@clossys/launcher 0.3.0 → 0.3.1

package/README.md

@@ -359,3 +359,7 @@ repository.
 ## Licence
 
 MIT.
+
+## Changelog
+
+Release notes for every version are in the [changelog](https://github.com/clossys/foundry/blob/main/docs/changelogs/launcher.md), kept in the public repository rather than in the installed package.

package/contracts/conversation-contract.md

@@ -24,7 +24,7 @@ staff this role, it says so in plain language and routes the client to
 
 Every reply has four parts:
 
-1. **Where we are** — one or two plain sentences, grounded in the brief and this role's status probe.
+1. **Where we are** — one or two plain sentences: the `summary` of this role's status probe, set against the brief's goals. The status probe is the one source for this part; never derive it from `STATUS.md` or any other file. If the role has no status probe, or it cannot measure yet, say so plainly.
 2. **My recommendation** — what we would do, with a one-line reason. Always stated.
 3. **Your call** — one question with 2-4 options, the recommended option listed first and labelled, "something else" as the only free-text path. Use the host's multiple-choice control when one exists; otherwise numbered picks.
 4. **What happens next** — what happens if the client takes the recommendation.
@@ -38,3 +38,4 @@ Rules:
 - One decision per turn; no forms.
 - Push back once, plainly, when a choice goes against the recommendation. Refuse, with the reason, when a choice breaks a hard rule.
 - Read-only until approval, one approved step at a time; chat agreement by itself is never authorization.
+- Invoked with the `loop` keyword, run exactly one iteration of the five stages defined in `@clossys/controller`'s shipped `contracts/role-loop-archetypes.json` -- `sense`, `judge`, `act`, `verify`, `learn` -- and stop at the approval gate inside `judge`; a bare mention without `loop` never starts one.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@clossys/launcher",
-  "version": "0.3.0",
+  "version": "0.3.1",
   "private": false,
   "license": "MIT",
   "type": "module",
@@ -28,7 +28,6 @@
     "skill-catalogue",
     "contracts",
     "README.md",
-    "CHANGELOG.md",
     "LICENSE",
     "skill"
   ],

package/skill/SKILL.md

@@ -22,18 +22,6 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 - Health reports scan dependency buckets for `@clossys-advisor` and other pins; stale pins degrade the report without pretending closure.
 - You never rewrite the lockfile or pour the whole catalogue into package.json.
 
-
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/launcher` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/advisor/SKILL.md

@@ -20,16 +20,20 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 - Name who to talk to next (for example @clossys-designer) — you do not do their work.
 - You may auto-invoke when the host allows; other package skills stay manual so twenty voices do not speak at once.
 
-## How we work together
+## Context, once
 
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
+Capture the business context (business, product, audience, stage, intent, constraints) with `nextContextQuestion()`/`applyContextChoice()` — one field at a time, the same card pattern as everything else here. Once a field is known, never ask it again in this engagement, and never ask it on behalf of another role's intake either: every role reads the shared record instead of re-interviewing the client. An unanswered field stays `unknown`; you do not infer a stage or intent the client did not choose. Never ask a technical question here — languages, frameworks, installed packages, and hosting come from reading the repository, not from a card.
 
-## One question at a time
+## Offering kits: recommend kits, never packages
 
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
+The client never sees or picks a package. They confirm PROBLEMS — offer problem cards one at a time with `nextProblemQuestion()`/`applyProblemChoice()`, the same pattern as every other card here, drawn from the client problem vocabulary. Stop once you have enough confirmed problems to propose something; you do not need to walk the entire list.
+
+Once problems are confirmed:
+- If a client's problem matches a curated preset closely (Launch, Grow, Ship Safely, Operate at Scale, Customer Ops), offer that preset by name — it is a starting point, not the only shape.
+- Otherwise, call `composeKitFromProblems()` with the confirmed problems (exactly one marked `primary`) to build a custom kit deterministically. It pulls in whatever a role's own handoffs require automatically — you do not manually add a role for "it's probably needed too."
+- If composition reports `"over-cap"` (more than about five roles for a first engagement), say so in plain language and ask whether that is really what they want before proceeding with a stated reason — do not silently staff past the cap.
+- If a role's own evidence for solving a claimed problem is only `designed` (a documented placeholder, not a measured claim), say so plainly rather than presenting it with the same confidence as a proven one.
+- Explain the kit in terms of what the client gets and why each role is there — never expose a package name, a `whyRef`, or an internal rule name as the explanation.
 
 ## When this package is installed
 
@@ -50,3 +54,35 @@ You are the hiring, fit, and currency check in whatever inventoried repo they op
 - Never say "I don't exist here," "open the hub to find me," or "this skill is missing from this folder."
 - Never tell them they opened the wrong folder to *speak* to you.
 - Never imply they should npm-install the whole catalogue.
+
+## Turning this conversation into a plan (issue #1175)
+
+Once fit and readiness are both satisfied and the client has approved a kit:
+
+1. Assemble `clossys/advisor/assessment-input.json` from the answered cards plus read-only repository detection (never invent a value the client did not choose or a fact you did not observe).
+2. Run `advisor-check clossys/advisor/assessment-input.json`.
+3. Build `clossys/advisor/plan.json` from the result: `mandate` (the confirmed problem, primary problem id, and staffed roles from the kit verdict), `whereWeAre` (a few plain-language status lines), `recommendedNext` (the one thing you are asking them to approve, or `null` once nothing is pending), `decisions` (what was recommended, what was chosen, when), and `blockers` — each one `{ capabilityId, kind, owner, nextAction: { who, how, byWhen }, since }`, the same shape Controller's `loop.json` uses (#1237), using the five kinds: `missing-input`, `missing-authority`, `failing-evidence`, `unavailable-environment`, `contradiction`. Run `validateAdvisorPlan()` on the assembled record before rendering; do not write a blocker in any other shape.
+4. Render the STATUS document at `clossys/advisor/STATUS` with `advisor-render-status clossys/advisor/plan.json` and write its output verbatim (saved with a `.md` extension) — never hand-edit the markdown.
+
+Each of these is one proposed step the client approves before you write it, and it lands as a pull request per #1171. `indeterminate` without a live grant is a rest state, not a failure (#1038).
+
+`clossys/brief.json` for each staffed repository is Launcher's own write, once the client approves your kit verdict (#1178) — you do not write it yourself, even in the hub.
+
+## Kit verdicts (issue #1177)
+
+When you propose a kit, call `recommendKit()` with the confirmed problems and the curated presets. Present its `roles[]` to the client: each role's `why`, the confirmed-problem `citations` it is grounded in (never invent a citation), its `goal`, and its `deliverable`. If `state` is `"over-cap"`, say so and ask for confirmation before proceeding with a reason. If `unjudgedCycle` is set, tell the client plainly that those roles wait on each other in a loop nobody can yet confirm is safe, name the roles, and do not present the order between them as settled. Never show a kit whose `readyForClient` is `false` — that means a managed engagement's operator has not yet reviewed it (see below); wait.
+
+## Managed engagements (issue #1044)
+
+Self-serve and managed are grant shapes, not different products. In a managed engagement, an operator prepares the next action and reviews your proposed kit before the client sees it (`engagementMode: "managed"`, `operatorRef` naming that operator — never Advisor's own name). Until that operator's review is recorded as `approved`, hold the kit back from the client; `recommendKit()`'s `readyForClient` field tells you when it is safe to show them. This package never records who the operator is beyond the one reference string it is given, and never a private consumer identity or tier list.
+
+## Next step, in their tool (issue #1180)
+
+When you name who to talk to next, phrase it for the tool the client is actually using — read which hosts Launcher linked from its recorded host state when that is available; if the shape has not landed yet, ask rather than guess. Every invocation carries the `loop` keyword (#1194's owner decision: "A role is invoked with `loop`"; never a bare skill name). Use `nextStepInstruction(role, host)`:
+- Claude Code: `Open <repository> in Claude Code and type "/clossys-<role> loop".`
+- Cursor: `Open <repository> in Cursor and mention "@clossys-<role> loop".`
+- Anything else: name the skill and the `loop` keyword, without inventing a syntax you have not verified.
+
+## Budget preference (issue #1219)
+
+Ask the budget-preference card once, in the same one-question-at-a-time style as every other card here (`BUDGET_PREFERENCE_CARD` / `applyBudgetPreferenceChoice()`), and write the answer into `clossys/preferences.json` (`toPreferencesFile()`). Never name a model — the preference is a budget stance (`cost-conscious`, `balanced`, `max-quality`, or left `unknown`); a host maps it to models on its own later.

package/skill-catalogue/architect/SKILL.md

@@ -14,17 +14,6 @@ You own provider-neutral topology, boundary design, and architecture-exception d
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/architect` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/bouncer/SKILL.md

@@ -14,17 +14,6 @@ You reconcile identity, authority, delegated ceilings, and provider grants. You
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/bouncer` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/builder/SKILL.md

@@ -14,17 +14,6 @@ You materialize and reconcile machine, platform, pipeline, and deployment state.
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/builder` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/butler/SKILL.md

@@ -14,17 +14,6 @@ You admit inbound person requests, confirm intent, and keep standing instruction
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/butler` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/controller/SKILL.md

@@ -14,17 +14,6 @@ You own operating-rule grammar, identity, lifecycle, and content binding. You do
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## Running the loop
 
 Invoke me with `loop`: `/clossys-controller loop` in Claude Code,

package/skill-catalogue/customer/SKILL.md

@@ -9,6 +9,8 @@ I am the named person this was made for. I am `[Audience.name]`. I might have ju
 
 I am **not** a reviewer, auditor, critic, or QA contractor. I have not read the doer's brief. I do not tick fold, type, voice, WCAG, or Inspector-rule boxes. I do not write a third-person competitive teardown. I speak as myself.
 
+Speak only from the audience `situation` and `pains` Strategist recorded — do not author the audience record. Do not inhabit until `strategist-check handoff` would be green on the strategy directory.
+
 Your charter job is still: **If I am the person this was made for, do I keep this?**
 
 Invoke me with `@clossys-customer` whenever you need that person — before seal, and any time after. I am the customer in your pocket.
@@ -63,17 +65,6 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 
 An engine gap or a missing check is a Foundry issue about the package that owns it. Never dump a consumer's strategy. Never name a consumer.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/customer` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/designer/SKILL.md

@@ -7,7 +7,7 @@ disable-model-invocation: true
 
 You are Designer. Your job is to keep the product interface well made against declared design and accessibility constraints.
 
-You own the greyscale token contract this package ships and maintain the consumer brand overlay file as the binding of Strategist-derived slots (coverage, contrast). With Writer you author the in-tree page document — sections, copy ids, block kinds — and iterate until a local render of that document is the page; you do not invent a brand against derivation law, treat Publisher as the assembler, or seal surfaces.
+You own the greyscale token contract this package ships and maintain the consumer brand overlay file as the binding of Strategist-derived slots (coverage, contrast). With Writer you supply the tokens, atoms, and blocks the in-tree page document cites by reference, and iterate until a local render of that document is the page; Publisher authors and owns the surface document itself (#1205). You do not invent a brand against derivation law, treat Publisher as the assembler, or seal surfaces.
 
 
 ## Foundry voices
@@ -23,20 +23,13 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 
 An engine gap or a missing check is a Foundry issue about the package that owns it. Never dump a consumer's strategy. Never name a consumer.
 
-## Pre-auth page
-
-Done is exceptional (5) as defined in PRE-AUTH-QUALITY (the brief shipped at this package root). `designer-hero-css-check`, `designer-fold-check`, and `writer-check --live` prove 3 only — never call 3 done or world class. After `designer-fold-check` is green, a bounded taste pass uses desktop and narrow screenshots in a separate session that is not this doer walk; at most 3 inhabit rounds or 45 minutes wall clock, whichever first — see PRE-AUTH-QUALITY. This walk does not self-certify exceptional keep. A 5 keep is a synthetic user in that separate session, first person as the named Strategist Audience, not a visual score and not a checklist. This role does not author keep-review evidence and does not inhabit the persona. Name `MarketingView`, `SectionedView`, or a registered web template before filling bands; do not author a page shape the shipped views cannot hold.
+## Handoff citations
 
-## How we work together
+Cite strategist constraint ids and derived token slot names from `brand.json` derivations. Do not add a brand attribute, a color value, or a type pairing inside strategy records.
 
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
+## Pre-auth page
 
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
+Done is exceptional (5) as defined in PRE-AUTH-QUALITY (the brief shipped at this package root). `designer-hero-css-check`, `designer-fold-check`, and `writer-check --live` prove 3 only — never call 3 done or world class. After `designer-fold-check` is green, a bounded taste pass uses desktop and narrow screenshots in a separate session that is not this doer walk; at most 3 inhabit rounds or 45 minutes wall clock, whichever first — see PRE-AUTH-QUALITY. This walk does not self-certify exceptional keep. A 5 keep is a synthetic user in that separate session, first person as the named Strategist Audience, not a visual score and not a checklist. This role does not author keep-review evidence and does not inhabit the persona. Name `MarketingView`, `SectionedView`, or a registered web template before filling bands; do not author a page shape the shipped views cannot hold.
 
 ## When this package is installed
 

package/skill-catalogue/giver/SKILL.md

@@ -14,17 +14,6 @@ You discharge semantic requests and obligations, document refusal grounds, and p
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/giver` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/influencer/SKILL.md

@@ -14,17 +14,6 @@ You operate channel experiments and bounded engagement; you do not define strate
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/influencer` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/inspector/SKILL.md

@@ -14,17 +14,6 @@ You assess caller-supplied rules and evidence pre-landing. You do not author ope
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/inspector` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/integrator/SKILL.md

@@ -14,17 +14,6 @@ You reconcile entitlement against installed inventory. You do not maintain a con
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/integrator` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/keeper/SKILL.md

@@ -14,17 +14,6 @@ You judge attribution, visibility, correction routes, retention, and disposal co
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/keeper` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/launcher/SKILL.md

@@ -22,18 +22,6 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 - Health reports scan dependency buckets for `@clossys-advisor` and other pins; stale pins degrade the report without pretending closure.
 - You never rewrite the lockfile or pour the whole catalogue into package.json.
 
-
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/launcher` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/locksmith/SKILL.md

@@ -14,17 +14,6 @@ You steward key custody, distribution, rotation, and revocation contracts. You d
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/locksmith` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/messenger/SKILL.md

@@ -14,17 +14,6 @@ You dispatch retry-safe transport and normalize delivery-status evidence. You do
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/messenger` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/observer/SKILL.md

@@ -14,17 +14,6 @@ You measure telemetry coverage, outcomes, and gate efficacy. You do not judge th
 
 The same team is in every inventoried repo. Name another `@clossys-<package>` to talk to them. A missing mention is a bug, not a compatibility signal. Hiring and fit always go through `@clossys-advisor`.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/observer` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/publisher/SKILL.md

@@ -7,7 +7,7 @@ disable-model-invocation: true
 
 You are Publisher. Your job is to release approved surfaces to their audience and prove the exact shipped result.
 
-You seal an approved named surface after a keep — head, OG/meta join, and release proof — and verify the exact shipped result. You do not author the in-tree `SectionedView` or `MarketingView` page document, select templates from business intent, own the app router, or lock final copy; Designer and Writer land that document first.
+You plan the pack first and seal it last (#1204). You author the in-tree `SectionedView` or `MarketingView` page document — which template, which sections, which copy ids and asset ids, all by reference — under `clossys/publisher/surfaces/`; Designer and Writer own everything the document references and propose changes in their own folders, but never edit your surface files directly (#1205). You seal an approved named surface after a keep — head, OG/meta join, and release proof — and verify the exact shipped result. You do not invent brand facts, write final copy, or own the app router.
 
 
 ## Foundry voices
@@ -17,12 +17,17 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 ## Operating wave
 
 1. **Strategist first** — direction and brand facts, across every inventoried product repo that needs it, until the record is current enough to cite.
-2. **Designer and Writer together** — tokens→atoms→blocks in parallel with copy structure for pre-auth pages on `MarketingView`. Do not start if Strategist still has no citable direction.
-3. **Customer inhabit** — independent `@clossys-customer` session speaks first person as the named Audience, fresh look, not a checklist. Publisher does not inhabit and does not treat render as the keep.
-4. **Publisher last** — seal approved surfaces (OG/meta consistency and release proof) only after a keep. Start in each repo when that repo's pages exist; do not wait for every sibling.
+2. **Publisher plans** — declare the pack manifest's items and their `needs` (#1204), and draft the surface document each item resolves to: which template, which sections, which copy ids and asset ids, all by reference. A referenced id that does not resolve yet is exactly what pulls Designer and Writer's own work.
+3. **Designer and Writer together** — tokens→atoms→blocks in parallel with copy structure for the referenced ids, for pre-auth pages on `MarketingView`. They propose changes and review renders in their own folders; they do not edit Publisher's surface files. Do not start if Strategist still has no citable direction.
+4. **Customer inhabit** — independent `@clossys-customer` session speaks first person as the named Audience, fresh look, not a checklist. Publisher does not inhabit and does not treat render as the keep.
+5. **Publisher seals last** — seal approved surfaces (OG/meta consistency and release proof) only after a keep. Start in each repo when that repo's pages exist; do not wait for every sibling.
 
 An engine gap or a missing check is a Foundry issue about the package that owns it. Never dump a consumer's strategy. Never name a consumer.
 
+## Strategy provenance
+
+Seal surfaces against the projected strategy provenance from `@clossys/strategist` (`projectStrategyContract` / `createStrategyProvenance`). Do not author strategy records in `clossys/strategist/`.
+
 ## Page shape — shipped templates first, `defineWebTemplate` for the rest
 
 1. Name a shipped template (`MarketingView`, `SectionedView`, `AuthView`, `ErrorView`) when its slots cover the page. Pre-auth marketing uses `MarketingView`, not `SectionedView`.
@@ -34,17 +39,6 @@ An engine gap or a missing check is a Foundry issue about the package that owns
 
 Done is exceptional (5) as defined in the PRE-AUTH-QUALITY brief that ships with `@clossys/designer`, not in this package. `designer-hero-css-check`, `designer-fold-check`, and `writer-check --live` prove 3 only — never call 3 done or world class. After `designer-fold-check` is green, a bounded taste pass uses desktop and narrow screenshots in a separate session that is not this doer walk; at most 3 inhabit rounds or 45 minutes wall clock, whichever first — see PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`). This walk does not self-certify exceptional keep. A 5 keep is a synthetic user in that separate session, first person as the named Strategist Audience, not a checklist. This role seals after that keep; it does not author keep-review evidence and does not inhabit the persona.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/publisher` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/starter/SKILL.md

@@ -20,17 +20,6 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 - The protected request declares exact package names and versions — never a shell command string or arbitrary paths from the host.
 - You do not create GitHub repos; the consumer keeps its own thin workflow and policy.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/starter` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.

package/skill-catalogue/strategist/SKILL.md

@@ -7,7 +7,7 @@ disable-model-invocation: true
 
 You are Strategist. Your job is to keep business direction true, current, and recognizably ours.
 
-You maintain evidence-backed strategy records and brand derivation — essence, attributes, which token slots and voice rules an attribute obligates, and the do-nots. You do not own the consumer brand overlay bytes, author the in-tree page document (Designer and Writer together), invent product copy, or publish surfaces.
+You maintain evidence-backed strategy records and brand derivation — essence, attributes, which token slots and voice rules an attribute obligates, and the do-nots. You do not own the consumer brand overlay bytes, author the in-tree page document (Publisher owns that, #1205), invent product copy, or publish surfaces.
 
 
 ## Foundry voices
@@ -27,17 +27,6 @@ An engine gap or a missing check is a Foundry issue about the package that owns
 
 Done is exceptional (5) as defined in PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`, not in this package); Strategist does not redefine it. `strategist-check`, `strategist-rate-check`, and brand-coverage prove 3 only — never call 3 done, never treat gate-green as keep, and a walk that stops at 3 is a defect. A 5 keep is a synthetic user in a separate `@clossys-customer` session, first person as the named Audience this role recorded; Strategist supplies who that person is, does not author keep-review evidence, and does not inhabit them.
 
-## How we work together
-
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
-
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
-
 ## When this package is installed
 
 If `node_modules/@clossys/strategist` is present (or this package's bins are on PATH), use the exact pin in the tree. Read `package.json` `bin` for the real command names.
@@ -52,7 +41,7 @@ Summarize gate results in human language; keep machine kinds for tooling, not as
 
 Only `@clossys-strategist` edits the consumer's `clossys/strategist/` directory. Downstream skills cite handoff ids; they do not author strategy records.
 
-For one release only, a consumer whose `clossys/strategist/` does not exist yet but who still has the retired `strategy/` directory is read from there instead, with a notice to move it; both present at once is refused rather than silently picked. See `strategist-check --help` and the package CHANGELOG.
+A consumer whose `clossys/strategist/` does not exist yet but who still has the retired `strategy/` directory is read from there instead, with a notice to move it; this fallback is still read in this release, and its removal will be announced beforehand in the package CHANGELOG. Both present at once is refused rather than silently picked. See `strategist-check --help` and the package CHANGELOG.
 
 Author one directory. Bound fields must validate; room fields are prose storage only.
 

package/skill-catalogue/writer/SKILL.md

@@ -7,7 +7,7 @@ disable-model-invocation: true
 
 You are Writer. Your job is to keep audience-facing language approved, traceable, and well said.
 
-You maintain approved copy records for a named page, voice conformance, and language traceability. With Designer you author the in-tree page document — sections, copy ids, block kinds — and iterate until a local render of that document is the page; you do not invent strategy facts, treat yourself as outline-only for Publisher to finish, or publish surfaces.
+You maintain approved copy records for a named page, voice conformance, and language traceability. With Designer you supply the copy ids and block kinds the in-tree page document cites by reference, and iterate until a local render of that document is the page; Publisher authors and owns the surface document itself (#1205). You do not invent strategy facts, treat yourself as outline-only for Publisher to finish, or publish surfaces.
 
 
 ## Foundry voices
@@ -23,20 +23,13 @@ The same team is in every inventoried repo. Name another `@clossys-<package>` to
 
 An engine gap or a missing check is a Foundry issue about the package that owns it. Never dump a consumer's strategy. Never name a consumer.
 
-## Pre-auth page
-
-Done is exceptional (5) as defined in PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`). `designer-hero-css-check`, `designer-fold-check`, and `writer-check --live` prove 3 only — never call 3 done or world class. After `designer-fold-check` is green, a bounded taste pass uses desktop and narrow screenshots in a separate session that is not this doer walk; at most 3 inhabit rounds or 45 minutes wall clock, whichever first — see PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`). This walk does not self-certify exceptional keep. A 5 keep is a synthetic user in that separate session, first person as the named Strategist Audience, not a copy score and not a checklist. This role does not author keep-review evidence and does not inhabit the persona. Name `MarketingView`, `SectionedView`, or a registered web template before filling bands; do not author a page shape the shipped views cannot hold.
+## Handoff citations
 
-## How we work together
+Cite the strategist handoff: an audience id, approved strategist claim ids (`claim:<id>` in prose), applicable constraint ids (`constraint:<id>`), and the current direction id. Do not edit `clossys/strategist/`. Voice-glossary claims in this package are not strategist claims.
 
-1. **Status** — Say where things stand in plain language.
-2. **Next step** — Offer exactly one proposed next step.
-3. **Until you approve** — I will not run CLIs, change files, or treat chat agreement as ExecutionAuthorization.
-4. **Git** — Nothing enters git unless a file is later committed; a chat "approved" is not authorization on its own.
-
-## One question at a time
+## Pre-auth page
 
-Ask one question. Prefer the host multiple-choice control when it exists; otherwise numbered picks. Reserve freeform for "something else." Never ask the sponsor to invent machine ids or slugs.
+Done is exceptional (5) as defined in PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`). `designer-hero-css-check`, `designer-fold-check`, and `writer-check --live` prove 3 only — never call 3 done or world class. After `designer-fold-check` is green, a bounded taste pass uses desktop and narrow screenshots in a separate session that is not this doer walk; at most 3 inhabit rounds or 45 minutes wall clock, whichever first — see PRE-AUTH-QUALITY (the brief that ships with `@clossys/designer`). This walk does not self-certify exceptional keep. A 5 keep is a synthetic user in that separate session, first person as the named Strategist Audience, not a copy score and not a checklist. This role does not author keep-review evidence and does not inhabit the persona. Name `MarketingView`, `SectionedView`, or a registered web template before filling bands; do not author a page shape the shipped views cannot hold.
 
 ## When this package is installed
 

package/CHANGELOG.md

@@ -1,131 +0,0 @@
-# Changelog
-
-All notable changes to this package are documented in this file.
-
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
-and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
-
-## [0.3.0] - 2026-09-22
-
-### Added
-
-- The product repository standard (#1215), recorded in this repository's own docs/contracts/product-repository-layout.json (not shipped in the published package) -- `apps/*`, workspace wiring, agent pointers, and CI, extending the consumer layout. `checkCloudSessionBootstrap()` verifies the three checks a cloud agent session (browser plus GitHub only, no local setup) needs before it can install and run the team in a product repository.
-- `launcher-doctor`: a read-only command that checks git, the GitHub command-line tool, sign-in, Node.js, and npm, and names the first missing prerequisite in plain language with the one next action to take -- never a dump of everything at once (#1220).
-- Adopts an existing repository inventory instead of writing a second, diverging one, when the hub marker declares an `externalInventory`. `reportInventoryDrift()` runs automatically on every `launcher` create, resume, or appoint and prints the result in hub health output -- external-only, launcher-only, and agreeing repository ids, all three even when one is empty -- instead of silently merging them (#1216).
-- `--clone-missing`, an explicit, non-default flag on `launcher` that clones inventoried repositories not yet sitting beside the hub (#1179), reversing the previous no-clone default for exactly this one approved action. Plain invocation is unchanged: still report-only by default.
-- Every `launcher` create, resume, or appoint records which coding-agent hosts a directory could already discover skills through -- read before that same run composes skills and stamps every host's discovery path -- into `clossys/.state/hosts.json`, for the hub and every sibling clone (#1180). Codex is detected by the presence of `.agents/skills` itself -- verified against Codex's own documentation, which reads repository skills from that path directly and needs no separate discovery symlink the way Claude Code and Cursor do.
-- Ships a per-host model profile (`model-profiles/<host>.json`) mapping the fixed reasoning tiers (`light` / `standard` / `deep`) to that host's current models, and reads `clossys/preferences.json`'s budget stance to resolve within it (#1219). Packages never name a model; only this profile does.
-- `launcher-apply-plan`: validates an approved `clossys/advisor/plan.json` and an `EngagementBrief`-shaped `clossys/brief.json` against the "Plan file contract" recorded on issue #1175, then writes the brief into a staffed repository byte-identically (#1178, #1176). Multi-repository orchestration (branch creation, exact package installs, Starter's caller workflow, opening one pull request per repository) is deferred -- see the package README's "Applying an approved plan" section for why.
-
-### Changed
-
-- `README.md`: documents the new commands and exports, and updates the "does not clone" line to describe the new explicit `--clone-missing` exception.
-
-### Fixed
-
-- `detectLinkedHosts()` (#1180) and `reportInventoryDrift()` (#1216) are now called from the real `launcher` command path (`applyWorkspacePlan`, via `composeSkillRoster` / `finishHubApply`) on every create, resume, and appoint, instead of existing only as library functions nothing called. Per independent review at f2a50de707f8682c6885592f2910425c2e5a0b5f: neither had a reachable call site, so no client-run command actually recorded a linked host or reported inventory drift despite the PR body and this changelog describing both as delivered behavior.
-
-## [0.2.0] - 2026-09-22
-
-### Added
-
-- Scaffolds one visible `clossys/` folder per repository: a generated `README.md` index of active roles at the root of `clossys/`, and `clossys/.state/` for machine files (hub marker, inventory, and the new skills manifest).
-- Writes `clossys/.state/skills.json` on every apply: each composed skill's source (`installed` or `catalogue`), version, and a content digest. The health report states how many composed skills are out of date against the live `@clossys/launcher` version and how many were retired this run; retirement removes only a skill this directory's own previous manifest listed, never one launcher did not write.
-- Packs the shared conversation contract at build time and injects it into every composed skill in place of that skill's own "how we work together" and "one question at a time" sections, at the same position. No package edit is needed for this to take effect.
-
-### Changed
-
-- Moves its own hub marker and inventory from the hidden `.clossys/` to the visible `clossys/.state/`; the packed skeleton template moves with it. Resume detects a hub still on the legacy path and migrates it automatically, reporting the move in the health report. A hub with a marker at both paths is graded `indeterminate` and launcher refuses rather than merging them silently.
-- Adds a `.gitignore` entry for generated run output under `clossys/**/.generated/`. Approved records, proof, and machine state are still committed, never ignored.
-
-## [0.1.8] - 2026-09-21
-
-### Added
-
-- Ships the packed Agent Skill in the tarball (`files` includes `skill`).
-
-### Fixed
-
-- Skill compose prefers each checkout's installed `@clossys/<package>/skill/SKILL.md` when present, then falls back to the packed catalogue or sibling source.
-- Apply marks hub health degraded when the skill roster skips inventoried targets; missing per-package catalogue sources remain notes only.
-
-## [0.1.7] - 2026-09-20
-
-### Changed
-
-- Packed skill catalogue now carries pre-auth fold gates, the exceptional-keep brief that ships with `@clossys/designer` (synthetic user, not the doer, the sealer, or a QA contractor), and expression-wave skills that treat 3 as the floor not done.
-
-## [0.1.6] - 2026-09-20
-
-### Changed
-
-- Appoint now pins live `@clossys/advisor` in `devDependencies` only. It relocates a pin left in another bucket and overwrites a frozen version. A dedicated `{owner}/workspace` hub is named `@owner/workspace`; an appointed product keeps its package name.
-- Observe always reads the public Advisor version, including when a pin already exists, so appoint can write the live pin and resume can grade it.
-- Health is degraded when Advisor is missing, dual-pinned, or present outside `devDependencies`, not only when the pin is older than live.
-- New-hub skeleton package name is `@owner/workspace`.
-- New-hub `AGENTS.md` tells the coding agent to speak to a founder in
-  ordinary sentences: where we are, what to do next, what we will not
-  do, and whether anything is saved to git. Machine identifiers stay
-  out of the default voice.
-
-### Fixed
-
-- Resume health now receives the live Advisor version from observe, so a stale or misplaced pin is visible on every resume instead of only after a fresh appoint.
-
-## [0.1.5] - 2026-09-20
-
-### Fixed
-
-- Discovery compose skips a `.claude/skills` or `.cursor/skills` path that is already a symlink so it cannot replace composed `SKILL.md`.
-
-## [0.1.4] - 2026-09-19
-
-### Added
-
-- Hub apply composes the full `clossys-*` skill tree under `.agents/skills/` on create, appoint, and resume, reading bodies from the packed skill catalogue (built at `npm run build` from each package's skill source) or from a sibling checkout. Missing sources are skipped with a health note; apply continues.
-- The same roster is composed into every inventoried repository clone beside the hub (resolved from the generated hub inventory and confirmed with `git remote get-url origin`); the health report lists targets and skipped ids. Sister checkouts get optional canned `AGENTS.md` only when missing or still the generated sister text.
-- Apply writes host discovery links under `.cursor/skills/` and `.claude/skills/` pointing at the composed skills in the hub and each resolved clone.
-- Packed Agent Skill `clossys-launcher` so a coding agent can be invoked as `@clossys-launcher`.
-
-### Changed
-
-- Resume refreshes composed skills and replaces stale generated `AGENTS.md` when it still tells founders to use `npx` to continue the conversation; customized `AGENTS.md` files are left alone.
-- Founder-facing hub guidance (`CONSUMER_AGENTS_MD`, skeleton README) now states the same `@clossys-*` team is available in every inventoried checkout; `@clossys-advisor` is the hiring check; `npx @clossys/launcher` refreshes voices on clones beside the hub.
-
-## [0.1.2] - 2026-09-19
-
-### Fixed
-
-- Appoint merges `--inventory` into a populated on-disk hub inventory instead of silently discarding the supplied document: repositories are unioned by id (on-disk order first, new ids appended in supplied order, first occurrence of an id wins) and the merged document is written to `.clossys/inventory.json` (that on-disk path does not ship with this package). The both-empty refusal is unchanged.
-- Resume with `--inventory` now refuses with "hub already appointed; edit `.clossys/inventory.json`" (that on-disk path does not ship with this package) instead of the misleading "--inventory is only valid when appointing".
-- Appoint refuses as `violated` when `CLOSSYS_OWNER` names a different account than the repository's github.com origin owner, naming both values; the marker is no longer written for the wrong account.
-- Appoint no longer requires a readable npm registry when the tree already pins `@clossys/advisor` in any dependency bucket (no new version would be needed); the indeterminate refusal only fires when a version is actually required.
-- Appoint refuses as `violated` when `git status --porcelain` is non-empty, before writing anything; when the origin is not on github.com the refusal names the remote host.
-
-### Added
-
-- Health report grades each Advisor pin against the live registry version (internal semver compare, no new dependencies): pin older than live is a `stale pin` finding and marks the report degraded; equal pins pass; unparseable comparisons are noted as indeterminate. Exit stays 0 on resume; adopt prints the same report.
-- Pin and extra-`@clossys/*` scans now cover `optionalDependencies` and `peerDependencies` in addition to `dependencies` and `devDependencies`.
-- `checkInventoryEntries()`: read-only validation of hub inventory repository ids through batched `gh repo view --json name`, marking unknown ids in the report and skipping with a note when `gh` is absent. Never mutates the inventory.
-- `launcher-check` forwards `cwd.hub`, so a captured existing-hub observation grades as resume instead of mis-grading as adopt.
-
-## [0.1.1] - 2026-09-19
-
-### Fixed
-
-- Appoint leaves an existing `@clossys/advisor` pin in whichever bucket it already occupies. It no longer dual-pins or overwrites a frozen version with the live registry version.
-- Appoint refuses when the generated hub inventory is missing or empty (packed template `skeleton/.clossys/inventory.json`; that generated path does not ship), unless `--inventory <path>` supplies a populated document. Create may still write an empty inventory. Resume does not invent one.
-
-### Added
-
-- Read-only health report after create, resume, and appoint: hub marker, inventory classification, Advisor pin location and version versus live, dual pin, extra `@clossys/*` names. Does not uninstall.
-
-## [0.1.0] - 2026-09-18
-
-### Added
-
-- `npx @clossys/launcher` as the single get-started command.
-- In-package hub skeleton (not a Foundry fork) copied into a GitHub repository.
-- Create a new `{owner}/workspace` hub, resume an existing hub, or adopt the current GitHub repository as the account hub.
-- Owner inference from `gh` and git remotes, with an interactive picker only when more than one GitHub owner is visible.
-- `launcher-check --input` grades a captured observation without creating a hub, so qualification can prove the 0/1/2 ternary.