Launches a local traceability app verifying research tags and findings against transcript lines.
npx skills add https://github.com/borderux/recursica-knowledge --skill stu-borderux
Please help me install this Agent Skill. Skill: stu Source: https://github.com/borderux/recursica-knowledge/tree/main/agents/stu Command: npx skills add https://github.com/borderux/recursica-knowledge --skill stu-borderux
proposed, untagged lines, count mismatches, and thin evidence — pulled from BigQuery before handoff. - Attributed Edits: Records every human edit in edit_log against a verified identity (Buzz pubkey or email), while the agent itself never edits or approves anything. - Use Case: After Claire ingests and analyzes interview transcripts, Stu launches the explorer, hands the reviewer a localhost URL, and summarizes which proposed terms and findings need a human decision. ## Quick Start Ask Stu to launch the traceability explorer for your research project and report which proposed terms and findings are waiting for review.High-intent search queries and answers about installing and using this skill.
Launch the traceability explorer, which links every tag, dictionary term, and finding back to the exact transcript line it came from. A person reviews the evidence in the web app and approves or edits items, with each change recorded against their identity.
Run ./start.sh --user-email <email> --user-name "<name>" from the app directory, with slug, project, and service-account key configured in stu.env. The command is idempotent and prints a localhost URL to share with the reviewer.
No. Stu never edits data and never moves items from proposed to active — only a human does that through the app. On Claude Code this is enforced by connecting Stu to a read-only BigQuery server rather than relying on prompt instructions.
Stu needs the same per-project fenced BigQuery access as Claire, using a read-only MCP server scoped to one project's dataset. The service account and allowedDatasets fence provide the isolation, not the prompt.
opencode's agent model uses a coarse permission block with no per-tool allowlist, so it cannot enforce read-only BigQuery access. Stu's core guarantee — that he never writes to the data — would have nothing behind it on that platform.
JoniPraiaReport row counts per SX table from the local SQLite index.
kernex-devTranslate technical proposals into plain-language business summaries with costs and risks.
ai-analyst-labPresent cached schema details for tables, columns, row counts, and relationships from active.yaml and schema.md files.
HosungYouGuide qualitative coding, codebook development, and theme identification for social science research.