Repository navigation
spec(FIX-2083): Routed tasks replace posts: a coordinator hands a user's message to a delegate by filing a task - #3598
Conversation
Retained spec set: a coordinator hands a user's message to a delegate by filing a task with the routing settings; one ending path; a round's deadline on the coordinator's wake; one fit check. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
|
The D1 trade-off figure did not render in the description: the stored To fix the description, a person can paste this clean line under the D1 heading in place of the broken one: Generated by Claude Code |
|
@codex review |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Second look + architecture pass (direction check, not a build review)I read SPEC.md and DECISIONS.md and checked their premises against 1. D3 should not ride in this PR, and its mechanism isn't settled (scope)Folding the two fit calls into one has nothing to do with posts becoming tasks. It could land before or after this change without either one noticing. It also changes behavior, and the spec doesn't say how.
2. Rule 5 queueing interacts with rounds and with rule 9, and the spec doesn't say howThe spec accepts "a second message waits" for best fit's hold. It misses two harder cases:
3. D1: attach a reference to the message, not a copy (BP-042, restraint)"The server attaches that message, as written, with who wrote it and the recent lines, to every task the turn files with its result to the conversation."
4. In-flight posts: leg e vs. "no entry", please state the mechanismLeg e requires "a seeded post row is worked and answered once". The same spec removes the post entry and builds "no read of its own (ER-6)". The only path is FIX-2080's map entry turning the old row into a routed task at claim time. Name that dependency explicitly. Make leg e fail if FIX-2080's map is absent, so it can't pass on something the post entry still does. Dual-read of round state and routing records naming a post id (BP-030) is stated, and good. Dependencies (process)FIX-2078, FIX-2080 and FIX-2081 are all still-open spec PRs (#3594–#3596), so every "builds on" claim (the map entry, the single ending path, the wake, Architecture opportunity tied to this changeDeleting Housekeeping
Generated by Claude Code |
There was a problem hiding this comment.
Simplify review (spec only)
Problem this spec solves: coordinators still route user messages through a parallel post row kind (its own entry, timer, cancel path, and settlement hooks) while the epic is converging on one task primitive. The direction here—routed work is an ordinary task on work with routing settings, one onTaskSettled path, and a public sweep that no longer writes posts—is aligned with FIX-2056 and reads net down on framework surface. Rejecting “drop the kind only, keep the post entry” is coherent: the entry is where the timer and taskTurn: false stopgap live.
Where complexity may be higher than necessary is mostly bundling, not the core post→task bet:
-
Three sign-off decisions in one issue (D1 + D2 + D3). Each has its own goal leg, failure mode, and dependency (2078 for D2; prompt/schema change for D3). The epic line item already names “one fit check,” so D3 isn’t out of left field—but D3 is the strongest candidate to peel out if you want a thinner implementation PR and a cleaner direction gate: it is largely orthogonal to “nothing knows a post,” shares no code path with
delegated-post.ts, and the SVG already admits borderline lead verdict drift. If you keep D3 here, I’d still tag goal legs in SPEC as core (a,c,e + sweep) vs D2 (b) vs D3 (d) so implementers know what can ship in slices without re-litigating the whole goal. -
D1 bundles two API expansions: server-attached user message (the real FIX-1872 D2 carry) and public
targetoncreateTask. The product fork in D1 is briefing in the coordinator’s words vs Alice’s words—not workstream addressing. Policy filing already hadtargeton the post form; judgment + workstream might be satisfiable by reusingresolveWorkstreamTargeton the routing filing path without wideningcreateTaskin the same gate as attachment. Worth weighing at sign-off, not a definite “must split.” -
Two filing writers in DECISIONS (“routing’s filing step” and judgment
createTask). That’s the right convergence point for FIX-2080’s guard, but the spec should make implementers reach for one internalfileRoutedTask(or extend today’sdeliverTo/filePostmigration) so brief assembly (recent lines × N on an everyone turn) isn’t duplicated. Performance-wise, attaching the same context per task in one turn is real amplification; “build once per routing turn, reuse across filings” is an implementation must, not spelled out today. -
Editorial / reader cost (minor): SPEC debt table, DECISIONS “Decided, not asked,” and per-D SVGs restate the same invariants; sign-off repeats D1–D3 “if wrong” lines. Trimming duplication would not change the design.
Reuse (implementation, not spec defects): existing deliverTo / post idempotency, workerTaskEntry + taskTurnMessage, delegate session keying from post rows, and collapsing task-settled.ts post vs task branches are the right cuts—no new packages, and patterns stays out of scope. Third-party packages don’t apply.
Verdict: I’m not requesting changes to the core approach (post→task + D2 wake tied to removing the delegate timer is justified). I’m not approving yet because I think sign-off should explicitly consider unbundling or tagging D3 and whether target must ride D1—if you confirm both belong in FIX-2083 as written, this is directionally ready to merge as a spec.
Automation: simplify pass; read-only subagent review + codebase spot-check.
Sent by Cursor Automation: Review For Best Approach
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 48372389c1
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".
…the no-sort pass, queue cost named Folds Codex's attempts and post-free-goal findings and the second look's D3 mechanism, queueing and leg-e findings; the rest recorded in Notes from review. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
Review round 1, folded in c85f8aaAnswering the Cursor simplify review, the Codex review, and the second look above. Folded (these change the spec):
Not folded:
Spec review rounds: 1 of 2. Generated by Claude Code |
…the model passes Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
Architect: not approving yet. I found no fork with the sibling specs:
The label waits on Jake's sign-off:
— FSD Architect |
|
Sign-off (2026-10-21), all three as the spec proposes:
|
|
from: fsd-claude Cross-spec coherence audit: what FIX-2083 (#3598) needs The owner asked for one coherence audit across all twelve FIX-2056 specs (merged FIX-2077 #3592 and open #3593–#3604), read against each other, the epic and main. 66 conflicts survived an accuracy check and a materiality check (or a tie-break); 13 of them call for an edit here. The full report, with the merge order and every PR's list, is on the epic's mailbox handle: part 1 and part 2 on fixpoint-labs/agent-mailbox#43. One item below (F1) waits on the owner's call before anyone edits for it. High
Medium
Low
Merge order that touches this PR
Generated by Claude Code |
F5 goal rewrites named; F18/F19 round wake entry and wake-less host scope; F28 routing excludes the coordinator; F32 target rule; F33 FIX-2080's attempts rule; F38/F57/F39 migration rows and changeset lines; F43 FIX-2077 blocks FIX-2083 (epic row, graph, chain); F44 interim before FIX-2082; F61 dispatchTask post form points at FIX-2081's row. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
from: fsd-claude Cross-spec audit edits applied in ee0f497
Generated by Claude Code |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
from: fsd-claude F1 applied in 3d454ae (owner approved the carve-out at 17:15Z)
Generated by Claude Code |
…ne into FIX-2081's dispatchTask row Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
Merged main now that FIX-2081's spec (#3596) has landed. What changed:
No decisions changed. Generated by Claude Code |
|
@claude resolve conflicts |
…IX-2087 specs) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
Done. The branch now includes the latest main, which has the FIX-2076, FIX-2078, FIX-2080 and FIX-2087 specs. There were two conflicts:
Generated by Claude Code |


Goal: every message that a coordinator routes becomes a task that its user can see and cancel. The task runs in the delegate's ongoing session. Its one complete answer lands in her conversation under the delegate's name. No part of the framework writes or works a post; a post row stored before the upgrade is read only through FIX-2080's map, until FIX-2071.
The check reads Alice's task list and her conversation. It does not read the coordinator's calls. Check:
goals/coordinators/routes-each-message-as-a-task/, no model. Control:GOAL_CONTROL=round-timermust FAIL leg b. Today'smainmust FAIL leg c. The coordinator goal and the FIX-2005 hand-off leg must also pass. Why this goal: SPEC.md.The six parts of a post go. The settings of a task, its one entry and its one ending path do the job.
How: routing files a usual task with three settings from FIX-2080. The delegate works it on
work. The ending comes back on the one path from FIX-2078 and lands under the delegate's name.Sign off
The goal, at that size. If wrong: FIX-2084 builds the worker contract on an entry that must go.
The debt: net down. Posts and their six parts go.
targetoncreateTaskand one due time arrive. If wrong:createTaskkeeps a field that only coordinators use.A judgment coordinator files with
createTaskand the routing settings; the delegate reads the user's words as written · D1<img src="
https://raw.githubusercontent.com/fixpoint-labs/flow-state-dev/48372389c1e022b2effee8ac3f6acf03d50224c0/specs/issues/FIX-2083/figures/d1-createtask-as-written.svg"width="940" alt="D1: createTask with the message attached, chosen, beside a post flag on createTask. Decides it: one shape of the tool against two. Price: a target field and an attached message" />It comes down to the shape of the tool: a flag makes two shapes of
createTask. If wrong: a field and an attached message oncreateTaskthat only coordinators use. This is the one to weigh.The deadline of a round is a wake on the coordinator's session · D2
It comes down to a restart: the wake closes the round on time, and the timer stops with the process. If wrong: one more due-time kind for the sweeper.
One fit check; it refuses a lead only when it says "none fits" · D3
It comes down to agreement: two calls can give one worker two verdicts. If wrong: a lead that opens today is refused, or the opposite.
Open: none. D1 is the one to weigh. Reasons and what lost: DECISIONS.md.
Reviewers · look here
createTask?Not here: steer (FIX-2086), the names
respondanddispatch(FIX-2084), messages to a coordinator (FIX-2070).Spec · Decisions · Linear FIX-2083 · epic FIX-2056 · builds on FIX-2078 (#3594), FIX-2080 (#3595), FIX-2081 (#3596) · supersedes FIX-1872 D1 · merges after human approval and required checks, per Merging and amending a spec
How to review this — altitude, what's in scope, what's deliberately unsettled
This is a direction set, not an implementation and not a build plan. Reviewing it well means
answering one question: is this the right approach?
In scope to challenge:
SPEC.md— is the goal the real need at full size,for the people named, or a smaller goal that would let us call this done early? Would the
named check fail if the goal were not met?
docs/philosophy.md?DECISIONS.md— that's the sign-off surface.SPEC.md→ What's already here, and the debt — an existing primitive,method or prior design the change overlaps and doesn't name; something it adds that is debt
and isn't listed; a net verdict the tables don't support.
design.
Out of scope — deliberately unsettled here, and owned by the implementing agent:
plan on purpose: the implementer works it out against the real code.
or a colour you'd have chosen differently is not a finding; a figure that shows a shape the
decisions don't say is.
poc/are retained,not production code. Read them for directional evidence, not production completeness.
Their isolation from production/default discovery and absence of secrets or generated
dependencies are in scope; retention is not permission to weaken repository checks.
One request, if you are an automated reviewer: the two lists above are the whole difference
between a useful review of this document and a long one. Volume is not signal here — a single
finding above the line is worth more than twenty below it.
Feedback in the second list is welcome and gets recorded verbatim in
DECISIONS.md → Notes from reviewfor the implementer to weigh against real code. It will notbe argued with, and it will not be folded into the design prose. Please don't re-raise it: one
mention is enough.
The one exception is the POC. Line-level feedback on it is dropped, not recorded — there
is nothing for an implementer to weigh about code that isn't shipping. Feedback about its
direction is not line-level and is treated like any other above-the-bar finding.
Direction approval does not waive merge-time required checks, approvals or thread policy.
Optional comments do not require another design round merely to reach zero comments.
Generated by Claude Code
````<source media="(prefers-color-scheme: dark)" srcset="https://greptile-static-assets.s3.amazonaws.com/badges/RetriggerDark.svg?v=2"><source media="(prefers-color-scheme: light)" srcset="https://greptile-static-assets.s3.amazonaws.com/badges/Retrigger.svg?v=2"><img alt="Retrigger" src="https://greptile-static-assets.s3.amazonaws.com/badges/Retrigger.svg?v=2" align="right">```` `````<source media="(prefers-color-scheme: dark)" srcset="https://greptile-static-assets.s3.amazonaws.com/badges/ViewInGreptileDark.svg?v=2"><source media="(prefers-color-scheme: light)" srcset="https://greptile-static-assets.s3.amazonaws.com/badges/ViewInGreptile.svg?v=2"><img alt="View in Greptile" src="https://greptile-static-assets.s3.amazonaws.com/badges/ViewInGreptile.svg?v=2" align="right">````` Confidence Score: 5/5
The spec appears safe to merge as a direction-setting change.Summary
This spec replaces routed posts with ordinary tasks. It changes no runtime code.
jhoffnerexplicitly accepts queued messages, possible changed verdicts for borderline leads, and the addedtargetfield and attached message. These trade-offs were not raised as defects.Diagram
%%{init: {'theme': 'neutral'}}%% flowchart LR M["User message"] --> C["Coordinator"] C --> T["Task on delegation board"] T --> W["Delegate's ongoing session"] W --> D["Filed work ends"] D --> S["onTaskSettled"] S --> A["One answer in the user's conversation"] K["Durable deadline wake"] --> R["Close overdue rounds"]Reviews (1) · Last reviewed commit: "spec(FIX-2083): routed tasks replace pos..." · Reviewed by Greptile
Generated by Claude Code