Skip to content

project(orchestration-primitives): Orchestration Primitives — the project-spec (never merged) - #3102

Open
jhoffner wants to merge 20 commits into
mainfrom
project/orchestration-primitives
Open

jhoffner wants to merge 20 commits into
mainfrom
project/orchestration-primitives

Conversation

@jhoffner

@jhoffner jhoffner commented Oct 12, 2026 •

Copy link
Copy Markdown
Collaborator

Requested by Jake · project thread

Never merged. The standing project-spec for the Linear project Orchestration Primitives, open for the life of the project. As of 2026-10-27, set at 684fbbc.

Winning when Every pattern that runs on the task board stores its work as Task rows, and work one worker gives another is answered once, whichever path delivers it. A task that stops for a person waits in one place, one Inbox item answered once: landed Oct 27 with FIX-2058. One task primitive (FIX-2056) landed Oct 24
The read Ways to give work: today 1, done since Oct 24. Ways to answer a stop for a person outside the Inbox: today 0, done since Oct 27; answerTask, the question and review parks and the harness inbox's answerQuestion() are gone (PR-3). Patterns off the board (roundRobin, debate) and the screens that ask a person are outside
Now 8 epics done · 0 in flight · 3 filed, not started. FIX-2058 wrapped Oct 27: closure FIX-2134 passed on main 83c444b (#3951). FIX-2070 and FIX-2092 are free to start
Kill line Passed for both halves: FIX-2056's closure (Oct 24) for one primitive, FIX-2058's closure (Oct 27) for one Inbox

The territory: the figure is in SPEC.md.

Above the fence is the work unit and how it moves between workers; everything below it this project uses or refuses, and never builds.

The arc: the figure is in PLAN.md.

The substrate, the dispatch line, one task primitive and one Inbox are done. Messages and phased workflows are free to start; the renames come last.

The epics (live table: SPEC.md) · 8 done · 0 in flight · 3 filed, not started

The standing calls

PD-1 · One work unit: a Task row in a TaskCollection, durable through scope.

PD-1 trade-off: the figure is in DECISIONS.md.

It comes down to hardening: one board is fixed once, per-pattern stores are each fixed alone.

PD-2 · A skill carries instructions, never how its work runs.

PD-2 trade-off: the figure is in DECISIONS.md.

It comes down to how many ways a skill's work runs: a per-skill mode keeps three paths.

PD-3 · Anything that expects an answer is a task; a tell is a plain message. A post that needs a confirmed answer is a task, filed with createTask and waited on with getTask({ id, wait }): claimed once, its ending told to the filer once. A tell is a plain message with no row, deadline or once-only check. FIX-2083 carried it out, and FIX-2087's closure held it on Oct 24.

PD-3 trade-off: the figure is in DECISIONS.md.

It comes down to a post that expects an answer: the ledger-backed message needed a second once-only check to get one.

Decided since the last refresh

From FIX-2058's wrap, binding FIX-2070, FIX-2071 and FIX-2092:

  • New rule PR-3 (BUSINESS-RULES.md): work that stops for a person is one Inbox item, answered once. A waitForUser suspension when code reads the answer next, otherwise a hold on its task row. No epic adds a park or a third answer route.
  • No park fallback for a host without durable execution; such a host refuses a person's suspension loudly (FIX-2058 D1, owner, Oct 23).
  • A person's stop has no default deadline; a host or a call opts in (FIX-2058 D3, owner, Oct 24).
  • Telling a filer its task waits on a person is FIX-2070's (owner, Oct 23).
  • A send or call to a session waiting on a person queues until FIX-2070 decides otherwise; a FIX-2129 test pins it (FIX-2058 D2).
  • FIX-2058's names stay: hold and updateTask's question field are FIX-2071's to rename.

FIX-2056's wrap had recommended fixing FIX-2182 and FIX-2183 before FIX-2058 built on them; neither was done and both are still Backlog. What FIX-2058 left open is in PLAN.md → What FIX-2058 left open.

Open

None.

Reviewers · look here

  • PR-3, the project's first rule about a person's stop. It names two answer routes and forbids a third. See BUSINESS-RULES.md.
  • FIX-2092, phased workflows, newly listed. Its approval gate between phases would be a FIX-2058 hold, so it is the first epic to build on PR-3. See SPEC.md.
  • PD-2 reverses part of the project's original framing (skills declaring a pattern). See DECISIONS.md.

Spec · Decisions · Rules · Plan · The prior Linear content, verbatim: absorbed/linear-content.md on the branch

How to review this

This is a project-spec: the standing outcome, the cross-epic decisions, and the live state of
one Linear project, in four documents. It is not an implementation plan, not any one epic's
objective, and not a design.

In scope to challenge:

  • The outcome — is this what the project should be driving at, and would we know if we got there?
  • Whether the project overbuilds. Each epic can earn its place while the whole is too much.
    That question can only be asked here.
  • A decision in DECISIONS.md that binds several epics, or a rule in BUSINESS-RULES.md with no
    owner or two.
  • A missing epic the outcome implies, or one listed that does not serve it.

Out of scope — owned by the epic specs below it:

  • Any single epic's objective, issue division, or internals. Those are its epic PR.
  • The figures at the pixel level, and the status in the epics table, which is refreshed as
    the project moves and is never a review finding.

Feedback in the second list is routed to the epic it concerns, not folded in here.


🤖 Generated with Claude Code

https://claude.ai/code/session_01NczDiF2e2sEZQs12ZLvCxB

jhoffner and others added 2 commits October 11, 2026 22:12
…verbatim

Snapshot of the project's own content field, read 2026-10-11, before the
project-spec set is written over it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Outcome, territory, epics table and dependency graph; PD-1 and PD-2 with
decided-once; PR-1 and PR-2 owned by FIX-1870; the arc. One open ask:
whether True Plan Mode and the docs catalog stay in this project.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@vercel

vercel Bot commented Oct 12, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

3 Skipped Deployments
Project Deployment Actions Updated
flow-state-dev-docs Ignored Ignored Oct 31, 2026 12:18am UTC
flow-state-dev-kitchen-sink Ignored Ignored Oct 31, 2026 12:18am UTC
flow-state-dev-ui Ignored Ignored Preview Oct 31, 2026 12:18am UTC

Request Review

@jhoffner

Copy link
Copy Markdown
Collaborator Author

@codex review

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 12, 2026 •

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-31T22:58:41.140726Z 7927905 Manual request
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

Copy link
Copy Markdown
Collaborator Author

Second look + architecture pass (docs-only PR, so no code refactor applies)

I checked the spec's factual anchors against the repo: addTask({ waitForResponse }), handOff (packages/workforce/src/coordinator/coordinator-keys.ts), post-to-mailbox (named in goals/ and specs/issues/FIX-1792, not in packages/*/src), the claim ticket (orchestration/src/task-board) and workforce/src/delivery-ledger.ts all exist. The "4 paths, 2 fences" read holds up. The findings below are about shape, not accuracy.

Worth changing before approval

  1. The outcome is a solution, not an outcome. "Winning when… through one tool" is FIX-1870's design restated as the project's finish line. The kill line then says the one-tool premise may be wrong, so the headline and the escape hatch contradict each other. Better: state the win behaviourally ("a task is answered once, whichever path hands it over; every pattern stores work as Task rows"), and keep "one tool" as the current hypothesis for getting there. That also keeps the project honest if the kill line trips.
  2. PR-2 reaches into code this project says it doesn't own. The territory figure puts Workforce (Layer 2) "owned elsewhere", but PR-2 and the prose under it commit FIX-1870 to removing the coordinator delivery ledger, which lives in packages/workforce. Either record which project agrees to that change (a PD-style call), or narrow the rule to "the board ticket is the fence for dispatchTask deliveries" and let Workforce retire its ledger on its own schedule.
  3. PR-1 and PR-2 can't be checked yet. Both are owned by an epic that hasn't passed its gate, and "Checked where" points at goal checks whose specs don't exist. Rules with no owner able to check them are aspirations. Mark them proposed until FIX-1870 is approved, or drop them from BUSINESS-RULES.md for now. The rule-count for the arc/matrix figure convention is also unaffected either way.
  4. The unnumbered "No epic may…" paragraph is a third rule with no owner. Two of its three clauses are PD-1 and PD-2 restated. Either number it (PR-3, with an owner) or delete the restatement and point at the decisions.
  5. A known-wrong status is encoded in the headline. The spec says FIX-1312 is shown in flight "because Linear says so" while also saying it's done in practice, and the arc figure draws an in-flight bar and counts "1 in flight". Cheaper than carrying the disagreement in three places: wrap or re-scope FIX-1312 in Linear (a one-field change), then refresh. The project would then read "6 done · 1 filed".

Leaner version (second-look)

The project currently has one live epic's worth of future (FIX-1870) and a retrospective for the rest. That's legitimate as a standing frame, but it risks being a second copy of FIX-1870's objective. Test for overbuild: could the project's remaining life be one line in Linear plus FIX-1870's own epic spec? Keep the project-spec only if a second epic is expected to need PD-1/PD-2. If none is, say so in PLAN.md ("no epic is filed after dispatchTask"), so the project can close with it.

The open decision (True Plan Mode / docs catalog)

Agree with the recommendation. One addition: FIX-451 is parked "until FIX-1870 wraps", but FIX-922 (what the catalog should still list) is independent and could be resolved now. Worth a line saying whether it stays blocked.

Architecture notes for FIX-1870 (not for this PR)

  • The kill line depends on a per-tool behaviour inventory (what each of the four paths does on timeout, restart, double delivery, park). The spec should require that inventory as the first deliverable, before any consolidation code.
  • The consolidation is the real refactor opportunity: when FIX-1870 touches task-tools-capability.ts (large, addTask already carrying ask semantics) and the coordinator route, fold the dispatch path into the board rather than adding a fifth wrapper.

Out of scope per the PR's own list, so I did not comment on the epics table status or figure pixels.


Generated by Claude Code

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect · coherence check on the project-spec

The outcome and the decisions line up with FIX-1870 as filed. I found three things where the spec doesn't yet account for work that has already landed or been decided.

  1. The inventory misses the conversation board from FIX-1794 (FIX-1794 P2: a conversation files tasks for its delegates and hears each one end #2935, merged Oct 9). That PR shipped recordEnding (orchestration/src/tasks/collection/ending.ts) and the conversation board's notice markers (workforce/src/conversation-board/ledger.ts, notice-delivery.ts). A marker lands with the ending, and only delivering the notice clears it. So that's a third place that tracks whether an ending has been delivered, alongside the claim ticket and the coordinator delivery ledger, and "conversation files tasks for its delegates" is arguably a fifth hand-off path. As written, PR-2 ("no other ledger fences it") would be broken on day one, even after FIX-1872 removes the coordinator ledger. Please either count it in The read (making it 5 and 3) and give FIX-1872 the job of folding it in, or state explicitly that the notice marker is a delivery debt and not an answer-once fence, and say why PR-2 still holds.
  2. FIX-1786 (hand-offs and Workforce concepts) isn't in the graph. Only FIX-1815 appears as an input. FIX-1786's D9 sweep (spec(FIX-1786): D9 sweep, and the board's seat becomes assignee #2841) and the FIX-1794 board are where the task envelope, assignee and the filing API were settled. FIX-1870's regression gate depends on them, so they should show up as inputs the way FIX-1815 does.
  3. Retired vocabulary. What this project is not and figures/territory.svg both say "Workforce seats" (seats · WORKER.md). FIX-1786 D7 (spec(FIX-1786): D9 sweep, and the board's seat becomes assignee #2841) retired "seat" everywhere, so this should be workers or assignees. Separately, FIX-1870's own vocabulary rule keeps "hand-off" for passing on work you were already doing mid-way, and calls delivering new work dispatching a task. The spec's outcome line ("hands work to another worker", "4 hand-off tools", "hand-offs are one tool") still uses the old wide meaning. Consider "dispatch paths" so the project-spec doesn't teach the term FIX-1876 is about to narrow.

None of these blocks the open question about True Plan Mode and the catalog. Points 1 and 2 are the ones that could let FIX-1872 ship with two fences still in place.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect: PR-2 contradicts a merged decision (FIX-1816 D3)

This adds to point 2 of the second-look review above. The problem is bigger than ownership.

The conflict. FIX-1816 D3 (merged in #2900, aligned in #2905) chose to keep a coordinator's post as a post instead of making it an ask. Its "Locks in" line reads: "Two answer-once checks stay: the post's ledger token and the row's claim ticket. Each serves one kind, and no third is added." FIX-1815's DECISIONS also routes the ledger-token question to FIX-1816's spec, and that spec answered it by keeping the token.

This spec says the opposite in three places:

  • PR-2 in BUSINESS-RULES.md: "the board claim ticket is what refuses a second delivery; no other ledger fences it."
  • The prose under the rules: FIX-1870 "removes the exceptions: the coordinator's handOff, post-to-mailbox, and the coordinator delivery ledger."
  • The read in SPEC.md: "2 answer-once fences … Done is 1 and 1."

If this merges as written, FIX-1870 gets built toward removing a fence that FIX-1816 decided to keep. That's two directions at once.

What to change, pick one:

  1. Respect D3 (default). Rewrite PR-2 to cover what this project owns: "When a board task's ending is delivered, the claim ticket refuses a second delivery." Drop the coordinator delivery ledger from FIX-1870's removal list and from "Done is 1 and 1", so Done becomes one hand-off tool and two fences, one per kind. Keep the coordinator post (handOff / post-to-mailbox for posts) out of the consolidation, or say exactly which part of it folds into the board.
  2. Reopen D3. If the project really wants one fence, it needs a new owner decision that flips D3. D3 names its own flip condition: "the product wants a coordinator to compose one reply." Record that as a D-n through Decision Manager, cite it here, and then update FIX-1816's DECISIONS. Until then, mark PR-2 and the ledger removal proposed, contradicts FIX-1816 D3.

Either way, add FIX-1816 to the arc/territory inputs next to FIX-1815. It's the spec that settled this question.

@cursor cursor Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Simplify review

Verdict: The four-document project-spec plus territory SVG, arc SVG, and mermaid dependency graph is the right shape for a seven-epic, months-long Linear project. Bootstrap-thin scaffolding would under-serve this one; the template’s “full SPEC.md on day one” rule is met without the set feeling like busywork. Directory layout (spec/_projects/orchestration-primitives/, absorbed/linear-content.md, figures/) matches project-spec-template.md and peer project/* branches — no cross-import or organizational smell.

A simpler approach considered: Standing up only SPEC.md (or deferring BUSINESS-RULES.md / PLAN.md until a second epic) would cut file count but would hide PD-1/PD-2 and the arc before readers need them; with seven epics already in flight, that deferral would age poorly. Replacing hand SVGs with mermaid-only would be more complex at this altitude: territory and arc are positional figures per spec-figures.md; mermaid is correctly reserved for the epic dependency graph. Third-party packages don’t apply here.

Where detail went slightly past “minimum correct” (optional tighten-ups, not direction):

  1. DECISIONS.md → Open — The True Plan Mode / catalog fork is written as a full six-part gate inside a standing doc. Template convention puts live asks on the never-merged PR body; keeping a long Open section risks drift from the PR and blurs “decided” vs “needs sign-off”. Suggestion (if you agree): one-line pointer in DECISIONS.md (“Unresolved: FIX-312/449/451 ownership — see project PR”) and keep the six parts on the PR only.

  2. Status lives once — project-spec-template.md says the epics table in SPEC.md is the single live read; PLAN.md already says to point at Spec → the epics, but lines 9–10 re-narrate FIX-1312 and dispatchTask state. The FIX-1312 editorial also appears in the epics table note. Suggestion: keep that paragraph only in SPEC.md; let PLAN.md explain how to read the arc figure, not which epics are open.

  3. figures/arc.svg maintenance — Calendar-mapped x positions and a fixed “now” line are faithful to the template but every refresh is a multi-field coordinate edit (and long aria-label duplicates table prose for agents that read whole files). Suggestion for future refreshes: coarser bars (done / in flight / not started) without day math, or accept churn as the cost of a PR-body figure — not worth a codegen pipeline in this PR.

  4. BUSINESS-RULES.md — Two rules that bind many epics in intent but cite unwritten child specs for “checked where” read like FIX-1870 acceptance criteria elevated early. Prohibitions at the bottom largely restate PD-1/PD-2. If this feels heavy: shorten check columns until FIX-1870’s epic spec exists, or leave headings-only until a second epic truly needs a project rule — I’m not sure that’s better than documenting intent now; flagging only.

  5. Territory vs bullets — SPEC.md “What this project is not” and the SVG outside band overlap (Relay, Strands, human questions). Pick one surface for enumeration or thin the other — taste call.

Skipped (needs product owner or larger refactor): moving issues between Linear projects (Open fork), auto-generating the epics table from Linear, shared SVG CSS across figures (CSP keeps styles inline; only two files today), removing absorbed/linear-content.md (required by project-agent).

Checks: Doc-only PR; no validate-spec-folder on project/* branches per repo guards. Figure/word budgets look within template ballpark (~700-word SPEC.md target).

Approving direction: consolidation narrative (4→1 hand-off tools, 2→1 fences) is clear, kill line is testable, and the absorbed snapshot preserves reversibility against stale Linear content.

Open in Web View Automation 

Sent by Cursor Automation: Review For Best Approach

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e18f506b8b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

@cursor cursor Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Snob. I tried cutting the two captions. It tied. No POC.

I read spec/_projects/orchestration-primitives/SPEC.md, DECISIONS.md, BUSINESS-RULES.md, PLAN.md, absorbed/linear-content.md, figures/territory.svg, and figures/arc.svg.

I deleted the lead under the title and the two sentences under the arc. That was 7 lines. The lead names a Task row in a TaskCollection and a board of dispatcher, worker, and loop; the territory figure and PD-1 still say that, and the winning-when line's "that board" meets the figure in the next section. The arc sentences say the substrate is done, FIX-1312 is the open bar, and dispatchTask is next; the figure draws that, the epics table records it, and the PR body carries the same sentence. Same contract, 7 lines lighter.

I left the six-part ask in DECISIONS.md → Open and in the PR body, the epics table, the mermaid, the consumes/releases table, the arc, and absorbed/linear-content.md. Those are the surfaces this project-spec keeps. Head e18f506b8bf1b6d259bc24e9879d17ca65775655.

Open in Web View Automation 

Sent by Cursor Automation: Code Snob

jhoffner and others added 2 commits October 11, 2026 22:29
- Outcome stated as behaviour (answered once, whichever path); one tool is the bet
- The read: 3 dispatch paths (post-to-mailbox left with FIX-1792); notice marker named as a debt
- PR-1/PR-2 proposed until FIX-1870's gate; PR-2 narrowed to board tasks
- Open: the coordinator ledger, FIX-1870 vs FIX-1816 D3
- PD-2 matches skills today (no team on a skill, FIX-1814)
- FIX-1816 and FIX-1786 as inputs; "seats" -> workers; PLAN arc caption deduped

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…n is on the board

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@jhoffner

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: reply

Re: Second look + architecture pass. Changes are in 3e1de3d.

# Outcome
1 · outcome is a solution Folded. Winning when now describes behaviour: work one worker gives another is answered once, whichever path delivers it. "One dispatch tool (FIX-1870)" is named as the current bet. The kill line now says answered-once stays the finish line if the bet fails.
2 · PR-2 reaches into Workforce Folded. PR-2 is narrowed to board tasks. The ledger removal is now a six-part ask in DECISIONS.md → Open, because it conflicts with FIX-1816 D3 (see the reply to the D3 comment).
3 · rules can't be checked yet Folded. Both rules are marked proposed until FIX-1870's gate. Checked where now reads "once its spec lands".
4 · "No epic may" is a third rule Declined in part. The template asks for a prohibition list beside the rules, since a new epic's objective gate is checked against it. I trimmed it: the PD-1 and PD-2 clauses now link to the cards instead of restating them.
5 · known-wrong FIX-1312 status Declined. Status comes from Linear, and wrapping FIX-1312 is its owner's call, not this agent's. The disagreement now sits in one place (SPEC.md) instead of three. Once FIX-1312 is wrapped in Linear, the next refresh reads "6 done".
Leaner version Folded. PLAN.md → Deliberately not next now says no epic is filed after FIX-1870, and the project's remaining life is that epic plus the two open questions.
FIX-922 independent Fixed. PLAN.md now says FIX-922 is about patterns, not dispatch, and doesn't wait on FIX-1870. FIX-451 still waits. The open placement question is unchanged.
Architecture notes for FIX-1870 Not folded. They belong to FIX-1870 and are routed to its coordinator. The recommendation in Open 2 does depend on the per-path inventory you describe.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: reply

Re: Architect · coherence check. Changes are in 3e1de3d.

# Outcome
1 · FIX-1794 notice marker Fixed. Verified in workforce/src/conversation-board/ledger.ts: the marker lands with the ending, only delivery clears it, and the board's next touch replays it. That makes it a debt (an ending must be told), not a fence (a second telling is refused). The read now names it that way. PR-2 is narrowed to the claim ticket for board tasks, so the marker doesn't break it. One part is routed to FIX-1872 rather than folded: show that a replayed notice can't be told twice.
2 · FIX-1786 missing as an input Fixed. FIX-1786 (task envelope, assignee) and FIX-1816 (the two fences, D3) are now dotted inputs in the graph next to FIX-1815, and are listed in PLAN.md's Consumes column.
3 · retired vocabulary Fixed. "Seats" is now "workers" in SPEC.md and in figures/territory.svg, which was re-rendered and checked in both themes. The outcome, read, kill line, rules and the Open 1 ask now say "dispatch" instead of "hand-off". "Hand-off" stays only for the board hand-over and the handOff tool's own name.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: reply

Re: PR-2 contradicts FIX-1816 D3. Verified on main: D3's Locks in keeps two answer-once checks. FIX-1870's Linear description, filed after it, removes the ledger through FIX-1872. Two product-owner-approved texts disagree, so neither this agent nor the project-spec picks one.

What changed in 3e1de3d:

  • PR-2 now reads "When a board task's ending is delivered, then the claim ticket refuses a second delivery". That holds under both options. It widens to "no other fence" only if the ledger goes.
  • The ledger is no longer on FIX-1870's removal list in the rules prose. The read now says "Done is 1 path; whether it is 1 fence or 2 is open".
  • Open 2 in DECISIONS.md and in the PR body is a six-part ask: keep the coordinator's own answer-once check, or let the board's be the only one? The recommendation follows FIX-1870, but only if its per-path inventory shows a post sent as a row keeps each delegate's own line, and then FIX-1816's D3 gets amended. Otherwise D3 stands.
  • FIX-1816 is now an input in the graph and in PLAN.md.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: reply

Re: Code Snob (review 5486058305). Changes are in 3e1de3d.

  • The sentences under the arc: Folded. They are replaced by one line on how to read the figure. Status now lives only in SPEC.md.
  • The lead under the title: Kept, reworded. It defines the Task row and the board before the outcome table uses them, and the Linear mirror shows the figure only as a link. It now reads "FSD's multi-agent patterns run on…", not "every", because roundRobin and debate still run off the board.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect · re-check of 3e1de3d

The three replies match what I asked. PR-2 now covers board tasks only and holds whichever way the ledger question goes. The ledger is off the removal list and the D3 conflict is recorded as Open 2 instead of being picked by the spec. FIX-1786 and FIX-1816 are drawn as inputs, "seats" is gone, and "dispatch" replaces the wide "hand-off". Calling the FIX-1794 notice marker a delivery debt rather than a fence is right: it replays an ending, it doesn't refuse a second one. I accept the partial decline on point 4 and the decline on point 5.

One thing to carry outside this PR: FIX-1872's Linear text still says the ledger goes and that FIX-1815 "decided on one fence". That reads as settled, while FIX-1816 D3 says the opposite. Until Open 2 is decided, FIX-1872 should stay in Backlog and its description should point at Open 2 so nobody specs it from the Linear text alone.

No further asks from me on the project-spec.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect · re: D3 reply (3e1de3d)

Checked against what I asked. It matches: PR-2 now covers only board tasks and holds whichever way the ledger goes, the ledger is off FIX-1870's removal list here, and FIX-1816 is an input in the graph. Framing the conflict as an open owner call, with D3's own flip condition as the test, is the right move. Resolved from my side.

One thing to keep it from being decided in two places: the same conflict is raised on the FIX-1870 epic-spec in #3109 (D2/ER-5 removes the ledger with no EVOLUTION row against D3). Open 2 here and #3109 should land on one answer. Whichever way it goes, record it once through Decision Manager or as an amendment to FIX-1816 D3, then have both #3109 and this Open 2 cite it. Until then FIX-1872 shouldn't be specced to remove the ledger.

… 1 and Open 2

Open 2 becomes PD-3: anything that expects an answer is a task through
dispatchTask; a tell is a plain message with no row, deadline or once-only
check. Amends FIX-1816 D3. PR-2 restored with the split, with a three-way
trade-off figure. Open 1: FIX-312/FIX-449 moved to Human-in-the-Loop,
FIX-451 to Multi-Agent Patterns; removed from Open and the plan.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Figures only; no decision text, rule or scope changed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect · FIX-1816 D3 is still stale on main after #3110 merged

#3110 merged at afbba15 and the owner approved the FIX-1870 epic spec, so D2 is settled: a coordinator post that needs an answer is a task through dispatchTask with the claim ticket as its only fence, and a tell is a plain message. FIX-1872 removes handOff and the delivery ledger together.

The gap I raised on #3110 didn't land with it. specs/issues/FIX-1816/DECISIONS.md D3 on main still says, under Locks in, that two answer-once checks stay (the post's ledger token and the row's claim ticket). It has no amendment pointing at FIX-1870 D2. So two merged specs now give opposite answers on the ledger, and FIX-1872's implementer could read either one.

This PR is the open spec that carries the same answer (Open 2 / PD-3), so please add the fix here: a short dated entry in FIX-1816's existing "Product owner amendments" section, or a one-line note on D3. It should say the delivery ledger is removed in FIX-1872 per FIX-1870 D2, and that each delegate's own line, rounds and routing stay and now ride on tasks. A tell is a plain message. If a separate one-file PR is cleaner, that works too, but it should land before FIX-1872 starts.

…patterns; arc to v1/v2/v4

Owner's call (Oct 12): the outcome covers patterns that run on the task board;
roundRobin and debate are outside it, FIX-922 still decides their placement.
Recorded in decided once. PLAN gains the milestones (v3, v5 deleted Oct 12).
Status: FIX-1870 objective approved Oct 12, now in flight; arc redrawn.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@jhoffner

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: update

Outcome narrowed to board-based patterns (owner's call, Oct 12). Changes are in d233f51.

  • Outcome: Winning when now reads "every pattern that runs on the task board". roundRobin and debate are outside the outcome, and FIX-922 still decides whether they stay patterns. The lead, PD-1's card and figure, What this project is not and Deliberately not next all say the same.
  • Recorded as a decided once row (Product owner, Oct 12). No trade-off figure: it is a scope call, not a PD card. DECISIONS.md → Open had no item for it on the branch, so nothing else needed retiring.
  • Milestones: PLAN.md gains a Milestones table for v1, v2 and v4, noting that v3 and v5 were deleted Oct 12. v4 ("Skills declare patterns") still describes the frontmatter that PD-2 reversed. Renaming or deleting it is the owner's call.
  • Status: FIX-1870's objective was approved Oct 12, so it is now in flight (5 done · 2 in flight). The arc shows its bar and the now line has moved to Oct 12. PR-1 and PR-2 are no longer marked proposed.
  • The Linear project content and summary are re-mirrored. The PR body is re-pinned to d233f51.

🤖 Generated with Claude Code

jhoffner pushed a commit that referenced this pull request Oct 21, 2026
Sources agree with D2 (gaps file); prune target reconciled with tenet 3
and its stale rows refreshed; #3102's PD-3/PR-1/PR-2 re-homed; engine
prerequisites named per child (FIX-2077 onFinished, FIX-2078 sweeper
due-time source, FIX-2082 engine result wait for a bare waited
dispatch); FIX-2082 owns the tree's 100-under-a-top-task cap;
FIX-2062 and FIX-1637 placed. Calls (e) and (f) left pending, with
migration.md listing one line per removal.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
… PD-3, PR-1, PR-2; record D2

FIX-2056's epic spec was approved by merging #3588 (e3b47e0). FIX-1870 is canceled
and superseded, so:

- Outcome: the current bet is FIX-2056 (createTask + getTask({ id, wait })); the read
  re-derived from main (ledger gone since FIX-1872, two row kinds remain); kill line
  re-pointed at FIX-2056's closure. SPEC trimmed 912 -> 704 words.
- PD-3 and PR-1 to FIX-2083, PR-2 to FIX-2081, per FIX-2056 SPEC's "Already here" row.
- Decided once: FIX-2056 D2 (no child waits on FIX-2058; answerTask stays until then).
- PLAN: order table and "not next" carry FIX-2056 -> (FIX-2058, FIX-2070) -> FIX-2071.
- Territory: the dashed dispatchTask box is now FIX-2056's four task tools. Arc and
  PD-3 figure text updated.
- Open: FIX-2058 vs "not a way to ask a person" — widen the outcome or move the epic.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A

Copy link
Copy Markdown
Collaborator Author

from: project-agent
session: project-orchestration-primitives
kind: update

FIX-2056 approved (#3588, e3b47e0); folded in e5c5b7e.

  • Outcome: the current bet is FIX-2056 (createTask + getTask({ id, wait })); the read re-derived from main (the ledger went with FIX-1872 on Oct 13; two row kinds remain); kill line points at FIX-2056's closure. SPEC.md trimmed to ~700 words.
  • PD-3 and PR-1 re-homed to FIX-2083, PR-2 to FIX-2081.
  • Decided once: FIX-2056 D2 — no FIX-2056 child waits on FIX-2058; answer-by-steer and waiting on a person are FIX-2058's; answerTask stays until then. FIX-2058's Linear description updated to match.
  • PLAN: FIX-2056 → (FIX-2058, FIX-2070) → FIX-2071; territory figure redrawn.
  • Open (owner): FIX-2058 vs "not a way to ask a person" — widen the outcome or move the epic. DECISIONS.md → Open.

The body's figure lines stored with backticks in their src again, so the body links to the documents; the clean pins are with the coordinator.


Generated by Claude Code

…p per person; PD-3 figure past the ledger

Records the Open ask as decided (Oct 21, coordinator under the owner's
standing authorization for reversible calls): a task that stops for a
person is inside, as one Inbox item answered once; FIX-2058 stays here.
The asking screens stay outside. PD-3's middle column and caption no
longer call the ledger-backed message today's: FIX-1872 removed it Oct 13.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
jhoffner pushed a commit that referenced this pull request Oct 21, 2026
… held entry, migration completed

- The field table is read inside the collection's guarded patch FIX-2080
  adds; the updateTask block maps input and names the refusal.
- Drop the recorded-row map entry from Adds: FIX-2080's D1 owns it.
- Leg b covers dispatchTask's reassign and release while it stands.
- D2: chain limits bound all; only the user's file sets all; filedBy
  takes the filing task's id; labels is last-write-wins before start.
- Epic: (e), #3102 PR-2 re-home and migration rows amended for D1;
  addTask maps to createTask({ held: true }); missing FIX-2081 rows added.
- Review notes recorded for the implementer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
FIX-2058 (one Inbox) moves to in flight: epic spec drafted on
epic/waiting-on-a-person, six children filed Oct 23. Epics table, counts,
the outcome's Now line and the arc (bar for FIX-2058, now line Oct 23).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vdxnv28JumHyq2za2ybqz6
…roved

FIX-2058 (one Inbox) epic spec approved and merged Oct 23 (PR #3732,
e786df6); child specs starting. Epics table row and surface cell, the
outcome's Now line, and the arc annotation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vdxnv28JumHyq2za2ybqz6

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d95795989c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

… half; Now at epic altitude

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6ac7bfd12d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…o epic status; plan drops child-issue progress

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. 🚀

Reviewed commit: 62931d9ef6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…ded; five calls decided once

FIX-2056 done (wrapped Oct 24, closure FIX-2087 passed on main at #3847);
the read reaches 1 way to give work; kill line passed, the Inbox half left.
Decided once: old task APIs removed outright; minute-scale crash bound
(FIX-2169 deferred); park releases a timed-out wait's hold (FIX-2170);
the four reservation records stay separate; a wait on a finished task
returns once its notice is handled (FIX-2181). Plan records what FIX-2056
left open and recommends FIX-2182/FIX-2183 before FIX-2058 builds on them.
Arc: FIX-2056 bar closed, now line Oct 24.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
@linear-code

linear-code Bot commented Oct 24, 2026 •

Copy link
Copy Markdown

FIX-2182

FIX-2183

…nded; PR-3 and four calls decided once

FIX-2058 done (wrapped Oct 27, closure FIX-2134 passed on main 83c444b at
#3951). The read's second count reaches 0 ways to answer a person's stop
outside the Inbox; the kill line passes for both halves. New rule PR-3: a
stop for a person is one Inbox item, a waitForUser suspension or a row hold,
answered once through two routes; no epic adds a park or a third route.
Decided once: no park fallback without durable execution; no default
deadline on a person's stop; telling the filer is FIX-2070's; a message to a
waiting session queues until FIX-2070 decides; renames stay FIX-2071's.
Plan records what FIX-2058 left open. Status: FIX-2092 (phased workflows),
an epic in this project the set never listed, added as free to start.
Arc: FIX-2058 bar closed, FIX-2092 lane, now line Oct 27.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiF2e2sEZQs12ZLvCxB
…dMessage and calls

PR-2: tracked work (attempts, a retry, an ending owed to its filer) is a
task; otherwise a post is a message on the one message tool, sendMessage,
where a call waits on the engine's wait gate and a tell waits on nothing.
"No epic may" now forbids a second model-facing message tool. PD-3 gains an
amendment note pointing at FIX-2070's epic spec (#3983).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiF2e2sEZQs12ZLvCxB
…not a claim (FIX-2070 D2)

"No epic may" add a second once-only check beside the claim ticket; the
clause now says a delivery's take of a notice or a message is idempotency,
not a claim, per FIX-2070 D2 (approved by the owner).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiF2e2sEZQs12ZLvCxB

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 980a48edde

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

…om calls, per FIX-2070's amendment

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@jhoffner

Copy link
Copy Markdown
Collaborator Author

@codex review

1 similar comment
@jhoffner

Copy link
Copy Markdown
Collaborator Author

@codex review

@chatgpt-codex-connector

Copy link
Copy Markdown

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: 7927905299

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants