ .../workforce-layer-2-abstraction/PLAN.md | 44 +++++++++ .../workforce-layer-2-abstraction/SPEC.md | 62 ++++++++++++ .../figures/arc.svg | 69 ++++++++++++++ .../figures/territory.svg | 54 +++++++++++ 6 files changed, 367 insertions(+) create mode 100644 spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md create mode 100644 spec/_projects/workforce-layer-2-abstraction/DECISIONS.md create mode 100644 spec/_projects/workforce-layer-2-abstraction/PLAN.md create mode 100644 spec/_projects/workforce-layer-2-abstraction/SPEC.md create mode 100644 spec/_projects/workforce-layer-2-abstraction/figures/arc.svg create mode 100644 spec/_projects/workforce-layer-2-abstraction/figures/territory.svg diff --git a/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md new file mode 100644 index 0000000000..71636cd0be --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md @@ -0,0 +1,43 @@ +# Rules — Workforce: Layer 2 Abstraction + +[Spec](SPEC.md) · [Decisions](DECISIONS.md) · **Rules** · [Plan](PLAN.md) + +What every epic under this project obeys. Each rule has **exactly one owner** — the epic that +implements and proves it; the others inherit it. A rule with two owners is a seam; a rule with none +is a wish. + +| | Rule | Owner | Checked where | +|---|---|---|---| +| **PR-1** | When a concept is proposed for this project, then it ships here only if there are **several valid assemblies** of it; exactly one → it is substrate and lives in `core` / `engine` / `orchestration` | FIX-1332 | The epic objective gate | +| **PR-2** | When a convention needs behaviour, then it **composes Layer 1 primitives** and never introduces a parallel primitive — Layer 2 stays transparently implemented, no black box | FIX-1332 | Package-boundary review | +| **PR-3** | When a convention ships, then it ships with a **reader and a consumer that exercises it**, not a format alone | FIX-1351 | The thin pentest lab | +| **PR-4** | When a file declares a thing, then every key the convention **derives** is refused by name — a file may not say where it lives, what its identity is, or what its state shape is | FIX-1351 | Reader tests, per-convention | +| **PR-5** | When a new surface names a Layer 2 concept, then it uses the **settled name** (Role, Strategy, Instructions, stream visibility), never the superseded one | FIX-1407 | The propagation pass | + +**No epic may:** + +- Add a **second claim system**. There is one task board and correctness depends on that (PD-2). +- Expose the **agent bundle** — Persona, Skills, Memory, Instructions, Strategy — as the public + API. An Agent is invoked or assigned work, not decomposed (PD-3). +- Run the **propagation pass** before the vocabulary locks (PD-4). Renaming an unratified model + costs the rename twice. +- Put a concept here **because it is convenient**. PR-1 is a test, not a preference. + +--- + +## Where PR-4 came from, and why it is worded this way + +PR-4 is the one rule here that was **corrected after it was first written**, and the correction is +the reason it names *derived keys* rather than a specific key. + +It began inside `FIX-1354 §7` as a table of seven inherited shape rules — a rule set living inside a +document it governs, where a sibling epic could not find it. Two of those rules were wrong; the +correction reached a comment on a Linear issue and never reached the table. Review then found, by +running it, that carrying the remaining frontmatter through verbatim let a file point itself at +another document's storage row. + +So the rule is stated as the general form — *every derived key is refused* — rather than as the one +key that happened to be found first. **That generalization is the rule; a list of refused key names +would rot the same way the original table did.** + +This is also the clearest argument for why this document exists at project altitude at all. diff --git a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md new file mode 100644 index 0000000000..6798c208d8 --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md @@ -0,0 +1,95 @@ +# Decisions — Workforce: Layer 2 Abstraction + +[Spec](SPEC.md) · **Decisions** · [Rules](BUSINESS-RULES.md) · [Plan](PLAN.md) + +Calls that bind **more than one epic**. A call that touches one epic belongs to that epic's own +`DECISIONS.md`; absorbing it here would make this a design nobody signed off. + +--- + +### PD-1 · The test is not complexity — it is whether there is one of it + +**Binds** every epic in the project, and every future one. + +A concept belongs to Layer 1 when there is **exactly one** of it and correctness depends on that. +It belongs to Layer 2 when it is **one opinionated assembly among several valid ones**. A +convention may still carry real capability and be far more than config; being more than config +does not make it substrate. + +Worked: **Task is substrate** — it is the claim system, and there is one of it. **Plan is a +convention** — there are any number of ways to express a plan with what FSD already has, including +a resource a flow writes to. Same for Strategy, Role, Archetype and Team. + +**Rejected:** sorting by complexity, which puts anything hard into the substrate and leaves Layer 2 +as config. **Costs** a concept that fails the test stays out even when it would be convenient to +put it here. + +### PD-2 · Task moved to Layer 1 + +**Binds** FIX-1333 (the MCP door over boards), FIX-1407 (work routing), and anything that claims. + +The task board began as an extension and is now core, because **claiming** is the thing correctness +depends on. Tasks are a unit of work with a goal, assigned to an agent — and there is one claim +system, so PD-1 puts it below the fence. It lives in `orchestration`. + +**Costs** Layer 2 cannot redefine what a task is; a convention that wants different claim semantics +is asking for a second substrate and the answer is no. + +### PD-3 · Agents are participants, not LLM-in-a-loop building blocks + +**Binds** FIX-1332 (thin Agent), FIX-1359 (the default agent flow), and every archetype. + +An Agent is a complete participant system. Internally it bundles a Persona, Skills, Memory, +Instructions and a Strategy; externally it is **invoked or assigned tasks, not decomposed**. That +is what makes `agent` a flow kind someone can replace wholesale rather than a pattern they wire. + +**Rejected:** exposing the bundle as the API, which makes every consumer re-assemble an agent and +makes "replaceable" meaningless. + +### PD-4 · The vocabulary is ratified by code shapes, not by agreement + +**Binds** every epic; it is why the project is sequenced the way it is. + +The model is not locked until concrete code ratifies it. This is why W2 (conventions and the seat +factory) ran first and why W3 carries **a thin pentest lab** rather than only readers: a convention +with no consumer has not been tested, it has been asserted. + +**Costs** the propagation pass — in-flight tickets, docs IA, the README, renaming the patterns +package to strategies, the `agentType` field — waits for the lock and gets more expensive the +longer it waits. Accepted deliberately; a propagation pass over an unratified vocabulary is worse. + +### PD-5 · The renames are settled, and old names are not alternatives + +**Binds** every epic, every doc page, and the propagation pass. + +Identity → collapsed to a system id field, Persona is the who-concept · Worker → **Role** · +Pattern → **Strategy** · Pattern-skill → *a Skill that carries a Strategy* · Approach → Strategy · +Prompt template → **Instructions** (concept) + `.md` + frontmatter (format) · `agentType` → **stream +visibility** · Actor → kept out of the primary vocabulary; Agent owns the slot. + +**Costs** code still carries old names until the propagation pass runs, so the tree and the +vocabulary disagree in the meantime. That gap is expected, not a defect to file. + +--- + +## Decided once + +Answers settled at project altitude, so no later epic reopens them. + +| Question | Answer | Where it was settled | +|---|---|---| +| Does a Plan get a substrate primitive? | No — convention (PD-1) | Vocabulary pass, absorbed from the project's Linear content | +| Who owns the claim system? | Layer 1, `orchestration` (PD-2) | Same | +| Is an Agent decomposable by its consumer? | No (PD-3) | Same | +| Does the propagation pass run before the vocabulary locks? | No (PD-4) | Same | + +## Open + +**Rooms have no declaration surface at all, and two documents disagree about it.** W3's +`FIX-1352` describes rooms as "written in code" in one section and reports zero matches for any +room declaration API in another. Verified: `RoomManifest`, `createRoom`, `roomRegistry` and +`defineRoom` return **no matches** across `packages/`. Whether the file convention is therefore +introducing rooms or *replacing* a code surface changes what W3 ships and what its docs claim. + +This is recorded as an ask rather than answered here, because it is a product call about what we +tell people rooms are — and it binds W3 and W4 both. It belongs on whichever epic reaches it first. diff --git a/spec/_projects/workforce-layer-2-abstraction/PLAN.md b/spec/_projects/workforce-layer-2-abstraction/PLAN.md new file mode 100644 index 0000000000..0a479ed1ea --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/PLAN.md @@ -0,0 +1,44 @@ +# Plan — Workforce: Layer 2 Abstraction + +[Spec](SPEC.md) · [Decisions](DECISIONS.md) · [Rules](BUSINESS-RULES.md) · **Plan** + +Sequencing, not building. What order the epics run in, what each hands the next, and what is +deliberately not next. Each epic's own plan owns its checks. + +![The arc](figures/arc.svg) + +Nine days, not three months. The project opened in June, but every epic here was filed from +**Sep 8** onward — before that, the work ran as 32 unparented vocabulary issues. Only one epic has +ever been in flight at a time, which is the cap doing its job rather than a queue forming. + +## What each epic consumes and releases + +| Epic | Consumes | Releases | +|---|---|---| +| **W2** · FIX-1332 | Layer 1: blocks, resources, scope state | The conventions, the seat factory, a thin Agent — everything below depends on this | +| **agent flow** · FIX-1359 | W2's thin Agent | A replaceable OOTB `agent` flow kind, so an app overrides rather than assembles | +| **W3** · FIX-1351 | W2's walk primitives; the agent flow as a consumer | The file surface — channels, resources, skills — and the pentest lab that proves PR-3 | +| **W1** · FIX-1333 | W2's conventions; Layer 1 boards + dispatcher | The MCP client door over intake, boards and dispatch | +| **W4** · FIX-1407 | W3's file surface; the settled vocabulary | Work routing, package cohesion, and the PR-5 propagation pass | + +**W1 is the one that unblocks nothing**, which is why it can sit unstarted at the bottom of the arc +without holding anything up — and also why it is the easiest to keep deferring. Its cost is not +delay to the others, it is that the client door stays shut. + +## Coordination seams + +- **W3 → W4.** W4's propagation pass (PR-5) cannot run until W3 locks the file-surface vocabulary. + Starting W4 early buys nothing and risks renaming twice (PD-4). +- **W3's lab is the proof for PR-3**, so any epic that ships a convention before the lab exists is + asserting the rule rather than checking it. +- **The rooms question in [Decisions](DECISIONS.md) → Open binds W3 and W4 both.** It should be + answered on whichever reaches it first, and recorded there, not answered twice. + +## What is deliberately not next + +- **The propagation pass.** Gated on the vocabulary locking (PD-4), not on capacity. +- **Tasks, Skills internals, Memory implementation.** Own projects; this one owns the vocabulary. +- **Re-parenting the 32 unparented issues.** They are the pre-epic era. Sweeping them under epics + retroactively would make the arc look tidier and tell you less about what actually happened. +- **A second harness or transport for the MCP door.** W1 has not started; widening it before it + does is scope with no consumer. diff --git a/spec/_projects/workforce-layer-2-abstraction/SPEC.md b/spec/_projects/workforce-layer-2-abstraction/SPEC.md new file mode 100644 index 0000000000..82ec59aaec --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/SPEC.md @@ -0,0 +1,62 @@ +# Workforce: Layer 2 Abstraction + +**Spec** · [Decisions](DECISIONS.md) · [Rules](BUSINESS-RULES.md) · [Plan](PLAN.md) + +Layer 2 is the everyday API most people who use FSD will live in: Agents, Teams, and the +conventions that assemble them. It is transparently implemented on Layer 1 — no black box — but +the README, the docs and the common path lead with it. This project owns the vocabulary and the +epics that establish it. + +## The outcome + +| | | +|---|---| +| **Winning when** | Someone stands up a working team of agents from files alone — no TypeScript to declare a room, a resource, or a skill — and Layer 1 stays fully available to anyone who wants past the conventions | +| **The read** | Layer 2 nouns that still require code to declare. Five at the start, and the count only moves when a convention ships with a reader and a proof | +| **Now** | 2 epics done · 1 in flight · 2 not started. 65 work issues, of which **32 still sit under no epic** — the vocabulary era this project began in, before the epic structure existed | +| **Kill line** | If real apps turn out to want to assemble Layer 2 themselves, this project is mis-shaped rather than unfinished: what changes is that the conventions become examples, not the remaining epic list | + +![The territory](figures/territory.svg) + +Above the fence is what this project owns; below it is the substrate it assembles but does not +own. The fence is one question — *is there exactly one of it?* — and it is the test every epic +under this project is checked against. + +## The epics — as of 2026-09-17 + +| Epic | What it owns | State | Issues | +|---|---|---|---| +| [FIX-1332](https://linear.app/fixpoint-labs/issue/FIX-1332) · **W2 foundation** | Conventions, seat factory, thin Agent | **done** | 5/5 | +| [FIX-1359](https://linear.app/fixpoint-labs/issue/FIX-1359) · **default agent flow** | OOTB replaceable `agent` flow kind | **done** | 7/11 | +| [FIX-1351](https://linear.app/fixpoint-labs/issue/FIX-1351) · **W3 file surface** | Channels, resources, skills + a thin pentest lab | **in flight** | 5/14 | +| [FIX-1333](https://linear.app/fixpoint-labs/issue/FIX-1333) · **W1 MCP door** | Client door over intake, boards, dispatcher | *not started* | — | +| [FIX-1407](https://linear.app/fixpoint-labs/issue/FIX-1407) · **W4 routing** | Work routing & package cohesion | *not started* | 0/4 | + +2 done · 1 in flight · 2 not started. + +**Two things in that table are worth reading twice.** The numbering is not the order — W1 was filed +the same day as W2 and has not started, while W2 finished in four days. And **FIX-1359 is marked +done with 7 of 11 issues closed**, which is either a wrap that outran its children or four issues +that should have moved out of it; it is recorded here as Linear has it rather than tidied, because +the discrepancy is the useful signal. + +```mermaid +flowchart LR + W2["FIX-1332 · W2 foundation"] --> AF["FIX-1359 · agent flow"] + W2 --> W3["FIX-1351 · W3 file surface"] + AF --> W3 + W3 --> W4["FIX-1407 · W4 routing"] + W2 --> W1["FIX-1333 · W1 MCP door"] +``` + +W2's conventions and seat factory are what everything else builds on. W1 is the one epic that does +not depend on the file surface, which is why it can sit unstarted without blocking anything. + +## What this project is not + +- **Not where primitives live.** Blocks, resources, the task board's claim system and dispatch are + Layer 1. They are assembled here and owned in `core` / `engine` / `orchestration`. +- **Not the implementation of Tasks, Skills internals, or Memory.** Those run in their own + projects. This one owns the vocabulary and the few epics that establish the surfaces. +- **Not a lock on the model.** The vocabulary is ratified by concrete code shapes, not by + agreement; see [Decisions](DECISIONS.md) → PD-4. diff --git a/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg new file mode 100644 index 0000000000..4af838f889 --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg @@ -0,0 +1,69 @@ + + + + + EPIC + SEP 8 + SEP 12 + SEP 16 + + + + + + W2 · foundation + FIX-1332 · 5/5 + + conventions, seat factory, thin Agent + + + + agent flow + FIX-1359 · 7/11 + + OOTB replaceable `agent` flow kind + + + + W3 · file surface + FIX-1351 · 5/14 + + channels · resources · skills · pentest lab + + + + W1 · MCP door + FIX-1333 · not started + filed Sep 8, before W2 — the numbering is not the order + + + + W4 · routing + FIX-1407 · not started + filed Sep 16, the day W3 started + + + + + NOW · SEP 17 + + diff --git a/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg b/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg new file mode 100644 index 0000000000..7855d5e4fd --- /dev/null +++ b/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg @@ -0,0 +1,54 @@ + + + + + + THIS PROJECT · LAYER 2 CONVENTIONS + Agent · Persona · Skill · Memory + the canonical nouns — one of each, assembled for you + Strategy · Role · Archetype · Team · Plan + conventions — one opinionated assembly among several valid ones + + + + + NOT HERE · OWN PROJECTS + Task & Plan implementation + Skills internals + Memory implementation + this project owns the vocabulary + + + + + Is there exactly one of it? One → substrate, below. Several valid → convention, above. + + + + + LAYER 1 SUBSTRATE · ASSEMBLED HERE, OWNED ELSEWHERE + Blocks · Scope state · Resources · Instructions format · Sub-flow invocation + Task Board · Dispatch · Stream visibility + correctness depends on there being exactly one — core / engine / orchestration + + From c012d78f71bf34cae09c18320d5cf5c5ccadf05d Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 17 Sep 2026 00:53:47 +0000 Subject: [PATCH 02/19] ci: port the project/* spec-folder exemption onto this branch Process guards failed here for the reason the machinery PR fixes: the exemption is keyed on branch name and main's copy lists only spec/* and epic/*, so a project/ branch carrying spec/_projects// is checked and fails. The fix is already on #1816. Ported verbatim rather than waiting for that to merge; it no-ops the moment main carries it. This branch never merges, so the duplicate hunk dies with it. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01MpacHaJN9Rba13WrsPJosZ --- .github/workflows/ci.yml | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 41a451c5ae..2ec6e1d399 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -75,9 +75,10 @@ jobs: with: node-version: 22 - # A spec PR (branch `spec/`) and an epic PR (branch - # `epic/`, carrying `spec/_epics//`) hold the in-flight spec - # by design and are never merged, so they are exempt. Keyed on the BRANCH, + # A spec PR (branch `spec/`), an epic PR (branch `epic/`, + # carrying `spec/_epics//`) and a project PR (branch + # `project/`, carrying `spec/_projects//`) hold the in-flight + # spec by design and are never merged, so they are exempt. Keyed on the BRANCH, # not the `spec` label: the label is applied after the PR is created, and # `labeled` is not a default `pull_request` event type — a label-keyed # exemption would fail on the opening run and never re-run. @@ -85,7 +86,8 @@ jobs: - name: Spec folder is clear if: >- github.event_name == 'push' || - !(startsWith(github.head_ref, 'spec/') || startsWith(github.head_ref, 'epic/')) + !(startsWith(github.head_ref, 'spec/') || startsWith(github.head_ref, 'epic/') || + startsWith(github.head_ref, 'project/')) run: node scripts/validate-spec-folder.mjs - name: Changesets name their Linear issue From 5f57332fd00004cff3e90f8f337982671e49908d Mon Sep 17 00:00:00 2001 From: Claude Date: Thu, 17 Sep 2026 01:53:09 +0000 Subject: [PATCH 03/19] spec(project): Channel is the L2 noun; room was retired before I wrote this MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The Architect review caught the project and its in-flight epic teaching two inventories, and verifying it showed the split was worse than flagged. W3's own record: "channel replaces room, folding the rooms / L2-channels dual into one file type" (Sep 11), with the objective approved that evening. D1 makes a channel L2 opinion — a replaceable flow kind, not a new L1 type — and FIX-1352 shipped the team-scope convention. CHANNEL.md exists and is done. Across W3's four documents: channel 88 mentions, room 1, and that one is the line recording the rename. So: - The territory figure's owned box gains Channel, which it never listed. - The winning-when line said "declare a room" — a dead noun — now "a channel". - PD-5 gains room -> Channel, which is exactly what that card is for. - Decided once gains the pair: a channel is not an L1 type, and room is not a second thing. The Open section is the real correction. It claimed rooms had no declaration surface and asked which way W3 should go. The grep behind it was sound — zero matches, re-run not recalled — but room had already been retired, so the zero was the fold working rather than a gap. I read a superseded noun out of one issue's spec PR instead of the epic's current record, which is precisely the drift this altitude exists to stop. Replaced with what W3 actually leaves open: org/channels/ locked open, read by nothing, owned by no issue. Rendering the figure caught the new caption overrunning its box into the neighbouring panel; shortened before commit. Verify: fences even, no risky mermaid labels, both SVGs self-contained with aria-labels, nav lines correct, budgets 595/896/515/478 against 700/900/600/700 (DECISIONS needed three trims to get back under). Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_01MpacHaJN9Rba13WrsPJosZ --- .../workforce-layer-2-abstraction/DECISIONS.md | 17 ++++++++++------- .../workforce-layer-2-abstraction/PLAN.md | 5 +++-- .../workforce-layer-2-abstraction/SPEC.md | 2 +- .../figures/territory.svg | 6 +++--- 4 files changed, 17 insertions(+), 13 deletions(-) diff --git a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md index 6798c208d8..13644191c0 100644 --- a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md +++ b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md @@ -67,6 +67,10 @@ Pattern → **Strategy** · Pattern-skill → *a Skill that carries a Strategy* Prompt template → **Instructions** (concept) + `.md` + frontmatter (format) · `agentType` → **stream visibility** · Actor → kept out of the primary vocabulary; Agent owns the slot. +**Room → Channel.** W3 folded the rooms / L2-channels dual into one file type on Sep 11. A channel +is declared as `CHANNEL.md` at team scope and is **L2 opinion — a replaceable flow kind, not a new +L1 type** (W3 D1). *Room* is a dead noun and may not appear in a new surface. + **Costs** code still carries old names until the propagation pass runs, so the tree and the vocabulary disagree in the meantime. That gap is expected, not a defect to file. @@ -82,14 +86,13 @@ Answers settled at project altitude, so no later epic reopens them. | Who owns the claim system? | Layer 1, `orchestration` (PD-2) | Same | | Is an Agent decomposable by its consumer? | No (PD-3) | Same | | Does the propagation pass run before the vocabulary locks? | No (PD-4) | Same | +| Is a channel an L1 type, and is *room* a second thing? | No to both — a channel is L2 opinion, a replaceable flow kind (W3 D1), and it replaced room | W3's objective and floor recut, Sep 11 | ## Open -**Rooms have no declaration surface at all, and two documents disagree about it.** W3's -`FIX-1352` describes rooms as "written in code" in one section and reports zero matches for any -room declaration API in another. Verified: `RoomManifest`, `createRoom`, `roomRegistry` and -`defineRoom` return **no matches** across `packages/`. Whether the file convention is therefore -introducing rooms or *replacing* a code surface changes what W3 ships and what its docs claim. +**`org/channels/` is locked open, read by nothing, and owned by no issue.** W3 ships channels at +**team** scope (FIX-1352, done); the `org/` door is declared and unread — the open half of its +D3/D7. It binds W3 and W4, so it is named here rather than left inside one epic. -This is recorded as an ask rather than answered here, because it is a product call about what we -tell people rooms are — and it binds W3 and W4 both. It belongs on whichever epic reaches it first. +*An earlier version asked whether rooms had a declaration surface. The zero it found was the +`room` → `channel` fold (PD-5) working, not a gap.* diff --git a/spec/_projects/workforce-layer-2-abstraction/PLAN.md b/spec/_projects/workforce-layer-2-abstraction/PLAN.md index 0a479ed1ea..200334cae0 100644 --- a/spec/_projects/workforce-layer-2-abstraction/PLAN.md +++ b/spec/_projects/workforce-layer-2-abstraction/PLAN.md @@ -31,8 +31,9 @@ delay to the others, it is that the client door stays shut. Starting W4 early buys nothing and risks renaming twice (PD-4). - **W3's lab is the proof for PR-3**, so any epic that ships a convention before the lab exists is asserting the rule rather than checking it. -- **The rooms question in [Decisions](DECISIONS.md) → Open binds W3 and W4 both.** It should be - answered on whichever reaches it first, and recorded there, not answered twice. +- **The `org/channels/` door in [Decisions](DECISIONS.md) → Open binds W3 and W4 both.** It is + declared in the tree and read by nothing, and no issue owns closing it. Whichever epic reaches it + first answers it and records it there, so it is not answered twice. ## What is deliberately not next diff --git a/spec/_projects/workforce-layer-2-abstraction/SPEC.md b/spec/_projects/workforce-layer-2-abstraction/SPEC.md index 82ec59aaec..a1f28e15cc 100644 --- a/spec/_projects/workforce-layer-2-abstraction/SPEC.md +++ b/spec/_projects/workforce-layer-2-abstraction/SPEC.md @@ -11,7 +11,7 @@ epics that establish it. | | | |---|---| -| **Winning when** | Someone stands up a working team of agents from files alone — no TypeScript to declare a room, a resource, or a skill — and Layer 1 stays fully available to anyone who wants past the conventions | +| **Winning when** | Someone stands up a working team of agents from files alone — no TypeScript to declare a channel, a resource, or a skill — and Layer 1 stays fully available to anyone who wants past the conventions | | **The read** | Layer 2 nouns that still require code to declare. Five at the start, and the count only moves when a convention ships with a reader and a proof | | **Now** | 2 epics done · 1 in flight · 2 not started. 65 work issues, of which **32 still sit under no epic** — the vocabulary era this project began in, before the epic structure existed | | **Kill line** | If real apps turn out to want to assemble Layer 2 themselves, this project is mis-shaped rather than unfinished: what changes is that the conventions become examples, not the remaining epic list | diff --git a/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg b/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg index 7855d5e4fd..19e27dc6d7 100644 --- a/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg +++ b/spec/_projects/workforce-layer-2-abstraction/figures/territory.svg @@ -1,5 +1,5 @@ + aria-label="The territory: Workforce owns the Layer 2 conventions — Agent, Persona, Skill, Memory, and as conventions Channel, Strategy, Role, Archetype, Team and Plan. Below the fence is the Layer 1 substrate it assembles but does not own: blocks, scope state, resources, the task board, dispatch, instructions. To the right, outside the project, sit the implementations of Tasks, Skills internals and Memory, which live in their own projects. The fence is one test: is there exactly one of it?"> From 37cead0bac37a841aece02b01c0a05dd488e0300 Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 20 Sep 2026 01:20:37 +0000 Subject: [PATCH 11/19] =?UTF-8?q?project(workforce-layer-2-abstraction):?= =?UTF-8?q?=20W5=20through=20its=20gate=20=E2=80=94=20and=20W3=20and=20W4?= =?UTF-8?q?=20both=20wrapped?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Dispatched as a status refresh for W5's objective gate. The live derivation found two more moves the dispatch did not carry: - W4 (FIX-1407) closed #1905 unmerged at 01:02 UTC with its Linear state still In Development -> done. A Linear-only read would have kept it in flight. - W3 (FIX-1351) completed in Linear at 01:05 UTC with #1718 still open -> done. The mirror case; a PR-only read would have missed it. So the table is 4 done / 1 in flight / 1 not started, and W5 is the only epic building. Arc redrawn: W3 and W4 bars close on Sep 20, W5's opens, now line moves to Sep 20. Dependency graph gets heavy borders on all four done epics. Status only. No cross-epic decisions were written: *decided once* belongs to the wrap Update, and no wrap Update has been dispatched for either epic, so W3's and W4's settled knowledge is owed rather than absent. Recorded as such in SPEC.md. Three surfaces that restated W4's or W3's status were corrected to match and nothing else: the org/channels open item (both binding epics wrapped without closing it), the W3->W4 and W4->W5 plan seams, and the propagation-pass prohibition's rationale in BUSINESS-RULES.md (the ER-14 ship fence went with W4). The two carried opens are untouched: W1 scope vs the countdown hold, and W4's growth past its ratified five. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz --- .../BUSINESS-RULES.md | 7 +- .../DECISIONS.md | 4 +- .../workforce-layer-2-abstraction/PLAN.md | 36 +++++----- .../workforce-layer-2-abstraction/SPEC.md | 65 +++++++++++-------- .../figures/arc.svg | 23 ++++--- 5 files changed, 73 insertions(+), 62 deletions(-) diff --git a/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md index 7df3f16264..2e96d7d048 100644 --- a/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md +++ b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md @@ -20,9 +20,10 @@ is a wish. - Expose the **agent bundle** — Persona, Skills, Memory, Instructions, Strategy — as the public API. An Agent is invoked or assigned work, not decomposed (PD-3). - Run the **propagation pass** before the vocabulary locks (PD-4). Renaming an unratified model - costs the rename twice. W4 going in flight does not start it: the pass rides FIX-1385, a ship - ticket, and W4's ship fence holds every W4 ship PR until every W3 child that carries an - implementation is merged to main (W4 ER-14). + costs the rename twice. The pass rode FIX-1385 behind W4's ship fence (W4 ER-14); **W4 wrapped on + Sep 20 and the fence went with it**, so the prohibition now stands on PD-4 alone with no epic + holding it. Whether the pass ran, and who owns it next, is W4's owed wrap Update — not settled + here. - Put a concept here **because it is convenient**. PR-1 is a test, not a preference. --- diff --git a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md index aea3666655..8ead1b5447 100644 --- a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md +++ b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md @@ -94,7 +94,9 @@ Answers settled at project altitude, so no later epic reopens them. **`org/channels/` is locked open, read by nothing, and owned by no issue.** W3 ships channels at **team** scope (FIX-1352, done); the `org/` door is declared and unread — the open half of its -D3/D7. It binds W3 and W4, so it is named here rather than left inside one epic. +D3/D7. It bound W3 and W4, which is why it was named here rather than left inside one epic — +and **both epics wrapped on Sep 20 without closing it**. It now belongs to whichever epic next +touches the channel surface, and nothing currently does. *An earlier version asked whether rooms had a declaration surface. The zero it found was the `room` → `channel` fold (PD-5) working, not a gap.* diff --git a/spec/_projects/workforce-layer-2-abstraction/PLAN.md b/spec/_projects/workforce-layer-2-abstraction/PLAN.md index cacd41eeb4..c248e8fb36 100644 --- a/spec/_projects/workforce-layer-2-abstraction/PLAN.md +++ b/spec/_projects/workforce-layer-2-abstraction/PLAN.md @@ -7,10 +7,10 @@ deliberately not next. Each epic's own plan owns its checks. ![The arc](figures/arc.svg) -Twelve days, not three months. The project opened in June, but every epic here was filed from -**Sep 8** onward; the vocabulary era before it left 45 issues under no epic. Until this week only -one epic had ever been in flight at a time. W3 and W4 were the first overlap; **W5, filed Sep 19, -makes three epics live at once** — two building and one at its gate. +Thirteen days, not three months. The project opened in June, but every epic here was filed from +**Sep 8** onward; the vocabulary era before it left 47 issues under no epic. The overlap this +project ran all week collapsed on Sep 20: **W4 and W3 wrapped three minutes apart**, and W5 is the +only bar crossing the now line. Four lanes closed, one open, one that never opened. ## What each epic consumes and releases @@ -31,21 +31,21 @@ issue says only *Todo* ([Decisions](DECISIONS.md) → Open). ## Coordination seams -- **W3 → W4 — the deferral ended on Sep 19; PD-4's concern did not.** This plan said starting W4 - early bought nothing and risked renaming twice. The objective gate went the other way, so the - reasoning was not overturned — it was **re-aimed from the start to the ship**. W4's D2 lets - filing, specs and POCs run now and fences every *ship* PR until every W3 child that carries an - implementation is merged to main (W4 ER-14); the PR-5 propagation pass rides FIX-1385 (W4 ER-19), - which is a ship ticket and therefore behind that fence. The rename-twice risk is held by a - mechanism, not retired. -- **W4 → W5 is soft, and only on the ship half.** W5 explores and specs now; it waits on W4's first - cut before it ships, because a human in a seat is routed to the same way an agent is, and routing - is W4's. Sanctioned-now versus held is per child: FIX-1458 runs, FIX-1455 is held. +- **W3 → W4 — both ends of this seam wrapped, and it closed unexamined.** This plan said starting + W4 early risked renaming twice; the Sep 19 gate re-aimed that concern **from the start to the + ship**, holding it in W4's ER-14 ship fence rather than retiring it. Whether the fence discharged + or simply stopped applying is a **wrap question**, and no wrap has been folded here — so the seam + is recorded open, not resolved. +- **W4 → W5 was soft, and only on the ship half — and its upstream is now done.** W5 explores and + specs now; its ship half waited on W4's first cut, because a human in a seat is routed to the + same way an agent is, and routing was W4's. W4 has wrapped, so what was a hold is now W5's own + sequencing call. Sanctioned-now versus held is still per child: FIX-1458 runs, FIX-1455 is held. - **W3's lab is the proof for PR-3**, so any epic that ships a convention before the lab exists is - asserting the rule rather than checking it. -- **The `org/channels/` door in [Decisions](DECISIONS.md) → Open binds W3 and W4 both.** It is - declared in the tree and read by nothing, and no issue owns closing it. Whichever epic reaches it - first answers it and records it there, so it is not answered twice. + asserting the rule rather than checking it. The lab shipped with W3 (FIX-1355, done). +- **The `org/channels/` door in [Decisions](DECISIONS.md) → Open outlived both epics that bound + it.** It is declared in the tree and read by nothing, and no issue owns closing it. The plan was + that whichever of W3 or W4 reached it first would answer it; **both wrapped on Sep 20 and neither + did**, so it is now an unowned door rather than a race — which is what the Open entry says. ## What is deliberately not next diff --git a/spec/_projects/workforce-layer-2-abstraction/SPEC.md b/spec/_projects/workforce-layer-2-abstraction/SPEC.md index 0c9f508bab..cf8a0f0fd5 100644 --- a/spec/_projects/workforce-layer-2-abstraction/SPEC.md +++ b/spec/_projects/workforce-layer-2-abstraction/SPEC.md @@ -13,7 +13,7 @@ epics that establish it. |---|---| | **Winning when** | Someone stands up a working team of agents from files alone — no TypeScript to declare a channel, a resource, or a skill — and Layer 1 stays fully available to anyone who wants past the conventions | | **The read** | Layer 2 nouns that still require code to declare. Five at the start, and the count only moves when a convention ships with a reader and a proof | -| **Now** | 2 epics done · **2 in flight** · **1 at its objective gate** · 1 not started. W5 was filed Sep 19 as a sibling of W4, not a child of it. 101 work issues, of which **45 sit under no epic** — the vocabulary era this project began in | +| **Now** | **4 epics done** · 1 in flight · 1 not started. W4 and W3 wrapped three minutes apart on Sep 20. 101 work issues, of which **47 sit under no epic** — the vocabulary era this project began in | | **Kill line** | If real apps turn out to want to assemble Layer 2 themselves, this project is mis-shaped rather than unfinished: what changes is that the conventions become examples, not the remaining epic list | ![The territory](figures/territory.svg) @@ -22,39 +22,45 @@ Above the fence is what this project owns; below it is the substrate it assemble own. The fence is one question — *is there exactly one of it?* — and it is the test every epic under this project is checked against. -## The epics — derived live 2026-09-19 23:46 UTC +## The epics — derived live 2026-09-20 01:30 UTC | Epic | What it owns | State | Issues | |---|---|---|---| | [FIX-1332](https://linear.app/fixpoint-labs/issue/FIX-1332) · **W2 foundation** | Conventions, seat factory, thin Agent | **done** | 5/5 | | [FIX-1359](https://linear.app/fixpoint-labs/issue/FIX-1359) · **default agent flow** | OOTB replaceable `agent` flow kind | **done** | 7/11 | -| [FIX-1351](https://linear.app/fixpoint-labs/issue/FIX-1351) · **W3 file surface** | Channels, resources, skills + a thin pentest lab | **in flight** | 19/22 | -| [FIX-1407](https://linear.app/fixpoint-labs/issue/FIX-1407) · **W4 routing** | Work routing & package cohesion | **in flight** | 4/9 | -| [FIX-1457](https://linear.app/fixpoint-labs/issue/FIX-1457) · **W5 living workforce** | Humans in seats, and reference apps that run on the surface | *at its objective gate* | 0/2 | +| [FIX-1351](https://linear.app/fixpoint-labs/issue/FIX-1351) · **W3 file surface** | Channels, resources, skills + a thin pentest lab | **done** | 19/20 | +| [FIX-1407](https://linear.app/fixpoint-labs/issue/FIX-1407) · **W4 routing** | Work routing & package cohesion | **done** | 5/9 | +| [FIX-1457](https://linear.app/fixpoint-labs/issue/FIX-1457) · **W5 living workforce** | Humans in seats, and reference apps that run on the surface | **in flight** | 0/2 | | [FIX-1333](https://linear.app/fixpoint-labs/issue/FIX-1333) · **W1 MCP door** | Client door over intake, boards, dispatcher | *not started* — **scope disputed** | — | -2 done · 2 in flight · 1 at its gate · 1 not started. Every number above is re-derived from Linear -and the epic PRs each refresh; none is carried forward. +4 done · 1 in flight · 1 not started. Every number above is re-derived from Linear and the epic PRs +each refresh; none is carried forward. -**The epic PRs are the other half of this table**, because an epic wraps by closing its PR unmerged -and leaves its Linear state alone. W2 and the agent flow read *done* that way +**This refresh is the case the two-source rule exists for.** W4 reads **done** from +[#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905) — closed unmerged 01:02 UTC — +while its Linear issue still says *In Development*, so a Linear-only read would have called it in +flight. W3 is the mirror: Linear completed it at 01:05 and left +[#1718](https://github.com/fixpoint-labs/flow-state-dev/pull/1718) open, which a PR-only read would +have missed. W2 and the agent flow wrapped the ordinary way ([#1664](https://github.com/fixpoint-labs/flow-state-dev/pull/1664), -[#1730](https://github.com/fixpoint-labs/flow-state-dev/pull/1730), closed unmerged); W3 is *in -flight* on [#1718](https://github.com/fixpoint-labs/flow-state-dev/pull/1718), W4 on -[#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905). **W5 has no PR cell yet** — -`epic/living-workforce` carried none at this read, so the branch is the handle until one opens. - -**W5 is a sibling of W4, not a stage inside it.** Filed Sep 19, at its objective gate: humans -occupying seats the same way agents do, and reference apps that run on the file surface rather than -demonstrate it. Two children — FIX-1458 (explore, runnable now) and FIX-1455, a nested child epic -for the kitchen-sink rebuild, held. Exploring and speccing are sanctioned now; **shipping is -soft-after W4's first cut** ([Plan](PLAN.md)). - -**W4's count moved because the epic grew, not because its set size reopened.** The five the owner -ratified on Sep 19 stand at 3 of 5. Four children have been attached since: FIX-1451 (done), -FIX-1381 (in spec review), FIX-1460 and FIX-1461 (backlog) — so the live derivation returns -**4/9**. Whether the newcomers sit inside W4's exit gate is -[#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905)'s call, not this table's. +[#1730](https://github.com/fixpoint-labs/flow-state-dev/pull/1730)). W5's cell, empty last refresh, +resolves to [#1944](https://github.com/fixpoint-labs/flow-state-dev/pull/1944). + +**Neither wrap has been folded at this altitude.** A wrap is when an epic settles cross-epic +knowledge, and recording it belongs to the `update` action, not to a status refresh. W3's and W4's +*decided once* rows are **owed, not absent**: the status above is current, the decisions behind it +are not yet here. + +**W5 is now the only epic building.** [#1944](https://github.com/fixpoint-labs/flow-state-dev/pull/1944) +carries `spec approved` and one folded review round. Neither child has started — FIX-1458 (explore) +and FIX-1455, a nested child epic that runs its own lifecycle and files no spec PR here (W5's +ER-15). Its ship half waited on W4's first cut, and **W4 has wrapped**, so that fence is W5's +sequencing call rather than a standing hold ([Plan](PLAN.md)). + +**W4 wrapped at 5 of 9 — growth, not a reopened set size.** The ratified five stand at 4 of 5 +(FIX-1405 in review, FIX-1430 spec-approved); FIX-1451 and FIX-1381 (done), FIX-1460 and FIX-1461 +(backlog) attached after the gate. Whether the newcomers sat inside W4's exit gate was #1905's call +and it closed without saying; not resolved here. **W1's status comes from the issue; a second surface disagrees.** FIX-1333 has sat in **Todo** since Sep 8 — never moved, nothing blocking it, priority **High**. The project's @@ -76,11 +82,14 @@ flowchart LR W3 --> W5["FIX-1457 · W5 living workforce"] W4 -.->|"first cut before W5 ships"| W5 W2 --> W1["FIX-1333 · W1 MCP door"] + classDef done stroke-width:2px + class W2,AF,W3,W4 done ``` -W2's conventions and seat factory are what everything else builds on. The dotted edge is the only -soft one — W5 explores now and waits on W4 only to ship — and W1, depending on no file surface, is -the one epic that can sit unstarted without blocking anything. +Four heavy borders and one live node: everything W5 was sequenced behind has now wrapped. The +dotted edge was the only soft one and it has gone slack — W4's first cut exists, so nothing +upstream is still holding W5's ship half. W1, depending on no file surface, remains the one epic +that can sit unstarted without blocking anything. ## What this project is not diff --git a/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg index 86ccc2b996..94531830d1 100644 --- a/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg +++ b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg @@ -1,5 +1,5 @@ + aria-label="The arc: six epic lanes against thirteen days. W2 ran Sep 8 to Sep 12 and is done. The default agent flow ran Sep 12 to Sep 16 and is done. W3 ran Sep 16 to Sep 20 and is done, with 19 of 20 issues closed. W4 stood up its epic-spec on Sep 18, cleared its objective gate on Sep 19, and wrapped on Sep 20 with five of nine issues done. W5 was filed on Sep 19, passed its objective gate on Sep 20, and is the only bar crossing the now line. W1 was created Sep 8 and has not started."> From 5f6532f93b626b26c8c9d55495f5b68fba225b2c Mon Sep 17 00:00:00 2001 From: Claude Date: Sun, 20 Sep 2026 02:05:25 +0000 Subject: [PATCH 12/19] =?UTF-8?q?project(workforce-layer-2-abstraction):?= =?UTF-8?q?=20fold=20both=20wraps=20=E2=80=94=20W3=20and=20W4?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The wrap Update both epics were owed. W4 closed #1905 unmerged and W3 completed in Linear on Sep 20; status was already current, the decisions behind it were not. Into *decided once* (eight rows): ER-20's PASS on the model path with both negative controls red; drain width 1, reversible on the knob; a board assignee is not a seat (W4 D3); dispatch's parentSessionId / opt-in history / brief payload, with the parent corrected to the draining seat's session (D4); runtime inventory as two layers with readDeclaredRoster public (D5); the ratified package format — Markdown, instructions + tools, not disk-only, documents org-scoped, building outside W4 as FIX-1459; the admission contract at four imposed keys, not three; and teams// as an address rather than a fence. Kept OUT of *decided once*, in a new section that records status rather than upgrading it: the ER-14 narrowing (recommended, never answered, run meanwhile as an operating default read off the owner's merges — and now expired unruled, since every W3 child is terminal and FIX-1435, the issue W4's re-gate named as the W3 child holding the fence, has no parent in Linear at all); ER-7's re-confirm, unrecorded though the key shipped; the FIX-1368 D2 arm, settled by what shipped rather than by a ruling; the three items W4 named out with no owner; and the "grown past its ratified set" orphan that #1905 closed without answering. Corrected in Open: org/channels/ is not merely unowned — FIX-1419 was filed for it and reads Done, while main still walks teams/ only and a committed test pins that silence. Status re-derived from Linear and the epic PRs: W4 7/9 (Linear In Review, PR closed unmerged), two strays deliberately deferred; W3 19/20 + 1 duplicate, #1718 left open, which is not this artifact's to close. W5's ship fence has lifted. Arc's W4 count 5/9 -> 7/9, re-rendered and looked at in both themes. Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz --- .../BUSINESS-RULES.md | 12 +-- .../DECISIONS.md | 76 +++++++++++++++++-- .../workforce-layer-2-abstraction/PLAN.md | 37 +++++---- .../workforce-layer-2-abstraction/SPEC.md | 50 ++++++------ .../figures/arc.svg | 4 +- 5 files changed, 126 insertions(+), 53 deletions(-) diff --git a/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md index 2e96d7d048..d2d31f6ac3 100644 --- a/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md +++ b/spec/_projects/workforce-layer-2-abstraction/BUSINESS-RULES.md @@ -12,7 +12,8 @@ is a wish. | **PR-2** | When a convention needs behaviour, then it **composes Layer 1 primitives** and never introduces a parallel primitive — Layer 2 stays transparently implemented, no black box | FIX-1332 | Package-boundary review | | **PR-3** | When a convention ships, then it ships with a **reader and a consumer that exercises it**, not a format alone | FIX-1351 | The thin pentest lab | | **PR-4** | When a file declares a thing, then every key the convention **derives** is refused by name — a file may not say where it lives, what its identity is, or what its state shape is | FIX-1351 | Reader tests, per-convention | -| **PR-5** | When a new surface names a Layer 2 concept, then it uses the **settled name** (Role, Strategy, Instructions, stream visibility), never the superseded one | FIX-1407 | The propagation pass, carried by FIX-1385 (W4 ER-19) | +| **PR-5** | When a new surface names a Layer 2 concept, then it uses the **settled name** (Role, Strategy, Instructions, stream visibility), never the superseded one | FIX-1407 | A check over the changing epic's own diff. W4 met it that way (ER-19); the repo-wide pass is **unowned** | +| **PR-6** | When an epic writes a hireable worker kind, then it **composes `workerConfigSchema()`** — the four imposed keys `instructions`, `teamInstructions`, `seatSkills`, `seatTools` — and a file that authors any of them is refused by name | FIX-1351 | The closed schema at every mint; a kind that cannot take the bag refuses at boot, naming the key | **No epic may:** @@ -20,10 +21,11 @@ is a wish. - Expose the **agent bundle** — Persona, Skills, Memory, Instructions, Strategy — as the public API. An Agent is invoked or assigned work, not decomposed (PD-3). - Run the **propagation pass** before the vocabulary locks (PD-4). Renaming an unratified model - costs the rename twice. The pass rode FIX-1385 behind W4's ship fence (W4 ER-14); **W4 wrapped on - Sep 20 and the fence went with it**, so the prohibition now stands on PD-4 alone with no epic - holding it. Whether the pass ran, and who owns it next, is W4's owed wrap Update — not settled - here. + costs the rename twice. **W4's wrap is folded and the answer is that the pass did not run:** what + FIX-1385 owed was a name check over W4's own diff, ER-19 was narrowed to exactly that, and the + repo-wide pass was **named out at the wrap with no owner** + ([Decisions](DECISIONS.md) → *Recorded at the wrap*). The prohibition now stands on PD-4 alone, + with no epic holding it and nobody scheduled to run it. - Put a concept here **because it is convenient**. PR-1 is a test, not a preference. --- diff --git a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md index 8ead1b5447..1d4cef72e0 100644 --- a/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md +++ b/spec/_projects/workforce-layer-2-abstraction/DECISIONS.md @@ -87,16 +87,78 @@ Answers settled at project altitude, so no later epic reopens them. | Is an Agent decomposable by its consumer? | No (PD-3) | Same | | Does the propagation pass run before the vocabulary locks? | No (PD-4) | Same | | Is a channel an L1 type, and is *room* a second thing? | No to both — a channel is L2 opinion, a replaceable flow kind (W3 D1), and it replaced room | W3's objective and floor recut, Sep 11 | -| Does W4 wait for W3's vocabulary lock before it starts? | No — it starts; only its **ship** PRs wait. PD-4's rename-twice risk is held by W4's ship fence, not by keeping the epic shut | W4's objective gate, Sep 19 | -| How many children does W4 carry? | Five — FIX-1394, FIX-1405, FIX-1385, FIX-1408, FIX-1430. FIX-817 and FIX-1415 are related-not-child | [#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905) sign-off 2, approved Sep 19 02:49 ET; re-homed in Linear 03:04 UTC | +| Does W4 wait for W3's vocabulary lock before it starts? | No — it starts; only its **ship** PRs wait. PD-4's rename-twice risk was held by W4's ship fence, not by keeping the epic shut. *(That fence's own fate is below, 1 — it expired unruled)* | W4's objective gate, Sep 19 | +| How many children did W4's gate ratify? | **Five** — FIX-1394, FIX-1405, FIX-1385, FIX-1408, FIX-1430; FIX-817 and FIX-1415 related-not-child. It **wrapped carrying seven**: FIX-1381 by the owner's word, FIX-1451 by a tracker correction. Whether that growth stayed inside the approved gate is *unresolved* (below, 5) | [#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905) sign-off 2, approved Sep 19 02:49 ET | +| Does routing work — a coordinator assigning a queue across seats their own files name? | **Yes, proved on the real path.** No later epic re-proves it | W4's ER-20 **PASS**: three runs of `goals/manager-queue-lab/it-routes-a-queue-to-the-seats-their-files-name`, `GOAL_FILER=model` on `vercel/openai/gpt-5.4-mini`, `e6ccb22` of `main`, **both negative controls red on that same path** | +| Does a busy seat take a second row concurrently? | **No — drain width is 1.** Work queues behind it. **Reversible:** the `MANAGER_QUEUE_DRAIN_WIDTH` knob stays | W4 ER-15, on the lab running at *both* widths rather than a number being picked | +| Is a board assignee a Workforce seat? | **No.** An assignee is a key on a ledger; a seat is a roster slot that mints a flow instance. They map **by composition** — no second `WorkerRegistry`, no assignable-channel routing | W4 D3 | +| How does work reach a child, and with what content? | Dispatch passes `parentSessionId`, bound **for life at mint**; parent history only by **opt-in tools**, never an ambient dump; the payload is the brief (`goal` / `constraints` / `acceptance` / `links`). **Linked is not nested** | W4 D4. Mechanism corrected by FIX-1430 reading the code: the parent bound is the **draining seat's** session, not the filer's | +| Is runtime inventory a helper or a resource? | **Two layers, and the fork is dismissed.** The declared roster composed at read time — public as `readDeclaredRoster(root)` — and the live org-scoped resource ChannelFlow updates as seats and channels open. They answer different questions | W4 D5; the export approved at the objective gate | +| What is a seat's package, and where do documents live? | One format, **authored in Markdown**, scoped to instructions and tools, two attachment modes (seat always-on, library opt-in), and **not disk-only** — packages will eventually be LLM-authored and stored as a resource, which binds the design now. **Documents are org-scoped**: v1 carries no per-seat document slot, and the exclusion is structural — a package attaches to one seat, a document installs at org scope | W4 ER-2, the owner's ratify Sep 19 · builds outside W4 as **FIX-1459** | +| What does a hireable worker kind have to accept? | **Four imposed keys, not three** — `instructions`, `teamInstructions`, `seatSkills`, `seatTools`. Compose `workerConfigSchema()`; a file that authors one is refused by name (PR-6) | Read off `packages/workforce/src/worker-config.ts` on `main`; `teamInstructions` shipped with FIX-1377 ([#1911](https://github.com/fixpoint-labs/flow-state-dev/pull/1911)) | +| Does `teams//` fence what a seat reads, or only address it? | **It addresses; it does not fence**, and the halves differ mechanically. Team *instructions* reach only that team's seats — they ride each seat's own config. Team *documents* install on a worker **kind**, so every seat of that kind reaches every team's | FIX-1377 shipped as written, FIX-1368 done, and the docs teach both rules out loud. **Settled by what shipped, not by a recorded ruling** — below | + +## Recorded at the wrap, not decided + +W3 and W4 both wrapped on Sep 20. These five came out of those wraps **with a status that is not +*settled***, and they keep it. Writing any into *decided once* would manufacture a ratification +nobody gave. + +**1 · The ER-14 ship-fence narrowing was recommended, never answered — and has now expired +unruled.** W4's ER-14 fenced its ship PRs behind W3's implementing children. The epic recommended +narrowing that to *per-PR file overlap*; the owner never answered; the epic meanwhile ran an +**operating default inferred from his merge actions** — he merged #1928 through a live fence — +rather than from anything he said, and said so itself. Two things changed at the wrap: every W3 +child is terminal, and **FIX-1435, the issue W4's re-gate named as the W3 child keeping the fence +live, has no parent in Linear at all**. On ER-14's own letter the release condition now reads +**satisfied**. Nobody ruled it discharged and nobody answered the narrowing; it stopped mattering +instead. An epic that wants a fence against another epic states its release condition in **merged +code**, not a package name, and inherits nothing from this one. + +**2 · ER-7's third framework-owned key shipped without its re-confirm being recorded.** W3 held +`teamInstructions` behind an owner re-confirm, because adding it changed a spec he had already +approved. The key is now on `main` and the contract carries a **fourth**, `seatTools`, besides. The +contract is settled and is in *decided once*; the re-confirm W3 said no child would ship without is +recorded nowhere I can find. Stated here so nobody later reads the shipped code as evidence the +gate was met. + +**3 · The `teams//` address-vs-fence fork was settled by what shipped, not by a ruling.** W3's +Open 6 was bound to FIX-1368's D2 — three live arms, two of three reviewers backing *address*. +FIX-1377 shipped as written, FIX-1368 is done, the published docs teach the address reading, and +**no artifact records the owner ruling on D2 by name**. The outcome matching the recommendation is +not the recommendation being accepted. Making `teams//` a real visibility boundary later is a +re-open of a shipped surface, not a correction. + +**4 · Three things W4 named out at its wrap have no owner, and closing W4 gave none of them one.** + +| Named out | State | +|---|---| +| **ER-19's repo-wide PR-5 rename** | Only the check over W4's own diff ran. The repo-wide pass is the project's, still gated on PD-4's vocabulary lock, and **has no owner** | +| **How a sub-agent's background work surfaces on a board row** | Orphaned when FIX-1385 shipped without a coordinator assign. Whoever takes it is bound by W4's ER-11 — a view, never a new status. **Unowned** | +| **A board id no channel minted** | Not refused: it silently resolves a second, empty ledger, with only `warnUnattendedBoards` as a signal. **No owner, no filed ticket** | + +**5 · "W4 has grown past its ratified set" was routed to #1905, which closed without answering it.** +The set went from the ratified five to seven — one by the owner's word, one by a tracker correction +— and whether the newcomers sat inside W4's approved exit gate was #1905's to say. It closed +silent. Re-homed here because its surface is gone; **still unresolved**, and it is the question a +later epic should ask before letting its own set grow after a gate. ## Open -**`org/channels/` is locked open, read by nothing, and owned by no issue.** W3 ships channels at -**team** scope (FIX-1352, done); the `org/` door is declared and unread — the open half of its -D3/D7. It bound W3 and W4, which is why it was named here rather than left inside one epic — -and **both epics wrapped on Sep 20 without closing it**. It now belongs to whichever epic next -touches the channel surface, and nothing currently does. +**`org/channels/` is still unbuilt — and the issue that owned it reads Done.** W3 ships channels at +**team** scope (FIX-1352, done); D3 locks `org/` and `teams//` as the same slots, and the +`org/` half of the channels reader was never built. It bound W3 and W4, which is why it is named +here rather than left inside one epic. + +**What changed at the wrap makes this worse, not better.** FIX-1419 was filed for exactly this gap +— its description says *"the owner has decided to close the gap rather than continue carrying it as +a documented limitation"* — and it reads **Done**, closed 2026-09-18 in FIX-1421's batch. The code +disagrees: on `main`, `readChannelsDirectory` walks `teams//channels/` only, its own header +says so, `packages/workforce/README.md` says *"There is no `org/channels/` level"*, the atlas lists +it under **not shipped**, and a committed test plants an `org/channels/` file and **pins the +silence deliberately**. A declared, unread door carrying a closed ticket is worse than an unowned +one, because the next reader takes Done at its word. Whoever next touches the channel surface +builds the reader or reopens FIX-1419; nothing currently does either. *An earlier version asked whether rooms had a declaration surface. The zero it found was the `room` → `channel` fold (PD-5) working, not a gap.* diff --git a/spec/_projects/workforce-layer-2-abstraction/PLAN.md b/spec/_projects/workforce-layer-2-abstraction/PLAN.md index c248e8fb36..9b8dadd524 100644 --- a/spec/_projects/workforce-layer-2-abstraction/PLAN.md +++ b/spec/_projects/workforce-layer-2-abstraction/PLAN.md @@ -20,7 +20,7 @@ only bar crossing the now line. Four lanes closed, one open, one that never open | **agent flow** · FIX-1359 | W2's thin Agent | A replaceable OOTB `agent` flow kind, so an app overrides rather than assembles | | **W3** · FIX-1351 | W2's walk primitives; the agent flow as a consumer | The file surface — channels, resources, skills — and the pentest lab that proves PR-3 | | **W1** · FIX-1333 | W2's conventions; Layer 1 boards + dispatcher | The MCP client door over intake, boards and dispatch | -| **W4** · FIX-1407 | W3's file surface; the settled vocabulary | Work routing, package cohesion, and the PR-5 propagation pass | +| **W4** · FIX-1407 | W3's file surface; the settled vocabulary | Work routing, proved on the real path (ER-20), and the **ratified** package format — whose build left the set as FIX-1459. **Not** the PR-5 propagation pass, which it named out unowned | | **W5** · FIX-1457 | W3's file surface; W4's routing, for its ship half only | Humans in seats on the same surface as agents, and reference apps that run on it | **W1 is the one that unblocks nothing**, which is why it can sit unstarted at the bottom of the arc @@ -31,25 +31,32 @@ issue says only *Todo* ([Decisions](DECISIONS.md) → Open). ## Coordination seams -- **W3 → W4 — both ends of this seam wrapped, and it closed unexamined.** This plan said starting - W4 early risked renaming twice; the Sep 19 gate re-aimed that concern **from the start to the - ship**, holding it in W4's ER-14 ship fence rather than retiring it. Whether the fence discharged - or simply stopped applying is a **wrap question**, and no wrap has been folded here — so the seam - is recorded open, not resolved. -- **W4 → W5 was soft, and only on the ship half — and its upstream is now done.** W5 explores and - specs now; its ship half waited on W4's first cut, because a human in a seat is routed to the - same way an agent is, and routing was W4's. W4 has wrapped, so what was a hold is now W5's own - sequencing call. Sanctioned-now versus held is still per child: FIX-1458 runs, FIX-1455 is held. +- **W3 → W4 — both ends wrapped, and the fence stopped applying rather than being discharged.** + This plan said starting W4 early risked renaming twice; the Sep 19 gate re-aimed that concern + **from the start to the ship**, holding it in W4's ER-14 ship fence. The wrap is now folded and + the answer is untidy: the narrowing the epic recommended was never answered, the epic ran an + unratified default read off the owner's merges, and ER-14's release condition now reads + **satisfied** — every W3 child is terminal, and the one issue W4 named as holding the fence turns + out not to be a W3 child at all. Nobody ruled it discharged + ([Decisions](DECISIONS.md) → *Recorded at the wrap*, 1). +- **W4 → W5 — the fence has lifted.** W5's ship half was soft-after W4's first cut, because a human + in a seat is routed to the same way an agent is, and routing was W4's. **That cut exists and W4 + has wrapped, so the condition is met** and W5's remaining sequencing is its own. Sanctioned-now + versus held is still per child: FIX-1458 runs, FIX-1455 is held. - **W3's lab is the proof for PR-3**, so any epic that ships a convention before the lab exists is asserting the rule rather than checking it. The lab shipped with W3 (FIX-1355, done). -- **The `org/channels/` door in [Decisions](DECISIONS.md) → Open outlived both epics that bound - it.** It is declared in the tree and read by nothing, and no issue owns closing it. The plan was - that whichever of W3 or W4 reached it first would answer it; **both wrapped on Sep 20 and neither - did**, so it is now an unowned door rather than a race — which is what the Open entry says. +- **The `org/channels/` door outlived both epics that bound it, and now reads closed while being + open.** The plan was that whichever of W3 or W4 reached it first would answer it; both wrapped on + Sep 20 and neither did. Worse, the issue filed for it (FIX-1419) reads **Done** while `main` + still walks `teams/` only and a committed test pins that silence + ([Decisions](DECISIONS.md) → Open). The next epic to touch the channel surface inherits a gap + the tracker says is shut. ## What is deliberately not next -- **The propagation pass.** Gated on the vocabulary locking (PD-4), not on capacity. +- **The propagation pass.** Gated on the vocabulary locking (PD-4), not on capacity — and now + **unowned**: W4 met PR-5 with a check over its own diff and named the repo-wide pass out at its + wrap. - **Tasks, Skills internals, Memory implementation.** Own projects; this one owns the vocabulary. - **Re-parenting the 45 unparented issues.** They are the pre-epic era. Sweeping them under epics retroactively would make the arc look tidier and tell you less about what actually happened. diff --git a/spec/_projects/workforce-layer-2-abstraction/SPEC.md b/spec/_projects/workforce-layer-2-abstraction/SPEC.md index cf8a0f0fd5..2f8c41c435 100644 --- a/spec/_projects/workforce-layer-2-abstraction/SPEC.md +++ b/spec/_projects/workforce-layer-2-abstraction/SPEC.md @@ -13,7 +13,7 @@ epics that establish it. |---|---| | **Winning when** | Someone stands up a working team of agents from files alone — no TypeScript to declare a channel, a resource, or a skill — and Layer 1 stays fully available to anyone who wants past the conventions | | **The read** | Layer 2 nouns that still require code to declare. Five at the start, and the count only moves when a convention ships with a reader and a proof | -| **Now** | **4 epics done** · 1 in flight · 1 not started. W4 and W3 wrapped three minutes apart on Sep 20. 101 work issues, of which **47 sit under no epic** — the vocabulary era this project began in | +| **Now** | **4 epics done** · 1 in flight · 1 not started. W4 and W3 wrapped three minutes apart on Sep 20, and **both wraps are folded** — what they settled is in [Decisions](DECISIONS.md), and what they left unowned is recorded there with that status attached. 101 work issues, of which **47 sit under no epic** — the vocabulary era this project began in | | **Kill line** | If real apps turn out to want to assemble Layer 2 themselves, this project is mis-shaped rather than unfinished: what changes is that the conventions become examples, not the remaining epic list | ![The territory](figures/territory.svg) @@ -22,45 +22,47 @@ Above the fence is what this project owns; below it is the substrate it assemble own. The fence is one question — *is there exactly one of it?* — and it is the test every epic under this project is checked against. -## The epics — derived live 2026-09-20 01:30 UTC +## The epics — derived live 2026-09-20 02:00 UTC | Epic | What it owns | State | Issues | |---|---|---|---| | [FIX-1332](https://linear.app/fixpoint-labs/issue/FIX-1332) · **W2 foundation** | Conventions, seat factory, thin Agent | **done** | 5/5 | | [FIX-1359](https://linear.app/fixpoint-labs/issue/FIX-1359) · **default agent flow** | OOTB replaceable `agent` flow kind | **done** | 7/11 | | [FIX-1351](https://linear.app/fixpoint-labs/issue/FIX-1351) · **W3 file surface** | Channels, resources, skills + a thin pentest lab | **done** | 19/20 | -| [FIX-1407](https://linear.app/fixpoint-labs/issue/FIX-1407) · **W4 routing** | Work routing & package cohesion | **done** | 5/9 | +| [FIX-1407](https://linear.app/fixpoint-labs/issue/FIX-1407) · **W4 routing** | Work routing & package cohesion | **done** | 7/9 | | [FIX-1457](https://linear.app/fixpoint-labs/issue/FIX-1457) · **W5 living workforce** | Humans in seats, and reference apps that run on the surface | **in flight** | 0/2 | | [FIX-1333](https://linear.app/fixpoint-labs/issue/FIX-1333) · **W1 MCP door** | Client door over intake, boards, dispatcher | *not started* — **scope disputed** | — | 4 done · 1 in flight · 1 not started. Every number above is re-derived from Linear and the epic PRs each refresh; none is carried forward. -**This refresh is the case the two-source rule exists for.** W4 reads **done** from -[#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905) — closed unmerged 01:02 UTC — -while its Linear issue still says *In Development*, so a Linear-only read would have called it in -flight. W3 is the mirror: Linear completed it at 01:05 and left -[#1718](https://github.com/fixpoint-labs/flow-state-dev/pull/1718) open, which a PR-only read would -have missed. W2 and the agent flow wrapped the ordinary way -([#1664](https://github.com/fixpoint-labs/flow-state-dev/pull/1664), -[#1730](https://github.com/fixpoint-labs/flow-state-dev/pull/1730)). W5's cell, empty last refresh, -resolves to [#1944](https://github.com/fixpoint-labs/flow-state-dev/pull/1944). - -**Neither wrap has been folded at this altitude.** A wrap is when an epic settles cross-epic -knowledge, and recording it belongs to the `update` action, not to a status refresh. W3's and W4's -*decided once* rows are **owed, not absent**: the status above is current, the decisions behind it -are not yet here. +**This is the case the two-source rule exists for, and each half caught one epic.** W4 reads +**done** from [#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905), closed unmerged +01:02 UTC, while Linear says *In Review*. W3 is the mirror: Linear completed it at 01:05 and left +[#1718](https://github.com/fixpoint-labs/flow-state-dev/pull/1718) **open** — the reverse of the +wrap convention, recorded rather than corrected, because that PR is not this artifact's to close. +W2 and the agent flow wrapped the ordinary way; W5 is +[#1944](https://github.com/fixpoint-labs/flow-state-dev/pull/1944). + +**Both wraps are now folded.** What W3 and W4 settled for their siblings is in +[Decisions](DECISIONS.md) → *decided once*. What they **did not** settle is in the section below +it, kept separate on purpose: an unratified operating default, three items named out with no owner, +and a door whose owning issue reads Done while the code does not carry it. A wrap that recorded +those as answers would have manufactured four ratifications nobody gave. **W5 is now the only epic building.** [#1944](https://github.com/fixpoint-labs/flow-state-dev/pull/1944) carries `spec approved` and one folded review round. Neither child has started — FIX-1458 (explore) and FIX-1455, a nested child epic that runs its own lifecycle and files no spec PR here (W5's -ER-15). Its ship half waited on W4's first cut, and **W4 has wrapped**, so that fence is W5's -sequencing call rather than a standing hold ([Plan](PLAN.md)). - -**W4 wrapped at 5 of 9 — growth, not a reopened set size.** The ratified five stand at 4 of 5 -(FIX-1405 in review, FIX-1430 spec-approved); FIX-1451 and FIX-1381 (done), FIX-1460 and FIX-1461 -(backlog) attached after the gate. Whether the newcomers sat inside W4's exit gate was #1905's call -and it closed without saying; not resolved here. +ER-15). **Its ship fence has lifted.** W5's soft-after condition was W4's first cut for ship; W4's +first cut exists and W4 has wrapped, so the condition is met and what remains is W5's own +sequencing, not a hold on anything upstream ([Plan](PLAN.md)). + +**W4 wrapped at 7 of 9 — growth, not a reopened set size.** The ratified five are all done; +FIX-1451 and FIX-1381 attached after the gate and are done too. What is left is two strays in +backlog — FIX-1460 (the fate of `unionAllowedTools`) and FIX-1461 (documenting +`registerCatalogTools`) — which the owner has said not to worry about, so they are **deliberately +deferred, not pending**. Whether the newcomers sat inside W4's exit gate was #1905's call and it +closed without saying; that orphan is carried in [Decisions](DECISIONS.md), still unresolved. **W1's status comes from the issue; a second surface disagrees.** FIX-1333 has sat in **Todo** since Sep 8 — never moved, nothing blocking it, priority **High**. The project's diff --git a/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg index 94531830d1..7b5f25e14a 100644 --- a/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg +++ b/spec/_projects/workforce-layer-2-abstraction/figures/arc.svg @@ -1,5 +1,5 @@ + aria-label="The arc: six epic lanes against thirteen days. W2 ran Sep 8 to Sep 12 and is done. The default agent flow ran Sep 12 to Sep 16 and is done. W3 ran Sep 16 to Sep 20 and is done, with 19 of 20 issues closed. W4 stood up its epic-spec on Sep 18, cleared its objective gate on Sep 19, and wrapped on Sep 20 with seven of nine issues done. W5 was filed on Sep 19, passed its objective gate on Sep 20, and is the only bar crossing the now line. W1 was created Sep 8 and has not started.">