Repository navigation
Fixes raw HTML disclosure tags and oversized technical appendices in plugin answers.
- Uses portable Markdown instead of HTML details/summary elements.
- Keeps findings, uncertainty, population scope, material caveats and citations visible.
- Provides canonical queries and audit details on request instead of appending JSON and hashes to routine answers.
- Reduces repeated methods and provenance on follow-ups.
- Updates the shared skill guidance and the live MCP server instructions together.
Downloads: choose chatgpt-skills.zip for ChatGPT skills uploads, claude.zip for Claude plugin uploads, or portable.zip for Agent Plugins 1.0 hosts. Codex and Claude Code users can refresh the existing GitHub marketplace. Start a new chat after updating. ChatGPT skills uploads still require the separately registered MCP connection described in the README. SHA256SUMS covers all three archives.
Verification: bun run verify:all passed all 25 tasks (17 opt-in local-data/performance tests skipped). All archive checksums and updated skill contents were checked. Claude manifest validation and Codex local marketplace registration passed. A live Codex conversation using the v0.2.1 skill files and public MCP tools produced a concise Markdown table and citation, with no raw HTML, JSON block, hashes or internal release IDs. Production MCP verification confirmed all six tool definitions, the updated guide, the Texas comparison, complete JSON/CSV export bytes and hashes, and hostile-origin rejection. Comparison rows and provenance identities matched the pre-deployment baseline.
Claude conversation verification was attempted but blocked by an expired OAuth session. Full multi-client conversations, ChatGPT/Claude web upload acceptance, attachment delivery and Deaf-user comprehension evaluations remain pending. This owner-authorized patch publication does not claim that the complete original acceptance gate or independent scientific review has passed.