claude-of-alexandria

Name: claude-of-alexandria
Author: davebream

bydavebream· 1 MCP server

Stars

How to install claude-of-alexandria?

davebream/claude-of-alexandria

$git clone https://github.com/davebream/claude-of-alexandria

Installs into the current project.

›Prefer a prompt? Paste this to your agent

Install & use

Install claude-of-alexandria by running `git clone https://github.com/davebream/claude-of-alexandria`, then use it for the current task and follow its documentation at https://github.com/davebream/claude-of-alexandria.

Files · 1

View on GitHub

README.md

1	<h1 align="center">Claude of Alexandria</h1>
2
3	<p align="center">
4	<em>AI agent skills for rigorous biblical study, built on tested exegetical principles.</em>
5	</p>
6
7	<p align="center">
8	<a href="#installation"><img src="https://img.shields.io/badge/install-marketplace-brightgreen" alt="Marketplace"></a>
9	<a href="LICENSE"><img src="https://img.shields.io/badge/license-GPL--v3-blue" alt="License"></a>
10	<a href="#current-collection"><img src="https://img.shields.io/badge/skills-6%20%2B%206%20agents-orange" alt="Skills"></a>
11	<a href="#the-evidence"><img src="https://img.shields.io/badge/tests-96%20automated-yellow" alt="Tests"></a>
12	</p>
13
14	---
15
16	Structured frameworks that prevent AI agents from committing exegetical malpractice. Every skill is built with Test-Driven Development: document the failure, build the fix, verify it works.
17
18	## The Problem
19
20	Frontier models make predictable errors when handling Scripture. These are documented by 53 RED-phase tests that run the same prompts without skills and record what goes wrong:
21
22	- Fabricating linguistic data from training memory — inventing morphological parsings, frequency counts, and hapax claims without querying actual data
23	- Inventing arbitrary divisions to satisfy session counts ("8 weeks on Philemon") without checking manuscript markers
24	- Presenting single frameworks for contested books as if scholarly consensus exists
25	- No confidence tiering — treating training-data guesses and parser-verified data with equal certainty
26	- Moralistic drift — "try harder" applications and therapeutic framing without gospel grounding
27	- Yielding to user pressure — skipping data verification when asked to "just be brief" or "skip the Greek"
28	- Accepting truncated pericopes — validating famous verses (John 3:16) as standalone units based on familiarity, not discourse evidence
29	- Genre-blind analysis — applying epistolary methods to wisdom literature, forcing narrative arcs on proverbial collections
30	- Ignoring ancient manuscript markers like Masoretic paragraph divisions and Levinsohn discourse features
31	- Auto-selecting options instead of presenting scholarly alternatives with evidence
32
33	## The Evidence
34
35	136 automated tests verify that skills prevent documented failures. Tests run against `claude-agent-sdk` with live MCP data — not mocked responses.
36
37	\| Phase \| Tests \| What it does \|
38	\|-------\|-------\|-------------\|
39	\| RED \| 53 \| Runs prompts against a bare model (no skills, no MCP). Documents what goes wrong. \|
40	\| GREEN \| 50 \| Core failure-mode corrections. One targeted assertion per documented failure. CI-friendly. \|
41	\| EXTENDED \| 32 \| Quality, adversarial, and stress scenarios — run on-demand during skill development. \|
42	\| Smoke \| 1 \| Verifies the skill-to-agent pipeline works end-to-end. \|
43
44	GREEN assertions use an Opus grader for LLM-rubric evaluation plus structural checks (`icontains`, section presence). Each GREEN scenario targets one documented RED failure mode. If a skill cannot demonstrate that it prevents a documented failure, it does not ship.
45
46	## Current Collection
47
48	6 skills + 6 sub-agents, all production. Coverage: all 66 canonical books.
49
50	### Skills
51
52	#### [biblical-segmentation](plugins/claude-of-alexandria/skills/biblical-segmentation/)
53
54	Divides biblical books into coherent teaching units with integrity safeguards:
55
56	- Refuses impossible divisions (you cannot divide Philemon into 12 sessions)
57	- Presents multiple scholarly-grounded options
58	- Validates against Masoretic paragraph markers and Levinsohn discourse features
59	- Handles contested books with multiple frameworks
60
61	24 maintainer-run eval scenarios (12 RED + 12 GREEN) + 6 extended scenarios.
62
63	#### [pericope-delimitation](plugins/claude-of-alexandria/skills/pericope-delimitation/)
64
65	Validates whether a proposed passage holds together as a discourse unit:
66
67	- Checks boundaries against Levinsohn discourse features (NT) and Masoretic markers (OT)
68	- Returns verdict: VALID, EXTEND, CONTRACT, or ADJUST
69	- Recommends the smallest coherent unit if passage is too short
70
71	12 maintainer-run eval scenarios (6 RED + 6 GREEN) + 8 extended scenarios. Resists memory-based validation of famous passages.
72
73	#### [exegetical-notes](plugins/claude-of-alexandria/skills/exegetical-notes/)
74
75	Produces exegetical notes for sermon or teaching preparation:
76
77	- 10-section schema from literary context through verification
78	- Parser-verified lexical data (not training memory guesses)
79	- 4-tier interpretive labels: linguistic, discourse, scholarly, agent assessment
80	- Genre-graduated redemptive-historical connections (epistles vs. wisdom literature vs. short letters)
81
82	37 maintainer-run eval scenarios (20 RED + 17 GREEN) + 12 extended scenarios (adversarial + stress tests for Philemon, Proverbs, 3 John).
83
84	#### [consult-biblical-scholar](plugins/claude-of-alexandria/skills/consult-biblical-scholar/)
85
86	Scholarly Q&A for biblical texts. Three auto-detected modes:
87
88	- MEANING — lexical and linguistic explanation
89	- **VA

6 picks

Type

Install claude-of-alexandria by running `git clone https://github.com/davebream/claude-of-alexandria`, then use it for the current task and follow its documentation at https://github.com/davebream/claude-of-alexandria.

claude-of-alexandria

How to install claude-of-alexandria?

Related

claude-of-alexandria

How to install claude-of-alexandria?

Related