Sixteen Buddhist masters, resurrected as citable AI personas

What it does

Master-skill is a Python-based AI persona framework that lets you converse with 16 Buddhist masters across four traditions — Indian, Chinese, Tibetan, and Theravada. Each persona is constrained to cite real sources (CBETA sutra numbers, BDRC/Toh IDs, SuttaCentral/PTS references) for every doctrinal assertion. It ships as an AgentSkills package for Claude Code, Cursor, Codex CLI, OpenCode, and Gemini CLI, plus a no-install browser chat at fojin.app.

The interesting bit

The project treats “don’t make things up” as an engineering problem, not a vibe: a HARD-GATE rule forbids fabricated source IDs, a two-stage review (doctrine, then style) records input hashes so any post-review edit forces a re-review, and a 240-fixture fidelity test suite — including crisis-referral and non-endorsement edge cases — runs offline citation audits on every scored answer. There’s even a /create-master generator that inherits the same guardrails, though it refuses to build personas for fictional figures.

Key highlights

• 16 masters spanning Madhyamaka, Yogācāra, Chan, Tiantai, Huayan, Pure Land, Kagyu, Gelug, and Thai Forest traditions, plus 4 teaching modes (multi-master comparison, debate, curriculum)
• Backed by FoJin’s corpus: 10K+ texts, 678K+ semantic vectors, 110K+ entity knowledge graph
• Offline fallback: sources/ ships canonical excerpts so citations survive even when the live retrieval service is down
• Honest about limits: the README explicitly labels all output as AI-synthesized, not the masters’ actual words, and points users in crisis to real hotlines
• Security posture is unusually thorough for a persona project — SHA-pinned CI, npm provenance, CodeQL, pip-audit

Caveats

• The two-stage review proves content wasn’t changed after review, but the README admits it can’t prove two independent humans actually did the reviewing
• Some Theravada and Tibetan sources lack per-passage numbering by nature; the citation contract explicitly allows this, so not every quote resolves to a precise locator
• One of the 16 masters (Daoxuan) and the comparison/debate modes are terminal-only, not available in the browser version

Verdict

Worth a look for anyone building grounded personas or working in digital humanities — it’s a rare example of roleplay with a citation audit trail. If you wanted a chatbot that freestyles spiritual advice, this project will disappoint you on principle.