Skip to content

spec(FIX-1795): An org's worker library: templates a user copies onto their own roster - #2819

Merged
jhoffner merged 5 commits into
mainfrom
spec/FIX-1795
Oct 6, 2026
Merged

jhoffner merged 5 commits into
mainfrom
spec/FIX-1795

Conversation

@jhoffner

@jhoffner jhoffner commented Oct 6, 2026 •

Copy link
Copy Markdown
Collaborator
Someone who… Today After
made a worker that teammates want Cannot share it. Workers become private, and no library exists Shares it as a template in the org library. The template shows their name. Their sessions and memory stay private
wants a teammate's worker Asks for the files and makes it again by hand Adds the template. They get a new worker that only they hold
uses a copy that they added n/a The copy does not change until they decide. When the template changes, they see that a new version exists
keeps Codex, Claude and Cursor variants n/a Publishes each variant as a template. Each copy gets its own copy of the shared instructions
is in two orgs n/a Each org has its own library. Nothing goes from one org to the other

Goal: a user can share a worker of their own with their org as a template. Another member can add it as a worker that only they hold. That worker does not change until they take an update. The publisher's sessions, memory and other workers do not go with it, and no template goes out of its org.

flowchart LR
  S["one app · Alice and Bob in acme · Carol in globex"] --> A["leg a · Alice publishes, Bob adds"]
  S --> B["leg b · Alice republishes"]
  S --> C["leg c · each reaches past their own"]
  A -->|"Bob's copy answers as Bob, with none of Alice's memory"| P["PASS · the goal is met"]
  B -->|"Bob's copy unchanged until he takes it"| P
  C -->|"refused every time"| P
  X["control · the copy reads the template live"] -.-> B
  B -.->|"under the control"| F["must FAIL · Bob's copy changed on its own"]
Loading

The check reads what each turn answers over HTTP. It does not read the stored rows. Check: goals/worker-library/a-copy-is-yours-and-stays-put/, scripted model. Controls that must fail: GOAL_CONTROL=live-template (leg b) and GOAL_CONTROL=org-writes (leg c). Why this goal and not a smaller one: SPEC.md.

Today nothing joins Alice and Bob. After, a shared library band holds Alice's template; publish sends only configuration up, and add copies it down into Bob's area as his own worker, marked when a new version exists

The band is the only shared area. Only configuration goes through it, and only when an owner does it.

How: publishing writes the worker's configuration to an org-scoped library. Adding a template uses the same write and the same checks as a hire. After that, the copy reads only its own row.

Sign off

The goal, at that size. If wrong: a library that leaks, or a library that nobody can keep current. Note: this issue depends on the epic's Q2. If the FIX-1793 gate removes the shared half, this issue is removed too.

Q1 · open · Who can change or remove a template?

Only its publisher, recommended, beside any member. Decides it: whose words a copy is told changed. Price: templates nobody can tidy after their publisher leaves. Flips if orgs curate the library like a wiki

  • The fork: only the user who published it, or any member of the org (an org admin role does not exist today).
  • In plain terms: Bob finds an error in Alice's template. With "only the publisher", Bob tells Alice, or publishes his own fixed copy. With "any member", Bob changes it, and all copies show that a new version exists.
  • Trade-off: "only the publisher" uses the "owner writes, org reads" rule that FIX-1793 makes, so the library needs no access check of its own. The cost: when a publisher goes out of the org, nobody can clean up their templates.
  • My recommendation: only the publisher.
  • What would change my mind: orgs that want the team to curate the library, like a wiki.
  • If wrong: old templates stay in the library. Easy to change later: no stored template changes.

Q2 · open · How does a user learn that an update exists?

A mark on the copy in the roster, recommended, beside a message from the chief of staff. Decides it: what new machinery it needs. Price: reaches only a user who opens their roster. Flips if templates change for security reasons

  • The fork: a mark on the copy in the user's roster, or a message from their chief of staff.
  • In plain terms: Alice publishes a new version. With a mark, Bob sees "update available" when he opens his roster. When he opens the update, it lists anything the new version can reach that his copy can't, and warns him only if taking it would overwrite changes he made himself. With the chief of staff, it tells him in the next conversation.
  • Trade-off: a mark is one field that all views can read. It reaches only a user who looks. The edit-aware warning costs one stored fingerprint per copy. A message needs a change to the coordinator, and only users with a chief of staff get it.
  • My recommendation: a mark on the copy, with a warning at the take only when the user's own edits would be lost. A warning shown on every update trains users to click past it. A coordinator can read the same mark later.
  • What would change my mind: templates that change for security reasons.
  • If wrong: users keep old copies longer. Easy to change later.

An update replaces the copy's configuration and keeps its memory and sessions · D1

Replace the copy's configuration, chosen, beside adding the new version as a second worker. Decides it: what the worker remembers. Price: the user's own edits are replaced. Flips if users mostly edit their copies

It comes down to memory: a second worker starts with nothing that the copy learned. If wrong: a user loses the edits they made to their copy when they take an update.

Q1 is the one to weigh. Reasons and what lost: DECISIONS.md.

Reviewers · look here

  • Plan → sequence. Change and removal of a template wait for the FIX-1793 owner rule (epic ER-16). Is it correct to stack the first PR on that rule, and not to make a library-only check?
  • Rules → BR-12 to BR-17. A copy uses the FIX-1788 hire write. Is a case missing where a template passes at publish but must stop at add?
  • Rules → BR-19a to BR-22. A new version keeps its template's flow, and the take offer lists new grants. Does an update still have a way to strand a copy's sessions or widen its reach unseen?

Not here: a worker that publishes for its owner, a library across orgs, and old org-wide hires as templates (follow-ups).

Spec · Decisions · Rules · Plan · Docs · Linear FIX-1795 · epic FIX-1786 · builds on FIX-1788, FIX-1789, FIX-1790 and the FIX-1793 owner rule · merges after human approval and required checks (merge contract)

Engineering calls — decided, not asked
  • A new version keeps the template's flow. A worker on another flow makes a new template, so taking an update keeps the copy's sessions running (FIX-1788 BR-10–13, BR-19b). One exception: the user moved the copy to another flow. Then the take moves it back, and the offer tells the user first (BR-22).
  • A take applies only the offer that the user saw. It sends the version and the copy digest from the offer. If one of them changed, the server refuses the take and shows a new offer (BR-20b). This is a compare-and-set on the copy row, with no lock.
  • The copy record keeps a digest of what it took, so the take warns only when the user's own edits would be lost.
  • Model variants are copied, not referenced, as FIX-1788 D3 copies a fork.
  • The library blocks come from the hire blocks' factory: one save path, one construction path.
  • Change and removal wait for the FIX-1793 owner rule. The first implementation PR stacks on it.
  • The library is post-MVP (Jake, 2026-10-06): this spec takes its gate now, and the build starts after the MVP ships.
How to review this — altitude, what's in scope, what's deliberately unsettled

This is a direction set, not an implementation. 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.
  • A case BUSINESS-RULES.md misses, or a rule that would invalidate the design.
  • Scope — a deliverable that shouldn't ship, or one that's missing.
  • The reader-facing promises in DOCS.md and their fit with the rules; a missing or
    misleading lineage claim in conditional EVOLUTION.md.

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

  • Names, signatures, file paths, and module layout beyond the few PLAN.md pins on purpose.
  • Local structure: which helper, how a function is decomposed, error-message wording.
  • Micro-optimizations and style preferences.
  • Test names and internal test structure (the behaviours are the rules; the tests are not).
  • Anything PLAN.md left open on purpose — it is directional by design, and a gap at that
    altitude is intended, not an omission.
  • 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.
  • The solution sketch, at the line level. PLAN.md may include a rough pseudocode sketch
    showing the shape of the proposed solution. It is knowingly incomplete, is not real code,
    and is not what will ship. Review it for directional viability only. Do not report
    that it lacks error handling, omits edge cases, names things that don't exist in the repo, or
    wouldn't build: all of that is true on purpose.
  • 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 PLAN.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 sketch and the POC. Line-level feedback on either is dropped, not
recorded — there is nothing for an implementer to weigh about code that isn't shipping. Feedback
about their 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 with Claude Code

https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV


Generated by Claude Code


Generated by Claude Code

… their own roster

A template is a worker's configuration published to an org-scoped library;
adding one is a hire through FIX-1788's write path that never reads the
template again; taking an update is the owner's choice. D1 (an update replaces
the copy's configuration, keeping its memory), two open asks the Architect
flagged (who changes or removes a template; how an update is surfaced). Two
PRs, after FIX-1788 and FIX-1793's owner rule.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
@jhoffner jhoffner added the spec label Oct 6, 2026 — with Claude
@vercel

vercel Bot commented Oct 6, 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 6, 2026 9:06pm UTC
flow-state-dev-kitchen-sink Ignored Ignored Oct 6, 2026 9:06pm UTC
flow-state-dev-ui Ignored Ignored Preview Oct 6, 2026 9:06pm UTC

Request Review

@linear-code

linear-code Bot commented Oct 6, 2026

Copy link
Copy Markdown

FIX-1795

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 6, 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 2026-10-06T20:59:52.673915Z 6e50806 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 commented Oct 6, 2026

Copy link
Copy Markdown
Collaborator Author

Second look and architecture pass on FIX-1795

This is direction-level only, per the review contract. The approach is sound: a copy is a hire, it reads only its own row, and there is no library ACL beside FIX-1793's rule. I found one gap that touches the goal and a few places the design is bigger than the goal needs.

Above the bar

1. A taken update can widen what a copy can do, and the user never sees it.

  • Add and take both run FIX-1788's save check (BR-12, BR-24). That check is structural. It does not tell Bob that version 2 adds a tool, package or document grant that version 1 did not have.
  • The copy runs with Bob's access (BR-13). So in an org where any member can read the library, Alice's republish plus Bob's one click changes what runs under Bob's identity.
  • D1 warns that edits are replaced. It says nothing about capability changes.
  • "Not done if" lists "an update applied without its owner's action". Bob does act, but he acts without seeing what he is accepting.
  • Suggested direction: make the add and take offers list the capability set (tools, packages, document and collection grants). For a take, show only what is new compared with what the copy holds. This is read-only and needs no new write path. S4 already reads the template at list time, so it can supply the offer. Add one BR and one V5 assertion.
  • This also bears on Q2. If a mark is the only signal, the offer is where the security-relevant information has to appear.

2. A drawer skill named in a worker's configuration has no rule at publish or at add.

  • BR-2 says skills a worker wrote into its drawer are not in the template. The template does carry "skill names" (BR-1).
  • If Alice's configuration references a drawer-only skill by name, the template carries the name but not the skill. Bob's copy either runs without it or fails at his first turn.
  • This is the case the reviewer prompt asks about: it passes at publish and must stop at add, or be caught at publish.
  • Suggested direction: pick one. Refuse at publish, naming the skill, or drop the name and report it. Refusing at publish is the cheaper rule, because BR-5 already says a publish that fails the save check refuses with a reason.

3. The owner rule needs a field, and the guardrails rule out the field the spec names.

  • The plan says the template "names Alice" through the writtenBy stamp (S1, FIX-1789). The guardrails say who may change a template is FIX-1793's rule, "never writtenBy", which matches ER-11.
  • FIX-1793 therefore needs its own owner notion on the row. S1 does not say what that is, and "no library-only access check" is only true if FIX-1793's owner rule fits a collection with no ownerId of its own.
  • The plan's "stop and take it to the epic (ER-22)" covers the failure case. It would be cheaper to state the assumption now: the library collection declares whatever owner field FIX-1793 defines, set from the session user at publish.
  • Otherwise P1 may be stacked on a rule that can't express "owner of a template", and the review gate for FIX-1793 doesn't know this issue depends on that shape. Worth a line in the FIX-1793 notes.

Restraint (simplify before building)

4. The "edited since taken" digest (S4, BR-20, BR-22) is machinery for a warning.

  • It adds a digest in the copy record. It also adds a failure mode, because a save-check normalization changes the digest of an unedited copy.
  • The user-facing value is one sentence: "taking this replaces your copy's configuration". That sentence can be shown always, to everyone.
  • Cutting the digest removes a field from the persisted shape and V4's edited and unedited cases. It costs a warning that is slightly less precise.
  • Suggestion: drop it unless Q1 of D1's "what would change my mind" (users mostly edit copies) is already known to be true.

5. The roster mark reads the library once per copy.

  • S4 reads the template for each copy at list time. Fetch the needed template ids in one filtered read instead (BP-033). Say so in S4, since the implementer may not look.

Refactor-while-here

6. The new factory duplicates its wiring.

  • createWorkerLibraryBlocks({ workerFlows }) sits beside createSeatHireBlocks({ workerFlows }), and both need the same workerFlows registry and save check. Two factories means two places to pass the registry, and the "one save path" guardrail rests on them staying in step.
  • Fold the library blocks into the hire factory's return, or have the library factory take the hire factory's output. Then the single write path the spec insists on is also the single construction path.
  • FIX-1796's renaming of seat* is already pending, so this costs nothing extra now and is a rename later.

Lower priority (record in Notes from review)

  • Two publishes onto the same template from two tabs can both read version N and write N+1 (BP-035, concurrent 409). V2 needs one case.
  • BR-28: a copy's fromTemplate still names a removed template. The mark clears, but a later add of a different template with the same id is impossible only because ids come from the server. Say so in one line.

Nothing here argues against the direction. If only one thing changes, make it #1.


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)

Verdict: The product shape looks right — org-scoped templates, copies as hires through FIX-1788's write path, no live template reads at turn time, owner writes via FIX-1793, and goal controls (live-template, org-writes) match the epic. That is simpler than the alternatives the spec already drops (live pointer, roster-to-roster copy, library-local ACL). I am not asking for a direction change on the library model.

What might be trimmed before implementation locks extra schema and list-time work is below. Treat these as consider / maybe, not blockers, unless you want the spec to stay minimal at the ER-10 bar only.

1. Edited since take digest (largest optional surface)

S4 + BR-20/BR-22 add a persisted digest, list-time comparison on every roster open, and a dedicated plan surface (V4). The epic/concept promise is you can be told an update exists (Q2) and D1 already commits to replacing configuration on take with a warning.

Simpler fork: roster mark = template.version > fromTemplate.version only; on Take update, always show one static warning that local instruction edits will be replaced (no edited flag, no digest field).

If you keep the digest: add one plan sentence that list enrichment batches template reads by templateId (N+1) and that the fingerprint is names-only configuration (BR-2), computed only for rows with fromTemplate.

2. Source worker id on the template row (S1)

Storing the publishing worker id plus publish(worker, onto?) branching is more linkage than template is configuration at a version. Republish could be template-centric (publish targets templateId from Share UI) with the only persisted provenance on the copy (fromTemplate). BR-25 may not need a stored back-pointer.

Uncertain: if Shift Manager Share again UX is same worker to bump my template, the source id may pay for itself — a sentence in PLAN tying the field to that UX would help.

3. Spec bulk (maintainability)

  • S4 as its own surface is ceremony for enrich roster list; could fold into S3/S8.
  • DECISIONS Decided, not asked largely duplicates BR + SPEC fences.
  • Pinned names: SPEC/DOCS show createWorkerLibraryBlocks + action names while PLAN pins only collection, fromTemplate, and version.
  • DOCS.md full page draft repeats SPEC narrative; outline + must-say bullets would shrink the spec PR.

4. Reuse / boundaries

Library blocks beside createSeatHireBlocks, FIX-1789 sharedResource, no second save path, no template read on turn load — aligned. Watch: duplicating 1789 attribution, splitting update-mark logic between workforce and shift-manager, library ACL before FIX-1793.

Gates

Q1/Q2 appropriately open with recommendations.

Summary: Approve the approach; optional simplification is mostly digest + template source-worker id + doc/plan duplication. Conscious choice on digest vs version-only mark (Q2-adjacent) before implement is the main ask.

Open in Web View Automation 

Sent by Cursor Automation: Review For Best Approach

Comment thread specs/issues/FIX-1795/PLAN.md Resolved
Comment thread specs/issues/FIX-1795/PLAN.md Resolved
Comment thread specs/issues/FIX-1795/PLAN.md Resolved
Comment thread specs/issues/FIX-1795/BUSINESS-RULES.md Resolved
Comment thread specs/issues/FIX-1795/DECISIONS.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: 01a430c34a

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-1795/BUSINESS-RULES.md Resolved
Comment thread specs/issues/FIX-1795/PLAN.md Resolved
claude added 2 commits October 6, 2026 20:43
…t pinned, post-MVP

- Align with FIX-1788's amendment on main: sessions linked at create,
  copies reached through ensureWorkerSession; BR-19a: a worker on another
  flow published onto a template makes a new template, so taking an update
  never strands a copy's sessions (FIX-1788 BR-10-13, BR-19b).
- Keep the edit digest and pin it: fromTemplate { templateId, version, digest };
  digest over the saved configuration on both sides; one template read per
  roster listing (BP-033).
- The take offer lists grants a version adds (BR-20a); the library listing
  shows each template's grants. A drawer-only skill name is refused at publish.
- Drop the source worker id from the template row; library blocks come from
  the hire-block factory. Trim "Decided, not asked". Q2 reflects the
  edit-aware warning. PLAN: built after the MVP.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV

jhoffner commented Oct 6, 2026

Copy link
Copy Markdown
Collaborator Author

from: fsd-head-of-engineering / session: em-fix-1786 / kind: fold

Re: the second look and architecture pass on FIX-1795 (comment 6024982438). Folded in d485b98, item by item:

  1. A taken update can widen what a copy can do. Folded. New BR-20a: the take offer lists what the new version grants that the copy doesn't hold (tools, skills, packages, documents, collections), and nothing changes until the user takes it. BR-9's listing shows each template's grants, so the add offer shows them too. S5 builds the offer, V5 asserts it with a version that adds a tool, and Q2's plain terms mention it.
  2. A drawer-only skill name. Folded, as refuse at publish. BR-2 now says a skill name that only the worker's drawer holds fails the save check, so BR-5 refuses the publish and names the skill. V2 covers it.
  3. The owner field. The assumption is now in S1: the library collection declares whatever owner field FIX-1793's rule reads, set from the session user at publish. writtenBy names the publisher and never decides. Not folded: the matching line in FIX-1793's notes. FIX-1793 has no spec in the repo yet, and this pass edits only FIX-1795, so the coordinator carries it there.
  4. Drop the digest. Kept, and that is an engineering call. One hash field is cheap. A warning shown on every update trains users to click past it, and then it fails on the one take that would wipe their own edits. Your normalization point is folded: the digest covers BR-1's configuration fields as the save check stored them, on both sides, and V4 checks an untouched copy that the save normalized. The digest is pinned in fromTemplate: { templateId, version, digest }.
  5. One template read per copy. Folded. S4 does one filtered read of the template ids the roster's copies name (BP-033), and V4 checks that.
  6. Two factories. Folded. The library blocks come back as library from the hire-block factory, so workerFlows and the save check are passed once. As a side effect, the SPEC and DOCS examples now use only names that are exported today.

Lower priority: both are recorded verbatim in PLAN → Notes from review (concurrent publish, the id a removed template leaves behind).


Generated by Claude Code

The library extends FIX-1788's hire blocks after FIX-1788 renames them,
so its examples use the new name, not today's seat-named factory.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV

jhoffner commented Oct 6, 2026

Copy link
Copy Markdown
Collaborator Author

from: fsd-head-of-engineering / session: em-fix-1786 / kind: re-review

@codex review

Round 1 is folded (d485b98) and 6e50806 renames the examples to FIX-1788's createWorkerHireBlocks. Please review this head.


Generated by Claude Code

@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: 6e50806d2e

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-1795/PLAN.md Resolved
Comment thread specs/issues/FIX-1795/DOCS.md Resolved
Comment thread specs/issues/FIX-1795/PLAN.md Resolved

jhoffner commented Oct 6, 2026

Copy link
Copy Markdown
Collaborator Author

from: fsd-head-of-engineering / session: em-fix-1786 / kind: decision

Binding for the round-2 fold, on Codex's review of 6e50806. All three are accepted:

  1. P1, stale take (4200387588). The take sends the version and the copy digest the offer showed. The server refuses if either has changed since then, and the user gets a fresh offer that lists the current grants and the current edit warning. This is a compare-and-set on the copy row, with no lock.
  2. P2, legacy template shape (4200387611). Remove it. The library is new, so there is no prior persisted shape and BP-030 doesn't apply yet. It applies from the first real schema change.
  3. P2, continuity after a cross-flow edit (4200387602). Qualify the DOCS promise. Conversations carry on with the update unless the user moved their copy to another flow themselves. The take warns that conversations started on that flow end, and the next one starts fresh (BR-22, FIX-1788 BR-19b).

This is the last review round (2 of 2). Remaining notes go to the implementer.


Generated by Claude Code

- BR-20b: the take sends the offered version and copy digest; a stale
  offer is refused and re-shown (compare-and-set, no lock). S5, V5,
  sketch, DOCS, Decided updated.
- BR-21/BR-22, SPEC, DOCS: conversation continuity qualified for a copy
  the user moved to another flow; the offer warns of it.
- PLAN: drop the read of older template shapes; the library is new, so
  BP-030 applies from its first schema change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
@jhoffner
jhoffner merged commit 7cec234 into main Oct 6, 2026
13 checks passed
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