Open WorkBuddy SkillHub
Go to Experts, Skills & Connectors, choose Skills, then open SkillHub.
Collect live status across four AI delivery dependencies with Lexmount WebFetch, verify key pages in a cloud browser, and return an auditable GO, CAUTION, HOLD, or UNKNOWN decision.
Before your first run, search for and install the required Skills inside the WorkBuddy app.
Go to Experts, Skills & Connectors, choose Skills, then open SkillHub.
Enter lexmount in the search field and confirm that both Lexmount Skills appear.
Click + on both cards. After the installed confirmation appears, verify that both Skills are enabled.
Skills load immediately after installation, so a restart is usually unnecessary. On first use, each Skill may download its Rust CLI and open a browser for Lexmount OAuth. Never paste an API key into the conversation. If the prompt returns SKILL_REQUIRED, confirm that both Skills are installed and enabled.
Use the installed and enabled WorkBuddy Skills: lexmount-webfetch and lexmount-browser.
Preflight: if either Skill is unavailable, stop with SKILL_REQUIRED, list the missing item, and tell the user to open Experts, Skills & Connectors → Skills → SkillHub in the WorkBuddy app, search for lexmount, then click “+” on the matching card to install and enable it. Do not open the external SkillHub website, create a replacement Skill, or switch to another web collection or browser tool.
When both Skills are available, follow their bundled instructions to bootstrap the Rust CLIs and run doctor; do not install the legacy Python/uv packages. Only when doctor reports missing credentials, start the built-in OAuth/PKCE login and wait for browser approval. Never request or ask the user to paste an API key. Continue only after doctor is ready.
Task: produce an “Are we safe to release today?” health brief for the AI development delivery chain. At execution time, check these four official public status pages for active outages, degraded performance, maintenance, or an important incident resolved within the past 24 hours:
- Model service: https://status.openai.com/
- Code and CI: https://www.githubstatus.com/
- Package management: https://status.npmjs.org/
- Edge network: https://www.cloudflarestatus.com/
Decision rules: return HOLD when a critical service reports a Major Outage or a critical component is unavailable; CAUTION for an active incident, degraded performance, partial outage, or relevant maintenance; GO only when all four links explicitly report normal operation; and UNKNOWN whenever evidence is incomplete. Never treat unknown as operational.
Flow:
1. Create artifacts/workbuddy-release-readiness in the working directory. If it exists, append a timestamp instead of overwriting prior results.
2. Use lexmount-webfetch first for all four URLs. Record requested_url, final_url, accessed_at_utc, overall_status, critical components, active incidents, the latest important incident resolved within 24 hours, maintenance, original source time text, and warnings. Convert a timestamp to UTC only when the source states its timezone.
3. Escalate to lexmount-browser when a source omits overall status, conflicts with another field, or reports a non-operational state. Even when all four sources are healthy, open the Cloudflare Status Overview with lexmount-browser and retain at least one visual screenshot. Browser activity is read-only: do not click Subscribe to updates or enter an email address, phone number, or webhook.
4. Write brief.md, facts.json, evidence/*.png, and run-summary.json. brief.md must contain the GO / CAUTION / HOLD / UNKNOWN decision, blockers, watch items, and recommended actions. Every major conclusion must reference a source_id in facts.json. Use null, an empty string, or an empty array for missing values; never invent data.
Acceptance: every input URL has an explicit result; every JSON file is non-empty and parseable; every major conclusion has a source and access time; every declared screenshot exists and is non-empty; and all temporary Browser Sessions are closed. Return SUCCESS only when all four sources and cleanup succeed, PARTIAL_SUCCESS for any source or cleanup failure, and FAILURE when fewer than two sources succeed. With fewer than two successful sources, release_readiness must be UNKNOWN.
Cleanup: on success, failure, timeout, or cancellation, close only temporary Sessions created by this task. Do not close WorkBuddy, pre-existing Sessions, or persistent Contexts.
Return only status, release_readiness, successful and failed source counts, the four artifact paths, screenshot count, cleanup result, and a sanitized error summary. Do not expose API keys, Project IDs, Session IDs, connection URLs, or OAuth details.