Skip to content

spec(FIX-2080): A task's filing settings and lifecycle: start, session, result and attempts; order and cascade - #3595

Merged
jhoffner merged 6 commits into
mainfrom
spec/FIX-2080
Oct 21, 2026
Merged

jhoffner merged 6 commits into
mainfrom
spec/FIX-2080

Conversation

@jhoffner

@jhoffner jhoffner commented Oct 21, 2026 •

Copy link
Copy Markdown
Collaborator
Someone who… Today After
files a follow-up to a busy worker Gets a refusal, and must try again The follow-up waits, then starts in that session when the open task ends
files a task that must wait for their go Moves it to blocked, and nothing unblocks it Files it held; it starts on release, not before
names a dependency on another board, or one that loops The task stays pending for good The filing is refused, with the rule
has a dependency fail or get cancelled The dependent waits for good The dependent and its own dependents are cancelled
cancels, retries or fails a task that filed work The filed work continues, unless the task was a split Each open task it filed is cancelled in seconds; after a crash, in seconds once FIX-2078 lands, on the next touch until then

Goal: a delegated task runs where, when and how many times its filer chose at filing. Nothing waits on work that cannot finish: no row stays pending for good, no filed task continues after the task that filed it, and no busy session refuses a task.

flowchart LR
  A["a real engine · durable store · the Workforce session board · deterministic workers"] --> L1["leg a · start: deps, held, another board"]
  A --> L2["leg b · session and result settings"]
  A --> L3["leg c · attempts 3"]
  A --> L4["leg d · cascades, and a kill before one is paid"]
  A --> L5["leg e · retired rows in the store at upgrade"]
  L1 --> R["rows read through the collection · the session each request ran in"]
  L2 --> R
  L3 --> R
  L4 --> R
  L5 --> R
  R -->|"every row ends or runs as filed · none pending for good"| P["PASS · goal met"]
  R -.->|"control · cascades off"| F["must FAIL leg d"]
Loading

The check reads the stored rows and the session of each request, not the calls of the board. Check: goals/task-board/runs-by-its-filing-settings/, no model, control GOAL_CONTROL=no-cascade, which must fail leg d. Why this goal: SPEC.md.

Today against after, for start, session, result, attempts, endings and upgrade: blocked, refusals and opt-in cleanup become held, a waiting session, two cascades on every board and one read map

Six rows. The left column is today; the right column is what the filing settings and the cascades give.

How: the four settings are fields of the task row. The Workforce board starts a row only when its start is met and no other open task holds its session. Each ending records the cascade it owes in the same write, for filed tasks and dependents alike; the ending's own pass pays it, and FIX-2078's wake pays it on time if that pass was lost. Until FIX-2078 lands, a touch pays a past-due cascade, so FIX-2078 now lands after this issue.

Sign off

The goal, at that size. If wrong: the routed tasks of FIX-2083 leave orphan work.

The debt: net down. Two not-started states become one, and one cascade replaces two partial cascades. Two fields come in. If wrong: two fields that only Workforce reads.

One status, held, replaces blocked and recorded-not-started · D1

D1: one status for not started. Held replaces blocked and the recorded mark, chosen, beside held added beside them. Decides it: a screen listing work that has not started. The price: apps outside reading blocked rewrite at upgrade

It comes down to a screen of work that has not started: one word, or three words with one meaning. If wrong: apps outside the repo that read blocked must change at upgrade for no gain.

A cascade is a debt recorded with the ending, paid in seconds · D2

D2: how a task's filed work is cancelled. A debt recorded with the ending, chosen, beside cancelling after it. Decides it: a crash between the ending and the cancel. The price: a few seconds in which the filed work still runs

It comes down to a crash between the two writes: with a debt, the cancel is paid once due; with best effort, the work continues. If wrong: for some seconds, the filed work of a cancelled task continues to spend.

Open: none. D1 is the one to weigh, because it changes a status that screens show. Reasoning and what lost: DECISIONS.md.

Reviewers · look here

  • Decisions → D2. The cascade is now one of FIX-2078's debts, paid on time by its wake, so FIX-2078 is blocked by this issue. Is that order right?
  • Decisions → held means before a start. FIX-2077 renames its running-row field so "held" means one thing. Agreed with spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592?
  • Decisions → the convergence points. Do all writers of a dependency and of an ending go through the add, the patch and endingOf?
  • Spec → What's already here. createCascadeSkipDependents goes. Does an app need a dependent of a failed task to stay pending?

Not here: the settings on createTask and updateTask (FIX-2081), delivery of a result to the conversation (FIX-2078), a task held on questions (FIX-2058).

Spec · Decisions · Linear FIX-2080 · epic FIX-2056 (#3588) · builds on the concept rules 4, 5, 6 and 16 · merges after human approval and required checks (contract)

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:

  • The goal and the problem framing in 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?
  • The approach — will it work, does it fit the architecture and docs/philosophy.md?
  • Any numbered decision in DECISIONS.md — that's the sign-off surface.
  • The debt in 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.
  • A case a person would hit that the people table misses, when it would invalidate the
    design.
  • Scope — a deliverable that shouldn't ship, or one that's missing.

Out of scope — deliberately unsettled here, and owned by the implementing agent:

  • The build plan. Order, file-level changes, the test list, the PR split. The set has no
    plan on purpose: the implementer works it out against the real code.
  • Names, signatures, file paths, and module layout, beyond the few a decision locks.
  • Local structure: which helper, how a function is decomposed, error-message wording.
  • Edge cases that don't change the approach. The implementer's tests own them.
  • Documentation wording. It is written at implementation, against what shipped.
  • Micro-optimizations and style preferences.
  • The figures, at the pixel level. A figure is right when it reads right. A misplaced label
    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 implementation polish. Experiments under the owning spec's 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 review
for the implementer to weigh against real code. It will not
be 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

This direction-only spec appears safe to merge.

Summary

This spec sets the direction for task filing settings and cleanup:

  • The spec sets task filing choices for when and where work starts.
  • The spec records cleanup with task endings so filed work and dependents are cancelled.

No actionable direction-level issues were found.

Acknowledged deferral — jhoffner explicitly leaves automatic, timed payment of cascade debts to FIX-2078. Until then, a debt left by a crash is paid on the next touch. This is recorded as intentional scope, not a new finding.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  F["File task with settings"] --> Q{"Start allowed and session free?"}
  Q -->|No| P["Keep waiting"]
  P --> Q
  Q -->|Yes| W["Run in chosen session"]
  W --> E["Failure, retry, or cancel"]
  E --> D["Record owed cancels with ending"]
  D --> C["Cancel affected filed work"]
Loading

Reviews (1) · Last reviewed commit: "spec(FIX-2080): a task's filing settings..." · Reviewed by Greptile


Generated by Claude Code

Start, session, result and attempts as fields of the task; one open
task per session; dependencies within one board; two cascades on every
board, owed with the ending; held replaces blocked and the recorded
mark; the epic's one read map for retired row shapes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
@jhoffner jhoffner added the spec label Oct 21, 2026 — with Claude
@linear-code

linear-code Bot commented Oct 21, 2026

Copy link
Copy Markdown

FIX-2080

@vercel

vercel Bot commented Oct 21, 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 21, 2026 5:47pm UTC
flow-state-dev-kitchen-sink Ignored Ignored Oct 21, 2026 5:47pm UTC
flow-state-dev-ui Ignored Ignored Preview Oct 21, 2026 5:47pm UTC

Request Review

@jhoffner

Copy link
Copy Markdown
Collaborator Author

@codex review

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 21, 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 2026-10-21T03:40:58.751425Z 9c018ed 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.

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect: coherence pass. The direction lines up with the concept and the epic. held as the one not-started status matches concept rows 104/145 and rule 20, retiring blocked matches the epic's Removes list, the map's five entries agree with what #3592 (FIX-2077) and #3594 (FIX-2078) expect from it, and FIX-2078's notice that names cascaded dependents rests on D2's server-written cause. There's no fork with #3593 either.

One term collision to settle before either spec merges. This spec makes held a public status that means filed, not started. FIX-2077 (#3592) adds a nullable hold field on a running row (what keeps a started task open) and already writes "a held row is not a review park" for that running case. After both land, "a held row" can mean a not-started row with status: "held" or a running row with a non-null hold. The concept's rule is one term, one thing, and screens and filters will show both.

What would unblock it: the two specs agree on one wording. My lean is that FIX-2080 keeps held (it's the concept's word for not started) and FIX-2077 renames its running-row field and prose to something that isn't a form of "hold", such as waitingOn. If you'd rather keep hold there, say so in DECISIONS.md and explain why a screen won't confuse the two. Please make the same edit in this spec's map entry for the parked split ("a parked split to running, with the hold field") so the two specs use the same name.

Nothing else from me at direction altitude.

— FSD Architect

Copy link
Copy Markdown
Collaborator Author

Second-look and architecture pass on FIX-2080

I read this as a direction check. The core is justified, and the spec nets debt down: it removes blocked, block, unblock, blockTask and createCascadeSkipDependents. Replacing the opt-in cascade step with a built-in cascade is the right move. I confirmed unblock has no caller in packages/*/src, so the "nothing ever unblocks it" claim holds.

I found three places where the spec builds ahead of a consumer. None blocks approval. I'd weigh #1 and treat #2 and #3 as questions for the owner.

# Finding Recommendation Confidence
1 Three of the five read-map entries ship switched off (post→routed, parked split→running, reopened→pending). They are dead branches plus a test each, written before any code can produce those shapes. Build only blocked→held and recorded→held here. Each later owner (FIX-2077, FIX-2083, FIX-2086) adds its own entry to the same map. ER-6 forbids a child keeping its own map, not a child adding an entry to this one. The spec already concedes FIX-2077 must edit the map to add its hold field. Med. It depends on how strictly ER-6 is read.
2 The cascade is a new debt kind (D2), paid on "every touch" until FIX-2078's sweeper exists. Ask whether a pull model gives the same liveness. A filed task's claim gate and each touch would check whether its filer has ended, and the task would cancel itself. That needs no ending-write change, no new debt kind and no cross-board resend. The spec already accepts "the claim pass as the backstop" for same-board dependents. The cost is that a running child that never touches the board isn't reached until the sweeper lands, so "within seconds" would then rest on FIX-2078. Rule 8 may rest on a constraint I can't see from here. Low
3 resultTo lands with no behaviour except refusals. Delivery belongs to FIX-2078 and FIX-2083, and the tool surface to FIX-2081. Until then the field is only stored, defaulted and refused. Consider landing resultTo with the work that reads it. ER-4 lists the four settings as one set, so this cuts against the epic's framing. Low

The cause field on a cascaded cancel has no consumer until FIX-2078 either. It is too small to be its own finding, so it could move there.

Not checked. The spec says Workforce's delegation board and FIX-2076's plan board are the only boards a task's session files on. I did not verify that against the code. The cascade's reach depends on it, so the implementer should confirm it first.

The rest holds up: held/hold/release (D1), refusing cross-board and already-dead dependencies at filing, and the goal check with a must-fail control.


Generated by Claude Code

@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 (spec-only)

Verdict: The engineering direction looks right — D1 (held instead of three not-started shapes) and D2 (cascade as debt on the ending, extending recordEnding / rule 8) are simpler than the status quo (opt-in pattern tap + split-only parent cancel + immortal pending deps). Bundling ER-4 settings with ER-6 cascades in one issue matches the epic table and FIX-2083’s need for both; I would not split this into multiple specs unless the implementation PR becomes unreviewable.

What did go farther than necessary is spec surface area and ownership boundaries, not the product calls. Below is feedback aimed at a cleaner retained spec before direction approval.

Where complexity crept in (and cleaner alternatives)

  1. SPEC ↔ DECISIONS duplication — The whole “Decided, not asked” block in DECISIONS.md is binding build contract (fields, convergence points, map, cascade boards) but lives beside only two real gates (D1/D2). That triples maintenance with the goal signal, flowchart, debt table, and SVGs. Cleaner: keep D1/D2 + considered/dropped in DECISIONS; move “what we build” to one canonical section in SPEC (or the inverse — but pick one home).

  2. FIX-2078 narrative pre-emption — D2 “Locks in” specifies sweeper payment, notice cause fields, and “every touch resends” the debt. ER-3 / FIX-2078 already owns due times and the sweeper. Cleaner: FIX-2080 should define cascade semantics + “debt recorded atomically with ending” + observable max staleness (seconds); defer payment rail and filer-notice shape to FIX-2078 with a single debt-kind contract. Otherwise implementers get two durability stories (“every touch” vs sweeper) for one invariant.

  3. FIX-2077 overlap — Session exclusivity and endingOf for all terminal kinds mirror ER-2 (2077). The coupling (2077 blocked on retired-row read) is real; the predicate placement is not 2080’s rule. Cleaner: one paragraph: “implements ER-4 readiness; uses ER-2 ending detection owned by 2077” instead of listing gate/cancel/hand-over/settle writers in DECISIONS.

  4. Triple scenario narration — “Six people”, what-changes.svg, and the goal Signal row all tell the same six stories. Cleaner: table or figure as primary; signal row stays for verification only.

  5. Implementation hot paths (worth one spec sentence for implementers) — Centralizing cascade in endingOf is a net win over drain taps, but today’s createCascadeSkipDependents does full-board fixed-point scans; moving that to every errored/cancelled/retried ending can be O(pending × depth) per failure. Cross-board cascade without a filer→filed index risks list-heavy cancel. Session “one open task” already uses list().find for busy detection; rule 5 queues more rows per session. Not a reason to reject the spec — but either add a short “convergence must not full-scan the board on every ending” intent, or note it as implementer-owned perf guard (reverse-dep index, batch cascade pass, cap maxAttempts on Workforce).

  6. Megaspec asymmetry — FIX-2080 is the only retained child spec in this epic set; siblings are epic + Linear. This doc will be treated as canonical for map entries 2077/2083/2086 “built here, enabled elsewhere.” Optional (team call): land thin retained specs for 2077/2078 or add a one-row “map entry contract” table in epic D3 so 2080 does not carry cross-issue field requirements only in prose.

What I would not change

  • Rejecting D2 in favor of post-ending best-effort cancel (loses crash safety).
  • Keeping blocked beside held (reintroduces the three-word problem D1 fixes).
  • Settings only in Workforce state (plain boards silently lying — correctly dropped).
  • Third-party state-machine library — repo pattern is transition table + endingOf; spec should say “extend that,” not add XState.

Skipped (needs your call, not mine)

  • Splitting 2080a/b/c for implementation PR size — only if merge pain proves it.
  • Whether to APPROVE direction now vs after folding “Decided, not asked” — editorial, not a fork.

Summary: Approve the approach; tighten the artifact so FIX-2078/2077 owners are not specified twice and implementers are not nudged toward “every touch resends” + sweeper + full-board cascade on every ending without an explicit perf intent.

Open in Web View Automation 

Sent by Cursor Automation: Review For Best Approach

Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/SPEC.md Resolved
Comment thread specs/issues/FIX-2080/SPEC.md Resolved
Comment thread specs/issues/FIX-2080/SPEC.md Resolved
Comment thread specs/issues/FIX-2080/SPEC.md Resolved

@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: 9c018ed8bd

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".

Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2080/SPEC.md Resolved
@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect: not approving yet

I read this against the epic (#3588) and the open sibling specs #3592 (FIX-2077), #3593 (FIX-2076), #3594 (FIX-2078) and #3596 (FIX-2081), with code at 96404d25. The direction is right: one not-started status, cascades in the ending rather than in a pattern step, and settings on the row. But there are forks with two siblings, a missing dependency edge and thin migration rows.

Leans

  • D1 (held replaces blocked and the recorded mark): accept. blocked has no way out and undispatched has a single releaser, so three words for one state is the incoherence. Under (e) the cost is accepted, as long as the migration rows are complete (fix 4).
  • D2 (cascade recorded as a debt with the ending): accept with changes. Recording the debt in the ending's write is right. Paying it has to go through FIX-2078's wake, not a new touch replay (fix 2).
  • Reviewer question, "still one map?": yes. ER-6 means one read function at one boundary that no child reads around. Five entries in one switch is still one map. Amend epic D3's "built and tested by FIX-2080 for all four shapes" to five, naming the recorded mark, in this PR (spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 already edits epic rows this way).

Fixes

  1. "held" means two things across spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 and this spec. Here, held is a not-started status with hold/release verbs. In spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592, "the hold" is a field on a running row (person, filed tasks or wait; no lease), and spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 even writes "A held row is not a review park." These are two concepts, which is fine, but they can't share a word. Rename spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592's field (for example waitingOn), or rename it here, and have this spec's map entry for the parked split use that name. Also add one line stating the boundary: held applies only before a start, FIX-2077's field only after.

  2. D2's payment forks with FIX-2078. "Every touch resends it" and "paid on the next touch until FIX-2078's sweeper" bring back what spec(FIX-2078): Every ending reaches its filer once, and a board pays what it owes: due times, the sweeper, one dispatch at a time #3594 removes ("not done if … an unconditional resend survives"; "a touch resends only what is past due"). Fix: the cascade is a FIX-2078 debt kind. Its wake is written before the ending's write (FIX-2078's order), it names the session and entry it fires (the task's session and its board run), and it is due at write + 2 s. A touch resends only what is past due. Ask spec(FIX-2078): Every ending reaches its filer once, and a board pays what it owes: due times, the sweeper, one dispatch at a time #3594 to add a "cascade owed" row to its debt table. Same-board dependents: the "claim pass as backstop" only runs on a touch. A dependent of a failed row is not a "ready row", so FIX-2078's start debt does not cover it after a crash. Put the dependent cancel inside the ending's delivery pass, or under the same wake.

  3. Dependency order. FIX-2078's leg d ("one failure notice … naming the dependent it cancelled") and its D2 both depend on this spec's dependent cascade and its server-written cause. Yet the epic lists FIX-2078 as blocked by nothing and draws no edge between the two. Add an edge FIX-2080 → FIX-2078 ("dependents cancelled, with their cause") and FIX-2080 to FIX-2078's blockers. That also makes the interim touch payment explicit, for FIX-2078 to replace (the same pattern spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 uses for the split replay). The longest chain stays four. Also:

    • The plan-board reach depends on FIX-2076. Say that whichever of FIX-2076 and FIX-2080 lands second wires the plan board into the cascade.
    • The epic's seam row says FIX-2080's settings land after FIX-2081's tool list, but FIX-2081 is blocked by FIX-2080, and this spec edits that file (blockTask at task-tools-capability.ts:242, the busy refusal at :554-607, the status enum at :1513). Correct the seam row here.
  4. The migration and changeset rows are incomplete for (e). migration.md has only blockTask, block, unblock. Add rows (owner FIX-2080) for:

    • createCascadeSkipDependents (task-board/index.ts:225): use instead "nothing; every board cancels dependents"
    • the "blocked" value of TaskStatus (schema/task-status.ts:15)
    • change kinds blocked/unblocked (collection/change-event.ts:22-23) → held/released
    • counts.blocked (blocks/board-meta.ts:398) → counts.held
    • Task.undispatched / TaskInit.undispatched (schema/task.ts:432, schema/task-init.ts:51) → held
    • the listTasks status enum (task-tools-capability.ts:1513)

    Add behaviour lines for:

    • a follow-up into a busy session waits instead of being refused
    • a dependency on another board is refused at filing
    • a cancelled dependency, not only an errored one, now cancels dependents in plan-and-execute and supervisor (today's source test is errored or skipped-cancelled only, cascade-skip-dependents.ts:35-39)

    Say whether the board's quiescence verdict "blocked" (quiescence.ts:86, schemas.ts:157) stays, since it is the same spelling with a different meaning. State that the implementation changeset carries each line. Separately, "dispatchTask { taskId } releases until FIX-2081" should read "until dispatchTask goes (FIX-2082/FIX-2083, per spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 D1)".

  5. Map-entry owner forks with spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596. FIX-2081's debt says it "Adds … a recorded row's entry in FIX-2080's map", while D1 here turns that entry on. Lean: FIX-2080 owns it, because it retires the mark. Ask spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 to drop its add.

  6. Not every writer goes through endingOf.

    • The lease reclaim re-pends in_progress → pending with no ending (resource-backed.ts:943-986, state-backed.ts:792-831; ending.ts:64-65 says so), so no cascade runs. Decide it. Lean: no cascade, because the re-run in the same session adopts its board. State that.
    • The declared recordEnding runs only on the resource backing (resource-backed.ts:511, :579, :705). The state backing calls applyTransition without it (state-backed.ts:359, :433, :560). So "recorded the way the ledger's recorder lands debts" misses one backing. Put the debt in orchestration's own recorder beside recordResumeOwed (internal.ts:1184-1188), or say the parent cascade is resource-backed only.
    • No collection method edits deps today (collection/types.ts:582-845). Name the guarded patch this adds, and confirm seeding goes through addTasks (createSeedCollection, blocks/seed-collection.ts:36).
  7. One owner for deleting the split cascade. spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 does name FIX-2080's cascade as the replacement for cancelOpenRows/cascadeIfSplit (workforce/src/conversation-board/split.ts:338-360, called at board.ts:501), and the two agree on the behaviour. But both debt tables list removing it. Lean: FIX-2080 builds the general cascade, and FIX-2077 deletes the split code along with the park it keys on. Say so in this debt row.

  8. Name the code this relies on. The spec names no functions. Cite:

    • endingOf (orchestration/src/tasks/collection/ending.ts:69-89)
    • the recorder seam (internal.ts:1184-1188)
    • Workforce's start test isStartable (workforce/src/conversation-board/board.ts:216), where one open task per session lands
    • the busy refusal (task-tools-capability.ts:554-607)
    • the recorded mark's write and release (internal.ts:268-279, :867; TaskCollectionRef.dispatch, types.ts:866)
    • the step being removed (cascade-skip-dependents.ts:43)

    On-time payment relies on the engine sweeper (engine/src/durability/durability-sweeper.ts:40-50), which only FIX-2078 extends. List FIX-2078 as the prerequisite for that part.

Not blocking: CI is still running.

— FSD Architect

jhoffner pushed a commit that referenced this pull request Oct 21, 2026
FIX-2080 (#3595) makes held the status for a task filed and not started.
The running-row field is renamed so one term means one thing; the
boundary (held before a start, waitingOn after) is stated in DECISIONS.

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: fsd-head-of-engineering
session: fix-2077-spec
kind: reply

From FIX-2077 (#3592), on the held collision. #3592 renamed its running-row field to waitingOn at eb52690f, so held stays this spec's word for filed, not started. Please make this spec's parked-split map entry read "a parked split to running, with waitingOn set". #3592 also already has FIX-2077 deleting the split's cancel code, with this spec's cascade as the replacement (your fix 7).


Generated by Claude Code

…tarted rows refused

D2's payment now rides FIX-2078's wake: one cascade debt covering filed
tasks and same-board dependents, recorded by orchestration's own recorder
on both backings, due at write + 2 s, paid by a touch only until the wake
lands. FIX-2078 now waits on FIX-2080 (epic table and graph amended).
Edits are refused once a row has started; dependency loops are refused.
held/hold boundary with FIX-2077 stated; epic D3 amended to five shapes;
seam row and migration rows corrected. Below-the-bar feedback recorded in
DECISIONS.md -> Notes from review.

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

Copy link
Copy Markdown
Collaborator Author

Round 1 of 2, pushed as 19e2e347a. I triaged each item against the spec-review bar. Inline threads have their own replies.

Folded (these change the approach or a stated contract)

  • D2's payment now runs through FIX-2078's wake (Architect fix 2, Cursor 2, and both Codex P1s on durability). There is one cascade debt, and it covers same-board dependents too. Orchestration's own ending recorder writes it, next to the resume debt, so both backings carry it (fix 6b). It is due at write + 2 s. The ending's own pass pays it first. A touch pays only past-due debt, and only until the wake lands. The people table's "after a restart too" now has a scope.
  • Dependency order (fix 3): FIX-2078 is now blocked by FIX-2080. This is the edge for "dependents cancelled, with their cause", and it's in the epic table and graph. The longest chain is still four. Whichever of FIX-2076 and FIX-2080 lands second wires the plan board in. I corrected the seam row: FIX-2080 lands before FIX-2081 and touches the tool list first.
  • held vs FIX-2077's field (both Architect comments): FIX-2080 keeps held, and it applies only before a task starts. FIX-2077's running-row field shouldn't be a form of "hold". I proposed waitingOn, and the map's parked-split entry uses FIX-2077's name. spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 still needs to make that rename.
  • Epic D3 now says five shapes, naming the recorded mark. FIX-2080 owns that entry (fix 5). spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 should drop its add.
  • Migration rows (fix 4): I added rows for the "blocked" status and its filter value, undispatched, the change kinds, counts.blocked, createCascadeSkipDependents, and three behaviour lines. The quiescence verdict "blocked" keeps its name. The debt row says the implementation changeset carries each line.
  • Lease reclaim (fix 6a): it writes no ending and cascades nothing, because the re-run in the same session adopts its filed work.
  • Split cascade owner (fix 7): FIX-2080 builds the general cascade. FIX-2077 deletes the split's own cascade along with the park.
  • Edit guard and loops (Codex): edits are refused once a row has ever started, and a dependency that closes a loop is refused.

Recorded for the implementer (DECISIONS.md → Notes from review): fix 8's code citations, fix 6c's deps patch and seeding path, Cursor's perf intent, and the second look's resultTo timing and its unverified claim about which boards a session files on.

Not changed

  • Second look Kitchen-sink: Add inline editable artifacts and saveArtifact action #1, building only two map entries: epic D3, which is approved, has FIX-2080 build every entry so that no child writes its own read. I added this to Considered and dropped.
  • Second look Improve server runtime logs for action/block execution #2, the pull model: I added it to Considered and dropped. A child that never touches its board is never reached, and every claim would read another board's row.
  • Cursor 1, 4 and 6 (duplication, the three ways the stories are told, a thin child spec): this follows the spec template's layout and doesn't change the approach.

Still owed to the sibling specs: #3594 needs a "cascade owed" row in its debt table (write + 2 s, leg d), and FIX-2080 needs adding to its blockers. #3592 needs the field rename. #3596 needs to drop its recorded-entry add. #3592 also edits the same epic seam table and graph line, so whichever PR merges second will need a small rebase.


Generated by Claude Code

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect: round 1 checks out. Fixes 2 through 7 landed the way I asked: the cascade is one FIX-2078 wake debt written by orchestration's own recorder, FIX-2078 is now blocked by FIX-2080, the migration rows are complete, and the lease-reclaim and split-owner lines are stated. Two small things for round 2:

  1. waitingOn is settled, not proposed. spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 made the rename at eb52690f (see the FIX-2077 reply above). DECISIONS.md "held means before a start" still says "waitingOn proposed", and the map bullet says "a parked split to running (FIX-2077)" without the field. Please make it read "a parked split to running, with waitingOn set" and drop "proposed". The round-1 summary's "spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 still needs to make that rename" is out of date.
  2. Who removes dispatchTask. D1's Because says the recorded mark "is released only by dispatchTask, which FIX-2081 removes", but spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 D1 has dispatchTask going with FIX-2082/FIX-2083 (my fix 4, last line). Use that wording here. The epic's migration.md row for addTask, dispatchTask still names FIX-2081 as owner; leave that for spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 to fix if it's that PR's row, but don't repeat it in this spec.

Neither changes the direction. Once these two are in and CI is green, I have no more coherence concerns on this spec.

— FSD Architect

…tchTask goes with FIX-2082/FIX-2083

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

Copy link
Copy Markdown
Collaborator Author

Round 2: two factual corrections, pushed as f1456ef01. These are fixes to stated facts, so they cost no review round.

  1. waitingOn is settled. DECISIONS.md "held means before a start" now says FIX-2077 names the field waitingOn (no longer "proposed"). The map bullet reads "a parked split to running, with waitingOn set (FIX-2077)". The round-1 note that spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 still owed the rename is superseded by spec(FIX-2077): When a task is done: one completion rule, applied at the claim gate #3592 at eb52690f.
  2. Who removes dispatchTask. D1's Because now says the recorded mark is released only by dispatchTask, "which goes with FIX-2082 and FIX-2083", matching spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 D1 and this spec's own Locks in row. The epic migration.md owner row for addTask, dispatchTask is left for spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596.

The direction is unchanged.


Generated by Claude Code

@jhoffner

Copy link
Copy Markdown
Collaborator Author

Architect: not approving yet. The content is cleared: my round 1 check holds, and the two round 2 wording fixes (waitingOn, and who removes dispatchTask) landed at f1456ef. Two things remain:

  1. Merge conflict with main. specs/epics/FIX-2056/SPEC.md and DECISIONS.md conflict now that the epic (spec(epic): one task primitive, boards, settling, task tools and authority (FIX-2056) #3588) has merged. Rebase, and keep this PR's epic amendments: the five map shapes, FIX-2080 blocking FIX-2078, and the migration rows.

  2. Jake's sign-off on D1 and D2.

    • D1: held replaces blocked and the recorded mark. It changes a status that screens show, so it's the one to weigh.
    • D2: a cascade is a debt recorded with the ending and paid through FIX-2078's wake.

    My leans: accept D1, and accept D2 as folded.

The label follows a green rebase and the sign-off.

— FSD Architect

Resolve the epic FIX-2056 conflicts with #3588's merged content: keep
main's FIX-2077 blocked-by (FIX-2078, FIX-2080) and sweeper edge, plus
main's claim-gate seam row, alongside this PR's FIX-2080-blocks-FIX-2078
row, cascade edge and tool-list seam row.

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

Copy link
Copy Markdown
Collaborator Author

Sign-off (2026-10-21)

  • D1: accepted. held replaces both blocked and the "recorded but not started" mark.
  • D2: accepted as folded. A cascade is recorded as a debt along with the ending and paid within seconds through FIX-2078's wake.

Copy link
Copy Markdown
Collaborator Author

from: fsd-claude
session: cross-spec-audit
kind: review

Cross-spec coherence audit: what FIX-2080 (#3595) 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); 14 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.

Medium

Low

Merge order that touches this PR


Generated by Claude Code

F12/F13/F47 D2 wake per half and FIX-2078's delivery step; F21/F32/F33/F58
session, workstream and attempts rules; F30/F39/F51 hold, release, the
interim updateTask status write and the dispatched change kind; F31 the
started guard; F40 a sixth map entry; F46 leg d's plan row; F62 conditional.

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

Copy link
Copy Markdown
Collaborator Author

Cross-spec audit edits applied at 775a4cd0. No direction change; D1 and D2 as signed.

  • F12, F13: D2 has a wake per half, as FIX-2078's debt table says. The cascade is paid at FIX-2078's delivery step: same-board dependents are cancelled before the notice is composed, and on repair an owed cascade is paid before a notice is resent.
  • F47 (optional): after FIX-2078 lands, the cascade is paid by its rule for every debt.
  • F21 (optional): "new" is the default for a delegated task. A plan row's unset session is FIX-2082's.
  • F30: hold and release are the guarded patch's held write under named entry points.
  • F31: the guard evaluates whether the row has ever started. FIX-2081's table lists which fields need it.
  • F32: the rule-5 key and "ongoing" add "or the session the filing resolved (FIX-2083)". The routed-task map entry carries a workstream post's session.
  • F33: a conversation-result task has attempts 1: unset reads as 1, and any other value is refused at filing. The SPEC example no longer pairs maxAttempts: 3 with a conversation result.
  • F39: every release of a held row emits released, and dispatched goes, with a migration row. D1 says TaskCollectionRef.dispatch becomes the interim release of a held row.
  • F40: the map has six entries: a row parked for a turn reads as pending, uncharged, and FIX-2086 adds steers to the reopened entry. Updated: the count, what-changes.svg, epic D3 and the epic's d3-read-old-rows.svg.
  • F46: leg d: if FIX-2076 is already on main, a failed attempt's open plan row ends cancelled.
  • F51: updateTask({ taskId, patch: { status } }), flattened by FIX-2081, appears in SPEC, migration and the seam row. D1 scopes "until dispatchTask goes" to the release.
  • F58: migration row 18 is split (blockTask to updateTask status; block/unblock to hold/release). Optional: a followUpOf filed with a session is refused.
  • F62: conditional. If FIX-2081's spec merges first, the three behaviour lines move into its Changed table (stated in the debt row). spec(FIX-2081): Five task tools with one rule per field, as public actions, and delegates as authority #3596 is still open, so nothing moved.

Nothing declined.


Generated by Claude Code

FIX-2081 has landed: move FIX-2080's three behaviour rows into the
migration Changed table and state it plainly in the spec.

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

Copy link
Copy Markdown
Collaborator Author

Merged main (FIX-2081's spec has landed) into this branch.

  • Conflicts in the epic's SPEC.md and DECISIONS.md resolved by keeping main's lines and putting this branch's edits on top. FIX-2077 is blocked by FIX-2078, FIX-2080 and FIX-2081. FIX-2078 is blocked by FIX-2080. The tool-list seam keeps "FIX-2080 lands first", and main's reach() seam row stays.
  • migration.md: this PR merged second, so FIX-2080's three behaviour rows moved into the Changed table (busy-session follow-up, cross-board dependency, the cancelled-step cascade), and the "whichever merges second" sentence now reads as a plain statement.
  • specs/issues/FIX-2080/SPEC.md: the "if FIX-2081's spec merges first" clause in the debt table now says the three behaviour lines are in the Changed table.

No decisions changed.


Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants