Changelog

Bug fixes and improvements

Smaller fixes and reliability improvements across the API and MCP.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, and MCP.

Automatic sync turns on once you set a sync target

Configuring a sync target now turns automatic sync on by itself, with no separate command. If you have turned it off, it stays off.

Sync your skills to Codex

A new codex preset points a sync target at the directory Codex reads personal skills from.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, and MCP.

Workflows are now skills

The stored object is now a skill: the CLI, MCP tools, and API all use the skill naming, and the older goodeye workflows commands and /v1/workflows REST paths keep working until 2026-10-01 before they are removed.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, and MCP.

Bug fixes and improvements

Smaller fixes and reliability improvements across the API and MCP.

Bug fixes and improvements

Smaller fixes and reliability improvements across the API, MCP, and web surfaces.

Buy credits on demand

Top up your balance any time, charged to your card on file or through a one-time checkout link.

Keep your balance from hitting zero

Turn on automatic top-ups to refill your credits below a threshold you set, with a monthly cap.

Upgrade to Pro and manage your subscription

Upgrade to Pro for a bigger monthly credit grant, and cancel or manage billing anytime by asking your AI agent.

Bug fixes and improvements

Smaller fixes and reliability improvements across the API, MCP, and web.

Newer models for verifier judges and search

Verifiers can now judge with the latest Claude and OpenAI models, including Claude Sonnet 5, Claude Opus 4.8, Claude Fable 5, GPT-5.5, and GPT-5.5 Pro.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI.

Authoring checks on every public template

Every public template now shows four authoring checks at a glance: Safety, Outcome, Runnable, and Well-formed, so you can judge a skill before you fork it. See how we check.

• Template reads include the check states, so your own tools can surface them too.
• Audit a skill against the same checks to improve it before or after publishing.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, and MCP.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI and API.

Legacy goodeyelabs.com API and MCP hosts retired

The old api.goodeyelabs.com and mcp.goodeyelabs.com hosts are retired; use api.goodeye.dev and mcp.goodeye.dev (recent CLI versions switch over automatically).

Saving a skill by name always saves your own

Saving a skill by name creates or updates your own and never edits one shared with you. To update a shared skill, reference it by its id instead.

Public images are screened for disallowed content

Uploading a public image or making one public now screens it first, so disallowed content cannot be served from a public link.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, MCP, and web.

Host images and get durable URLs for generated images

Upload, host, and manage images from the CLI, API, and MCP, and get a stable URL that never changes.

• Upload and host an image, then get a stable URL you can share or embed; set visibility to public or private and flip it anytime.
• Set a TTL so an image expires automatically, or mark it permanent; change the TTL or remove the expiry anytime.
• Manage hosted images from the CLI (`goodeye images ...`), REST, or MCP with a consistent five-command suite: upload, list, get, update, delete.
• Generated images now return durable hosted URLs automatically, so the link you get stays valid long after the provider session ends.

Clearer, more consistent guided skill sessions

The design, teaching, optimize, and audit sessions now follow one standard, so each is easier to follow and gives stronger guidance.

A faster, more focused skill design session

The guided design session asks fewer upfront questions and adds more built-in guidance for writing effective skills.

Sign in with Google

Use your Google account to sign in or create an account, alongside the existing email option.

Earn bonus credits by referring new users

Share your referral code so new users get bonus credits when they redeem it, and you earn bonus credits once they activate.

Auto-pull keeps local sync targets fresh without manual runs

Enable auto-pull once and the CLI pulls new and updated skills in the background. It never overwrites local edits and never deletes.

Bind an image generator to a skill

Attach an image generator to a skill so it produces images on the same terms everywhere, across the CLI, API, and MCP.

• Bind a generator with the repeatable --image-generator name=ref flag, or clear all bindings with --clear-image-generators.

Bug fixes and improvements

Smaller fixes and reliability improvements across the CLI, API, and MCP.

Visual demos and social cards for template pages

A template's public page can now show a visual demo, so visitors see what a skill produces before they try it.

• Add a demo/README.md to a skill bundle to show a visual demo on its public template page: a writeup with images, short animated clips, and an embedded video from YouTube, Loom, or Vimeo.
• Every template page gets a custom social-share card, either a demo/og image you provide or one generated from the template's details.
• Every image in a published template is screened for disallowed content, and a disallowed image blocks the publish.
• Saving or publishing flags demo issues to fix first: an external image that will be dropped, a video link that will not embed, or a demo image missing from the bundle.

Skill audits and trigger-description tuning

Two ways to sharpen a skill before you rely on it, across the CLI, API, and MCP.

• Audit a hosted skill, or an unsaved local skill file, against a clear rubric: get a priority-ranked report that names each issue with the exact text to change, then apply the fixes you pick.
• Tune the description that decides when a skill fires; an agent-driven loop measures how reliably it triggers and proposes a sharper one. Only the description changes, and nothing saves without your approval.

Archive, permanent delete, and version-history sharing

More control over the full life of a skill or template, across the CLI, API, and MCP.

• Archive and unarchive a skill or template to hide it from your lists and from public view without deleting it; your name and slug stay reserved.
• Delete is now permanent: erase an item and all its versions, or a single old version, with no recovery. Archive instead to keep a way back.
• Grant access with the full version history, not just the version current at share time.

Multi-file skill bundles publish reliably through the safety scanner

Publishing a skill with attached files no longer fails when the bundle is clean. The safety check now handles empty fields correctly and respects the judge's all-clear verdict.

Generate and host images from a skill

Deploy an image generator, produce images on demand, and snapshot image settings into a published template. Available from the CLI, API, and MCP, billed per image.

Attach a full file tree to any skill or template

Skills can now carry multiple files alongside the main body. Save a whole directory, publish it as a template, and fork it back with every file intact. The CLI can keep that file tree in sync.

Keep a local folder of skills in sync with Goodeye

The new skills sync commands mirror a local directory to your hosted skills: pull to download, push to upload edits, and status to see what has drifted.

Patched several security advisories in third-party dependencies

Routine security updates across our third-party dependencies, all within the current major versions.

Restore a deleted skill with undelete

Recover a skill you removed by mistake. Collaborators who had access before the deletion get it back automatically.

Goodeye moved to goodeye.dev

The API is now at api.goodeye.dev and the MCP server at mcp.goodeye.dev, with public templates at goodeye.dev. Older CLI installs keep working and update their saved address on first use.

Optimize a skill through agent-driven iteration

Run a structured improvement loop on any skill you own. Your agent works through scenarios and saves only the version you approve.

Check skill or template safety on demand

Run the safety checks against any saved skill or published template without going through a publish step.

See your full spendable balance, including one-off credits

Your usage summary now shows your available balance in one place, with monthly grant and any one-off credits listed separately.

LLM judge requests limited to a curated set of supported models

Verifier deploys and skill search now accept a vetted list of judge models, and reject anything outside it with a clear message.

Initial public release

The first installable release of the Goodeye CLI, API, and MCP server.

• Design private skills and share them with specific users and teams.
• Publish a skill snapshot as a public template with a stable @handle/slug link that anyone can fork.
• Deploy, run, and revoke LLM-judge verifiers, and run a public template's verifier with no account.
• Search your skills and public templates in natural language.
• Manage the full template lifecycle: delete, restore, deprecate, and transfer.
• Rename your handle, with old handles reserved so existing links keep resolving.
• Track model usage against a monthly credit grant, with an automated safety screen on every published template.