Add ulysses-check (DELLIGHT AI search measurement skill)

[arthurelgindell]

What this PR does

Adds one new remote-source plugin entry.

• Plugin name: ulysses-check
• Type: remote source
• Source URL + pinned SHA: https://github.com/dellightai/grok-plugins.git @ c82df8748b7072017846d17a741d3fc018acaee8, path ulysses-check
• Homepage: https://www.dellight.ai/

Ownership

• I own this plugin or have the right to distribute it.
• The source repo is published under our official org (dellightai — DELLIGHT.AI Limited).

Checklist

• Added/updated exactly one entry in .grok-plugin/marketplace.json (valid JSON, kebab-case name).
• Remote source pins a full 40-char lowercase commit sha, and that commit is public + reachable.
• Regenerated .grok-plugin/plugin-index.json (python3 scripts/generate-plugin-index.py).
• python3 scripts/validate-catalog.py passes locally.
• python3 scripts/generate-plugin-index.py --check passes locally.
• homepage + clear description set; local plugins include README.md + .grok-plugin/plugin.json.
• License is stated (MIT, in the source repo).

Security

• No curl | bash, remote-code download/exec, or postinstall RCE.
• No reading/exfiltration of secrets, tokens, .env, or env vars.
• Hooks and MCP scope are least-privilege (none exist: skills-only plugin — one manifest, one SKILL.md, no executables).
• Network endpoints this plugin calls (and why): optionally reads https://www.dellight.ai/ in-session, solely to quote live pricing accurately when the user asks; no other endpoint. The skill explicitly bans web search, arbitrary fetches, and third-party APIs.
• Credentials/permissions it requires (and why): none. The skill unconditionally bans use of credentials, sessions, files, environment variables, uploads, and telemetry, and only reads engine output the user makes visible in the session.

Notes for reviewers

Skills-only instruction plugin from DELLIGHT.AI's official org. It reports how AI engines describe a brand using ONLY raw engine output visible in the session, with mandatory "not observed" / "not explicitly ranked" labels — it is designed to be unable to fabricate rankings. The source went through a four-round adversarial pre-publication review (truth boundaries, hostile-input isolation, credential bans) before this submission. A sibling PR adds sentinel-feed as its own single-entry change per the template rule; whichever lands second will be rebased to regenerate the index.

🤖 Generated with Claude Code