Repository navigation
Conversation
…e MCP server's setup text and /start/open-source say so
init --agent adds the dev-only agent link next to the import it writes (every framework; one if block in the files
it creates for Next.js; plain HTML gets a module script after the tag that loads dist/agent.js from the script URL's
directory on localhost), and remove takes it out, wrapped by a formatter or not. The READMEs' quick start, INSTALL.md
("How do I set up the open-source parts only?"), the FAQ, AGENT.md, the MCP server's README and setup text, and the
start page's prompt and commands name the two commands (npx @genclass/runtime@beta init --no-telemetry --agent, then
npx -y @genclass/mcp@0.2.1 init) instead of a nine-step hand edit; every @genclass/mcp launcher stays pinned, with
INSTALL.md and FAQ.md added to the sync list.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
… line, the script-tag agent link in AGENT.md, what you will see on /start/open-source, the skill's listings - INSTALL.md names 0.3.0-beta.5 and 0.2.3, pins its script tags to beta.5 (SINKS.md too; beta.0 is inside the mode-raise advisory's range), gains the --sync and --no-network rows its help already prints, and says to reload a dev page once the MCP server runs. - AGENT.md no longer says the script-tag build has no agent entry: init --agent writes the module script. - README.md names @genclass/mcp 0.2.1; the worker's MCP_PKG follows. - /start/open-source says what the developer will see (the Model ready line, the overlay, genclass_status). - docs/adoption/listings/skills.md: where the genclass skill gets listed (skills.sh by installs, gh skill, cursor.directory, the curated lists) and what each needs; mcp.md points at it. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
…me; the CLI's MCP launcher is pinned; two wording fixes From the adversarial review of the diff: - agentUrl returns null when the script URL does not end in genclass.global(.min).js, so init --agent with such a --cdn writes no module script (it imported the runtime itself as a module, and connectAgent was undefined) and says so in a note; test in cli.test.ts. - The runtime CLI's help and Next line name the MCP server's launcher pinned (bin/lib/version.mjs MCP_PKG, in the manifests sync), as every other copied command does; the CHANGELOG entry too. - The READMEs say what the telemetry opt-out is for plain HTML and Astro; skills.md cites the MCP CHANGELOG's Unreleased entry for the licence field. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
…ent entry for the READMEs' bare package URL; the prompt's webhook sink by name
From the adversarial review:
- A block form in remove's recogniser could match an empty if (...) {} window, so after a developer edited the
agent line, a user's line below it could be taken out with it; a block now needs one of init's two calls.
- --cdn https://cdn.jsdelivr.net/npm/@genclass/runtime@<version> (the READMEs' script URL, with no file name)
gets its dist/agent.js next to it; other file names still get a note instead of a module script.
- /start/open-source: the logs choice says --sink webhook:<url> (which init and the runtime handle, the endpoint
kept out of what GenClass watches) instead of a hand-written sink; step 1 says what the plain-HTML and Astro
forms are.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
Contributor
Author
|
CI note on Generated by Claude Code |
…dule headers name the agent line and the fourth inline form; skills.md cross-references Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
The adversarial review's last confirmed item: every command block named --agent as the setup that works today, while npm's beta tag (0.3.0-beta.4, then beta.5) rejects the flag; the blocks now carry the version, and the merge is tied to the 0.3.0-beta.6 release. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Fqwfq3CizvxvgyXp6wY8WR
… documented flag Owner decision 2026-11-02: the two commands are `npx @genclass/runtime@beta init --agent` and `npx -y @genclass/mcp@0.2.1 init`, with the runtime's anonymous diagnostics on and the app's free dashboard link, as with every init; `init --no-telemetry --agent` is the documented way to send nothing. Every surface that showed `init --no-telemetry --agent` now shows `init --agent` and says what is sent: both READMEs, INSTALL.md, the FAQ, the CLI's --agent help, the CHANGELOG, and the worker's /start/open-source (its prompt, lead, "What leaves your network" and commands), with its tests. The MCP-side mentions (AGENT.md, the MCP README, tools.ts) already showed the bare `init --agent` and are unchanged. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…e; the CHANGELOG heading names the release The two-command setup (`init --agent`) ships in 0.3.0-beta.6, so the package version moves from 0.3.0-beta.5 to 0.3.0-beta.6 in `packages/runtime/package.json`, in `RUNTIME_VERSION` (`src/version.ts`; the release workflow checks that the tag, package.json and this string agree, and `test/telemetry.test.ts` keeps the two equal) and in the root `package-lock.json`'s `packages/runtime` workspace entry, edited by hand so the committed lockfile stays in sync (`ONNXRUNTIME_NODE_INSTALL=skip npm ci --no-audit --no-fund` succeeds against it). The CHANGELOG heading "Unreleased (0.3.0-beta.6)" becomes "0.3.0-beta.6 (2026-11-03)", the tag date assumed for the release; nothing in the entry changes. No runtime code, no model-visible text. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…ipt-tag pins move from beta.5 to beta.6 The same pass commit b740e43 made for 0.3.0-beta.5, one release on. The shipped README, TELEMETRY.md and INTERCEPTION.md describe and pin the version they ship with (a published tarball cannot be republished), so every "newest line" and "installs" mention and every jsDelivr script-tag pin names 0.3.0-beta.6, with the beta line's publish dates as 2026-10-22 to 2026-11-03 (0.3.0-beta.5 published 2026-11-03, beta.6 tagged the same day, both assumed); the repository README and docs/runtime (API, ARCHITECTURE, OPTIONS-SPEC, README, INSTALL, SINKS, ANALYTICS) follow. The package README's `init` version-check example moves one step (its own version beta.5, the newer one beta.6) and its "installs its own version" range reads `^0.3.0-beta.6`. The pins carry no integrity hash, as before. Historical mentions stay: "since 0.2.3 and 0.3.0-beta.5" in OPTIONS-SPEC and THREAT-MODEL, the measured visual checks' "from 0.3.0-beta.5" in the READMEs, TELEMETRY.md, PRIVACY.md and VISUAL.md, the CHANGELOG's beta.5 section, RESULTS.md's limit rows, SECURITY.md and docs/security, and the test comments. No runtime code, no model-visible text. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…r-send, the hand-back format and the review loop) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
…citing-bohr-yivjyo Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One conflict: telemetry-worker/src/pages.ts, the /start/open-source prompt. Kept this branch's rewritten prompt, which already names genclass-dev/GenClass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The rename on main left these out because the sections they sit in are this branch's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
A developer who wants only the open-source parts (no account, no token, nothing sent) had three stories that disagreed: the README quick start (which creates a dashboard token by default), a nine-step hand-edit prompt on genclass.dev/start/open-source, and an agent link that
initcould not write. Now it is two commands, everywhere:init --agent(bin/lib/plan.mjs,edit.mjs,init.mjs): writes the dev-only agent link next to the devtools line, the same shape for every framework (if (<dev>) import("@genclass/runtime/agent").then((a) => a.connectAgent(genclass)); oneifblock in the files created for Next.js; plain HTML gets a module script after the tag that loadsdist/agent.jsfrom the script URL's directory onlocalhostand connectsGenClass.runtime).removerecognises it, wrapped by a formatter or not, and refuses an edited one.--agent --no-devtoolsstill binds the runtime. Tests:test/install/cli.test.ts"init --agent" (Vite, Next 16, Nuxt 4, Astro, plain HTML, formatter-wrapped removal).--agent,--sync,--no-networkrows, a FAQ entry, AGENT.md, the MCP README; the MCP server'sinitandgenclass_setupnow lead with the one command; INSTALL.md is on the current line (beta.5, 0.2.3);docs/adoption/listings/skills.mdsays where thegenclassskill gets listed.telemetry-worker/src/pages.ts): the prompt is the two commands plus the dev-server check (4 steps, 591 words; was 8 steps, 853), every choice is aninitflag, and the page says what the developer will see. The site's embedded copy: genclass-dev/genclass-site PR 15.Checks
npm run typecheck -w @genclass/runtimeandnpm run build -w @genclass/runtimepasspackages/runtime(1270 passed, 15 skipped),telemetry-worker(361) andpackages/mcp(152; INSTALL.md and FAQ.md joined the pinned-launcher sync list)packages/runtime/test/review-*.test.tswas editedsituation-v2.3); nothing underpackages/runtime/srcchangedpackages/runtime/CHANGELOG.md"Unreleased (0.3.0-beta.6)" has the entrysim/anddemos/authors did not read each other's codeNotes for the reviewer
Sequencing. The published
betaCLI (0.3.0-beta.4 and beta.5) rejects--agent(unknown option), so the README and the site must not name it before 0.3.0-beta.6 is on npm. Plan: push the pendingv0.3.0-beta.5tag, run its post-publish pins, then merge this PR with the version bumped to 0.3.0-beta.6 and tag it. Draft until then.Not run: the browser model suite, the install suite's real-framework builds (
test/install/frameworks.sh), anything on Azure. A fresh-eyes review (five walkthroughs, two judges) and an adversarial review of this diff ran in the session; the latter's findings, if any, will be pushed here.Generated by Claude Code
🤖 Generated with Claude Code