copilotkit 4.18.0 → 4.20.0

package/onboarding/prompts/proof/complete.md

@@ -1,217 +1,115 @@
 # Close out onboarding
 
-Report the selected framework, frontend, model, and complete round-trip evidence to the developer.
-Report the exact interaction, visible result, validation commands, and evidence locations.
+Send the friction reports first, then run `onboard complete`, then write the handoff. The
+command decides whether this run is complete, and it prints the facts the handoff carries.
 
-Do not complete onboarding without this evidence.
+## What the reports can say
 
-Report whether the real UI was driven on this journey's own surface -- in a browser for a
-web frontend, on a device for React Native -- using the outcome the proof subagent
-returned. Do not soften it and do not leave it out: a run that never drove the UI proved
-the agent, not the surface, and the developer needs to know which they have.
+Report only the project this run worked in. The handoff and every friction report describe
+that project and nothing else. Do not name or describe a file outside the project
+directory. That covers its path, its contents, how long a value in it is, and whether it
+holds a value at all. A path the developer named is theirs to give this run, not this run's to
+record. A friction report leaves the machine, and the handoff is durable.
 
-**Whether this run is complete is decided by the command at the end of this prompt, not
-here.** Run it before you write the summary, and let its output decide which summary you
-write. A run whose surface was never driven is blocked, not complete, unless it cloned a
-starter and records `skipped-cloned-starter`. For a blocked run, say so in the first
-line, name the evidence the command lists as missing, and do not describe the run as
-finished, working, or ready. Everything else below applies to either outcome.
+The handoff is the developer's handoff, not this run's retrospective. Say nothing about
+what this run found awkward in the onboarding graph, the documentation, or the CLI. A
+defect this run hit appears only as the thing the developer must do about it. Give no
+count of what went wrong, no list of fixes made along the way, and no argument that a
+failed check is not a defect. Report a failed check as part of the evidence.
 
-Report only the project this run worked in. The summary and every report below describe
-that project and nothing else. Do not name or describe a file outside the project
-directory: not its path, not its contents, not how long a value in it is, not whether it
-holds one. A path the developer named is theirs to give this run, not this run's to
-record. A friction report leaves the machine, and a completion summary is durable. The
-file's owner chose neither.
-
-The summary is the developer's handoff, not this run's retrospective. Report what was
-built, the round trip and its evidence, the validation commands, where the evidence is,
-and how to start it again. Report a failed check as part of that evidence. Say nothing
-about what this run found awkward in the onboarding graph, the documentation, or the CLI.
-A defect this run hit belongs in the summary only as the thing the developer must do
-about it: keep this version pin, wire this constructor, run this build. It never appears
-as an account of hitting it. No count of what went wrong, no list of fixes made along the
-way, no argument that a failed check is not a defect.
-
-Where this run connected and proved Rich Threads, explain in your own words that they
-preserve messages, generative UI, tool interactions, and application state so users
-can return to earlier work. Explain that these interactions can also provide evidence
-for Automatic Learning. Describe only the capabilities supported by this run's result.
-This is required content within the handoff, not a separate message or a new check.
-
-The handoff must include:
-
-- The URL of the running app.
-- The command that starts the app in the future.
-- The debugging surface this journey's frontend can reach, named below.
-- How to use https://intelligence.copilotkit.ai to manage the CopilotKit Intelligence
-  installation. Do not open it. This item makes the developer aware the dashboard is
-  there, and it is a sentence to write rather than a check to perform. Nothing in this
-  graph gates on the dashboard, so a dashboard this run never opened is not a finding,
-  not a caveat, and not a blocker. Where this run settled a Learning Container, tell the
-  developer they can use their project there to follow progress, review Insights and
-  supporting conversations, and approve proposed Skills as they become available.
-- What changed since the developer approved the plan. Name every path, version, or step
-  the run departed from, and why. A deviation can be the right call and still has to
-  reach the person who approved the thing it departed from. Where nothing departed, say
-  that in one line rather than leaving the item out.
-- Every change to the existing agent's behavior: its system prompt or instructions, its
-  tools or what those tools do, its model or provider configuration, or its memory or
-  state handling. Name each one, name the file, and say the developer approved it. Where
-  none of them changed, write `none` on that line rather than leaving the item out. Such a
-  change answers differently everywhere the agent is used, not only in CopilotKit, and the
-  developer is the only one who can tell whether that is wanted.
-- The process IDs and stop commands for the running agent and frontend.
-- The `@copilotkit/*` versions this conversion changed, and what they were before.
-- On a conversion, the criterion this run was judged against, in the words the run was
-  given.
-- The Learning Container this run settled, or the fact that Learning is not available to
-  this organization and no container was created.
-- Any path this run created that git does not ignore, and what to do with each.
-- Any protected path this run accepted as changed from outside it, and what changed. The
-  command at the end of this prompt prints each one, so copy them from its output rather
-  than from memory. This run did not change them and cannot say what did, so the developer
-  decides what to do about each.
-
-A run leaves paths beside the credential file. Tool output such as `.playwright-mcp/` and
-`.langgraph_api/` is regenerable, so the developer normally wants it ignored.
-`.copilotkit/` is not debris: `project.json` and `channels.json` are the project's own
-declared state and belong in the commit, while `.copilotkit/artifacts/` holds regenerable
-manifests and does not. Name each path and say which of the two it is. Do not add an
-ignore line for any of them yourself. None of these carries a secret, unlike the
-credential path, so the developer decides.
-
-Name the debugging surface this journey's frontend can reach, rather than the one the rest
-of the documentation leads with. For a web frontend it is the CopilotKit Inspector. For
-React Native there is no Inspector: it is a browser overlay built on a DOM custom element,
-and `@copilotkit/react-native` does not ship it. Give a mobile developer
-`npx --prefer-offline --yes copilotkit@4.18.0 verify --round-trip`, the runtime's own log, the AG-UI
-Event Inspector in the CopilotKit VS Code extension, and the CopilotKit Intelligence
-thread view
-instead. Naming the Inspector to a developer who cannot open it costs them the time it
-takes to conclude their own wiring is broken.
-
-Where this run settled a Learning Container, explain in your own words that new
-conversations are routed to it, where Automatic Learning can find recurring patterns
-and propose reusable Skills to help the agent improve. The developer controls which
-Skills get approved. Explain that Learning results appear as new conversations provide
-enough evidence.
-
-These are required explanations within the existing handoff, not fixed wording or
-additional actions. Describe only the setup supported by this run's evidence. Do not
-claim this run produced Insights or Skills or configured automatic Skill delivery
-unless its evidence proves that. If no container was configured, report the actual
-reason and retain the project-management destination without implying that Learning
-is collecting evidence.
-
-Where the Learning step was skipped because this organization cannot use Learning, say so in
-one line and name what was not created. A skip nobody names reads as a container that
-exists, and the developer then waits for insights from a container this run never made.
-
-State that the servers remain running after proof, unless the command at the end of this
-prompt prints `environment_kind: container` or `environment_kind: sandbox`. That run is on
-an isolated host, such as a Docker container or a cloud agent session, which is not a
-Learning Container. The servers end when that host ends, and a `localhost` URL opens on the
-developer's machine only through a forwarded port. Write the items that output lists in
-place of this sentence.
+## Report friction
 
 Report each thing that slowed this run down. Send at most four reports, worst first. The
 friction commands follow the telemetry setting the developer already chose, so they need no
 separate question.
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard friction --category <slug> --cost-seconds <seconds> --message "<sentences>"
-```
-
-Put one or two sentences in `--message`. Pick one category from
-docs-missing, docs-wrong, docs-sequential, cli-gap, sdk-gap, environment,
-port-collision, credential, validation-loop, and other. For the seconds, give your
-own estimate of what that one papercut cost this run.
-
-Pass --docs-path only for a docs-missing or docs-wrong report, naming the page the report
-is about:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard friction --category docs-wrong --cost-seconds 300 --docs-path /docs/threads/drawer --message "<sentences>"
+npx --prefer-offline --yes copilotkit@4.20.0 onboard friction --category <slug> --cost-seconds <seconds> --message "<sentences>"
 ```
 
-Give the page's site-relative path or its full URL, with no spaces, query string, or
-fragment. A report that names its page can be counted against that page. A report that
-describes the page in prose cannot. Leave the flag off for every other category, and leave
-it off when no single page is at fault.
+Put one or two sentences in `--message`. Pick one category from docs-missing, docs-wrong,
+docs-sequential, cli-gap, sdk-gap, environment, port-collision, credential,
+validation-loop, and other. For the seconds, give your own estimate of what that one
+papercut cost this run. For a docs-missing or docs-wrong report about one page, add
+`--docs-path <site-relative path or full URL>`, with no spaces, query string, or fragment.
 
-Send no secrets, source code, logs, or command output. The command refuses a report
-that carries any of those, prints the reason, and exits zero. A refused report is not
-a failed step and not a failed onboarding run. Reword it and send it again, or move
-on. A run that proves a round trip is complete whether or not it reported friction.
+Send no secrets, source code, logs, or command output. The command refuses a report that
+carries any of those, prints the reason, and exits zero. A refused report is not a failed
+step. Reword it and send it again, or move on. A run that proves a round trip is complete
+whether or not it reported friction. When you send a report, tell the developer. Do not
+quote it unless they ask. If the CLI says the report was not sent, say what it said and go
+on without another question.
 
-Tell the developer when you send a friction report. Do not quote or summarize the report
-unless the developer asks. If the CLI says the report was not sent,
-state what it said and continue without another question.
+## Run the completion
 
-When the evidence is gathered, run `npx --prefer-offline --yes copilotkit@4.18.0 onboard complete`, carrying
-the surface-check outcome the proof subagent returned. Pass exactly one of `--visual-check`
-or `--device-check`, and pass the one that matches this journey's surface.
-
-For a web frontend -- React SPA, Next.js, Angular, Vue:
+Pass the surface-check outcome the proof subagent returned, not the one you wanted. For a
+web frontend, pass `--visual-check` and the URL the browser opened, unchanged:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard complete --visual-check <outcome>
-```
-
-The outcome is one of `performed`, `skipped-no-browser-tool`, `skipped-cloned-starter`, or
-`failed`. Use `skipped-cloned-starter` only for a run that cloned a starter and was told to
-open no browser. It is the one skip that does not block.
-
-For React Native:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard complete --device-check <outcome>
-```
-
-The outcome is one of `performed`, `skipped-no-device`, `skipped-cloned-starter`, or
-`failed`.
-
-The two flags are not interchangeable and neither takes the other's outcomes. A browser
-proves nothing about a React Native view tree, and a device capture proves nothing about
-browser-origin CORS, so the flag you pass is how this run states which surface it proved.
-
-For a web frontend, also pass the URL the browser opened:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard complete --visual-check <outcome> \
+npx --prefer-offline --yes copilotkit@4.20.0 onboard complete --visual-check <outcome> \
   --frontend-url <the url you opened>
 ```
 
-Pass the URL from the proof subagent's record, unchanged. Only the kind of host is
-recorded -- a loopback hostname, a loopback IP literal, another IP literal, or another
-name -- and never the host itself. A run that opened the loopback IP literal loses every
-static chunk to a refusal, and the field is how that stops being invisible. Leave the flag
-off for React Native, which opens no URL.
-
-Pass the outcome you were given rather than the one you wanted. Anything but `performed` or
-`skipped-cloned-starter` prints what the missing check leaves unverified and ends this run
-as blocked. That output
-is the developer's finding, so carry it into the summary rather than restating it as a
-smaller caveat.
-
-If the round trip proved and something after it still blocked this run, add `--blocked-by`
-to the same command:
+The outcome is one of `performed`, `skipped-no-browser-tool`, `skipped-cloned-starter`,
+or `failed`. For React Native, pass `--device-check` with one of `performed`,
+`skipped-no-device`, `skipped-cloned-starter`, or `failed`, and no URL. Use
+`skipped-cloned-starter` only for a run that cloned a starter and was told to open no
+browser. Use `skipped-no-browser-tool` for a run with no browser tool loaded in this
+session.
+
+`skipped-cloned-starter` does not block. `skipped-no-browser-tool` does not block when this
+run recorded a passed round trip with `onboard proof`. The command then prints what no
+browser checked and how to add a browser for the next session. `failed` and
+`skipped-no-device` print what the missing check leaves unverified and end this run as
+blocked.
+
+If the round trip proved and something after it still blocked this run, add
+`--blocked-by <cause>`: `inspector` for a debugging surface that did not open,
+`plan-excluded-capability` for a capability the approved plan excluded,
+`browser-control-unresponsive` for a browser tool that stopped answering during the surface
+check, or `other`. Do not use it for the CopilotKit Intelligence dashboard, which nothing in
+this graph opens.
+
+`browser-control-unresponsive` is the one cause that goes with the failed check, because
+nothing drove the frontend:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard complete --visual-check performed --blocked-by <cause>
+npx --prefer-offline --yes copilotkit@4.20.0 onboard complete --visual-check failed --blocked-by browser-control-unresponsive
 ```
 
-The cause is one of `inspector` for a debugging surface that did not open,
-`plan-excluded-capability` for a capability the approved plan had already excluded for this
-framework, and `other` for anything else. Pick `other` rather than leaving the flag off.
-The command ends the run as blocked and prints what the blocker leaves unverified.
-
-The managed Intelligence dashboard is not among these, because nothing in this graph asks a
-run to open it. A run that records a blocker for a surface it was never sent to looks worse
-than an identical run that wrote the same handoff sentence, and the two are then not
-comparable.
-
-This is the flag for a journey that worked. It is not a way to report an unsupported path,
-and a blocked run here is still owed the full handoff above: the URL, the start command,
-the process IDs, and the versions this conversion changed.
+## Write the handoff
+
+If the command reports the run as blocked, say so in the first line and name the evidence
+it lists as missing. Do not call the run finished, working, or ready. A blocked run is
+still owed the full handoff.
+
+For `skipped-no-browser-tool`, the command's list of what no browser checked is that report.
+Give it once, in the section about what was proved, with how to add a browser, as the
+command prints them. Do not add a separate line, heading, or opening sentence about the
+missing browser.
+
+Paste the **Handoff facts** block the command printed, as it is. Copy any accepted or
+authorized path it lists too. The dashboard line there only tells the developer where the
+dashboard is. Do not open it. Nothing in this graph gates on the dashboard. Then add what
+only you can say:
+
+- What the project now does, and the first thing to try in it.
+- What changed since the developer approved the plan: every path, version, or step the run
+  departed from, and why. Where nothing departed, say so in one line. A file the approved
+  plan named under `Files the run writes itself` is not a departure, and neither is a
+  Skills install the developer declined.
+- Every change to the existing agent's behavior: its instructions, its tools, its model or
+  provider, or its memory. Name each one and its file, and say the developer approved it.
+  Where none changed, write `none`.
+- On a conversion, the criterion this run was judged against, in the words the run was
+  given.
+- Where this run proved Rich Threads: they keep messages, generative UI, tool interactions,
+  and application state, so users can return to earlier work. The same conversations give
+  Automatic Learning its evidence.
+- Where the facts name a Learning Container: new conversations go to it. Automatic
+  Learning can find recurring patterns and propose Skills, and the developer approves which
+  ones apply. Results appear as conversations give enough evidence. Claim no Insight or
+  Skill this run did not produce.
+
+Say that the servers keep running, unless the command printed `environment_kind`. Then
+write the items it lists instead.

package/onboarding/prompts/proof/oss-baseline.md

@@ -1,7 +1,7 @@
 # Prove the existing OSS baseline
 
 Do not prove the baseline yourself. Spawn one proof subagent. Tell it to run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read subagent/prove-oss-baseline` first and follow the
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read subagent/prove-oss-baseline` first and follow the
 prompt it returns. If that read fails because the subagent cannot use the shell, stop that
 subagent. Run the same command yourself, then spawn a fresh subagent with the returned prompt
 and the same handoff. Give it the repository findings and exact CLI package spec.
@@ -17,7 +17,7 @@ Wait for the subagent to finish.
 Record what that proof returned before you route on it:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard proof --step oss-baseline --outcome <passed|failed|skipped> [--predicate <1-6>]
+npx --prefer-offline --yes copilotkit@4.20.0 onboard proof --step oss-baseline --outcome <passed|failed|skipped> [--predicate <1-6>]
 ```
 
 Report the gate whatever it returned. Use the list below. A proof that never ran is `skipped`, not failed. The command prints one line and sends
@@ -39,12 +39,12 @@ Do not change project files before this proof ends. Starting existing developmen
 processes and their ignored runtime files is allowed.
 
 If the subagent proves the `both-oss` predicate, keep its evidence with the plan and run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read conversion/plan`. That project already works.
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read conversion/plan`. That project already works.
 What it needs is the conversion, not a build.
 
 If the proof does not establish the baseline, record the starting state
 `both-copilotkit-unproved` and run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read credentials/plan`. This prompt is served
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read credentials/plan`. This prompt is served
 whenever a project looks like an OSS integration, so a baseline that did not prove is an
 ordinary starting state rather than a failure. Keep the failing predicate with the plan.
 
@@ -58,5 +58,5 @@ and the plan preserves it rather than repeating it.
 
 If it cannot identify the running process safely, exposes a secret, or finds a baseline
 failure that cannot be classified, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read stopped/run-failed`. None of those mean the
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read stopped/run-failed`. None of those mean the
 project is unsupported: they mean this run did not establish what it needed to.

package/onboarding/prompts/proof/round-trip.md

@@ -1,7 +1,7 @@
 # Prove the user journey
 
 Do not do the proof work yourself. Spawn one proof subagent. Tell it to run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read subagent/prove-round-trip` first and follow the
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read subagent/prove-round-trip` first and follow the
 prompt it returns. If that read fails because the subagent cannot use the shell, stop that
 subagent. Run the same command yourself, then spawn a fresh subagent with the returned prompt
 and the same handoff. Give it the selected framework, frontend, model, approved plan, selected
@@ -13,6 +13,8 @@ Give it the browser or device control you recorded in the preflight as well. The
 drives the surface and cannot see your environment, so without that finding it spends the
 step discovering what you already know. A subagent told it has no control for the surface
 this frontend needs reports the skip outcome rather than looking for a way around it.
+A web run with no browser tool reports `skipped-no-browser-tool`. That outcome completes the
+run once the round trip is recorded as `passed`.
 
 A run that cloned a starter is the exception. Tell that subagent that this run cloned a
 starter, to finish after `verify --round-trip`, and to open no browser. The cloned code is what this repository's
@@ -22,7 +24,7 @@ expensive step in this setup. `verify --round-trip` reads the answer back off th
 so it holds for every runtime mount and needs no browser. Give that subagent no browser or
 device control, and have it report `skipped-cloned-starter` rather than a missing
 capability: nothing was unavailable, the run declined to spend it. That outcome completes
-the run. Every other skip says nothing drove the surface, and blocks it.
+the run.
 
 Give the subagent this guide for continued-development tools:
 https://docs.copilotkit.ai/build-with-agents.md
@@ -30,16 +32,10 @@ https://docs.copilotkit.ai/build-with-agents.md
 Wait for the subagent to finish. Its result arrives as a notification. Do not sleep to
 pass the time.
 
-Report each attempt at the journey as it ends, counting from one:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase journey-attempted --attempt 1
-```
-
 Record what that proof returned before you route on it:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard proof --step round-trip --outcome <passed|failed|skipped>
+npx --prefer-offline --yes copilotkit@4.20.0 onboard proof --step round-trip --outcome <passed|failed|skipped>
 ```
 
 Report the gate whatever it returned. A proof that never ran is `skipped`, not failed. A
@@ -48,7 +44,7 @@ to record it as `skipped`. The command prints one line and sends nothing else. W
 record each attempt as it ends.
 
 For every protected-path audit in this prompt, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard audit` from the target app directory. If its result
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard audit` from the target app directory. If its result
 starts with `Status: blocked`, report the printed reason and use the route-out rules below.
 A blocked audit proved nothing changed and is not a preservation failure. If a
 protected-path audit reports a changed path, decide it the way the implementation prompt
@@ -57,7 +53,7 @@ returns none, so a finding with no Files changed section to test against routes
 path one of those sections names is this run's own change and routes out too. Accept a
 path only when a section this run collected covers the step that wrote it and does not
 name it:
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard protect --accept-external --path <path>`. Then run
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard protect --accept-external --path <path>`. Then run
 the audit again and name the path in the closing summary. Never repair, reset, or revert a
 protected path.
 
@@ -65,17 +61,23 @@ That holds for a repair cycle too. When the fix for a failing check lands on a p
 path, the path is still the developer's, however right the diagnosis is and however small
 the fix. Reading the file never settles who wrote it. Ask the developer to allow the
 change, and record their answer with
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard protect --authorize --unplanned --path <path> --reason "<what the developer said>"`,
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard protect --authorize --unplanned --path <path> --reason "<what the developer said>"`,
 or route out. Never repair it, and never send it to a repair worker.
 
 If the proof result starts with `Status: passed`, run the protected-path audit. Continue to
 `proof/complete` only if that audit passes. After the audit passes, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read proof/complete`. A performed surface outcome with
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read proof/complete`. A performed surface outcome with
 the full round trip is core success even if a continued-development tool fails. A skipped
-surface outcome still enters `proof/complete` so the CLI records the blocked result.
-`skipped-cloned-starter` is the exception: the CLI records that run as complete. Do not
-describe a skipped surface as proved. Keep the Skills and MCP results separate from the proof
-result.
+surface outcome still enters `proof/complete`. The CLI records `skipped-cloned-starter` and
+`skipped-no-browser-tool` after a passed round trip as complete, and `skipped-no-device` as
+blocked. Do not describe a
+skipped surface as proved. Keep the Skills result separate from the proof result.
+
+A passed result can carry the surface-check outcome `failed` with the cause
+`browser-control-unresponsive`. It means that the browser tool stopped answering, not that
+the journey failed. Record the round trip as `passed`, and carry the cause to
+`proof/complete`. Do not send it to a repair worker, and do not run the proof again. No
+repair to the project reaches the tool, and a second attempt waits on the same hang.
 
 If the round trip proves and something after it blocks this run anyway -- the debugging
 surface, or a capability the approved plan already excluded for this framework -- read
@@ -144,14 +146,7 @@ authorized list from that audit.
 Run the protected-path audit again after validation passes. Continue only if its result
 starts with `Status: passed`.
 
-Restart each project-owned process changed by the repair. Report the cycle, counting from
-one:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase repair-attempted --attempt 1
-```
-
-Then spawn a fresh proof subagent
+Restart each project-owned process changed by the repair. Then spawn a fresh proof subagent
 with the full original proof handoff, failed proof evidence, and new validation evidence.
 This handoff includes the plan, documentation, policy, current process IDs, ports,
 surface-control state, the continued-tools guide, and the failed attempt's pinned Step 1
@@ -166,7 +161,7 @@ proof cycles, use the route-out rules below.
 
 Route out only when the failure is not yours to fix, when the same proof still fails after
 three attempts, or when no evidence of the round trip can be produced. In those cases run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read stopped/run-failed`. The stack is supported:
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read stopped/run-failed`. The stack is supported:
 this run did not finish, which is a different ending and a different report. All three are
 about the round trip itself. A round trip that proved is not one of them, whatever failed
 after it.
@@ -177,7 +172,7 @@ friction command follows the telemetry setting the developer already chose, so i
 separate question.
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard friction --phase stop --category <slug> --message "<sentences>"
+npx --prefer-offline --yes copilotkit@4.20.0 onboard friction --phase stop --category <slug> --message "<sentences>"
 ```
 
 `--message` takes one or two sentences: the step you stopped at and what stopped it.

package/onboarding/prompts/research/gather.md

@@ -44,7 +44,7 @@ Before you end your turn to wait for the developer where the prompt states no de
 report the pause:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase awaiting-developer
+npx --prefer-offline --yes copilotkit@4.20.0 onboard checkpoint --phase awaiting-developer
 ```
 
 When the answer or the result arrives, continue from the step that paused. Do not read an
@@ -64,7 +64,7 @@ Both research subagents failing means this harness has no working subagent at al
 is worth recording once:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase delegation-unavailable
+npx --prefer-offline --yes copilotkit@4.20.0 onboard checkpoint --phase delegation-unavailable
 ```
 
 Then say once, in your own words, that this environment has no working subagents, so you
@@ -77,7 +77,7 @@ Before you ask the developer any setup question, finish every read-only investig
 preflight check in this section.
 
 Prepare two research assignments. Give each research subagent one assignment. Tell it to run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read subagent/inspect-repository` first and follow the
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read subagent/inspect-repository` first and follow the
 prompt it returns. If that read fails because the subagent cannot use the shell, stop that
 subagent. Run the same command yourself, then spawn a fresh subagent with the returned prompt
 and the same handoff. Require only its assigned packet.
@@ -88,20 +88,11 @@ Start both research subagents in parallel:
 2. Spawn a second research subagent. Assign it the environment evidence packet.
 3. Start both research subagents. Then continue.
 
-Report that both subagents started:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase research-dispatched
-```
-
-A refused checkpoint prints its reason and leaves onboarding unaffected. It is not a
-failed step.
-
 The research is under way. Continue to the surface-control preflight while it runs:
 
 ```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard read research/preflight
+npx --prefer-offline --yes copilotkit@4.20.0 onboard read research/preflight
 ```
 
 If inspection stops onboarding, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read stopped/run-failed`.
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read stopped/run-failed`.

package/onboarding/prompts/research/merge.md

@@ -18,15 +18,6 @@ whatever holds that port answers.
 
 Project selection is where the settled port is written down, through `--runtime-url`.
 
-Then report that the research came back:
-
-```text
-npx --prefer-offline --yes copilotkit@4.18.0 onboard checkpoint --phase research-returned
-```
-
-A refused checkpoint prints its reason and leaves onboarding unaffected. It is not a
-failed step.
-
 Continue only if both research results start with `Status: passed`. For `Status: failed`,
 retry only that packet with its failed items, up to three attempts. For `Status: blocked`,
 or a third failed result, use the stop route at the end of this prompt.
@@ -39,7 +30,8 @@ verifier result.
 
 A target project directory that holds no project has no app directory to key on, and
 that is a merged result rather than a failed merge. Record that this run has no target app
-directory and no environment row, then take the read below. Do not send a focused directory
+directory and no environment row. Keep the toolchain reading: it is about this machine, not
+a directory, and the framework question reads it. Then take the read below. Do not send a focused directory
 check, and do not use the stop route: neither worker can find a directory a developer has
 not created yet, and the route below is where such a project picks its starter.
 
@@ -54,7 +46,7 @@ worker a focused directory check. Continue only when both workers return the sam
 app directory. Both results must start with `Status: passed`. Otherwise, use the stop route.
 
 When both research results are merged, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read research/route`.
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read research/route`.
 
 If inspection stops onboarding, run
-`npx --prefer-offline --yes copilotkit@4.18.0 onboard read stopped/run-failed`.
+`npx --prefer-offline --yes copilotkit@4.20.0 onboard read stopped/run-failed`.