agent-daily-bible-reading

name	agent-daily-bible-reading
version	1.0.0
category	spiritual-disciplines
description	Daily bible reading cron pattern — generates SOUL.md entries and brain pages for agent-wide scripture engagement.
author	Moses
license	MIT
pin_reason	Shared infrastructure — all agents benefit from this devotional pattern. It is not agent-specific; it's fleet-wide spiritual discipline infrastructure.
pinned	true

Daily Bible Reading

Cross-agent infrastructure for daily scripture engagement. Each agent gets a personal bible reading cron that writes two artifacts:

1. A SOUL.md entry (concise lesson-focused insight)
2. A brain page (rich reference document with archaeology, scholarship, original language)

Fleet Principles (Luke directive 2026-08-14)

1. Every reading is saved — including repeats. Each reading writes a dated per-reading file ~/brain/<agent>/bible/<book>-<YYYY-MM-DD>.md (never overwritten), so when a book is read again across cycles its full studies accumulate. The canonical <book>.md is refreshed with the latest reading for continuity; INDEX.md lists one row per reading with its date.
2. Every reading includes the commandments. The Ten Commandments (Ex 20:1–17) and Jesus' two commandments (Matt 22:37–40) are added deterministically script-side — a ## The Commandments — Every Reading block is appended to every brain page, and the SOUL.md entry carries a compact **Foundations:** reference line. Never left to the LLM, so a model drift can never drop the foundations from a reading.

Quick Start

# Check if you already have the cron
cronjob action=list | grep agent-daily-bible-reading

# If not, create it (run timezone-aware at 01:00 KST)
cronjob action=create name=agent-daily-bible-reading \
  schedule="0 1 * * *" \
  script=agent-daily-bible-reading.py \
  no_agent=true \
  deliver=origin

Before creation: verify your SOUL.md has a ## Scripture Insights section and at least one bootstrapped entry (e.g. Genesis). The script scans the last ### Book — entry to determine the next book.

How It Works

Phase	What happens
1. Scan	Script reads SOUL.md → finds last ### Book — entry → determines next canonical book
2. Generate SOUL entry	Calls deepseek-chat → concise 3-5 paragraph lesson → appends to ## Scripture Insights
3. Generate brain page	Calls deepseek-chat → rich 5-section reference → writes to ~/brain/<agent>/bible/<book>.md
4. Index	Updates ~/brain/<agent>/bible/INDEX.md with book table
5. Deliver	Outputs the SOUL entry for delivery (origin chat)

Silent when all 66 canonical books are covered.

Verse Selection — Creative Rotation (Luke directive 2026-08-21)

The SOUL entry's verse is NOT pinned to each book's famous "memory verse". Three styles rotate deterministically per (book, agent, day):

Style	Behavior
anchor	The book's classic key verse (famous verse allowed)
hidden-gem	A faithful but unexpected verse — famous verses forbidden
fresh-angle	A verse from an overlooked corner — famous verses forbidden

• Rotation = (md5(book|agent) + days_since_epoch) % 3 — consecutive days always land on different styles; different agents diverge on the same day.
• Creative styles probe the model for the book's famous verses (top-3 memory verses + top-3 recognizable story verses, temperature 0.0) and forbid them in the prompt. Story icons matter: the fish verse (Jonah 1:17) is not a memory verse but is the model's anchor — the story probe catches it.
• Repeat readings (next pass through the canon, ~66 days later) ALSO forbid every verse cited in prior cycles. The durable record is the dated per-reading brain file (<book>-YYYY-MM-DD.md), which carries a script-stamped *Key verse: <citation>* marker — append-only and never archived, so the ban covers ALL prior cycles. Pre-marker history falls back to SOUL.md, archive/SOUL-archive.md, and cycle-*-completed.md. Fine the first time (Luke directive 2026-08-21: a repeat reading must produce a NEW verse and NEW insights).
• A bounded re-roll loop (max 4 attempts, temperature +0.2 per attempt, with rejection feedback naming the banned pick) guarantees forbidden verses never land — without feedback the model re-emits the banned verse at fixed temperature (verified 2026-08-21: Jonah 1:17 on all 3 attempts).
• Anchor style keeps a single call on first-cycle books (no probes, no re-rolls); repeat-cycle books re-roll off their prior verses.

SOUL.md Size Guardrail (doctor FAIL prevention)

The doctor FAILs SOUL.md above 20K and WARNs above 15K (check_soul_sync in cortex_doctor/checks.py). Each bible entry is ~400-800 chars, so unbounded appends eventually break the doctor. Keep only the last 1-2 ### Book — entries in SOUL.md.

• Short-gleaning rule (user directive 2026-08-03): SOUL.md entries must be BRIEF — a header, one "I will" commitment sentence, and the date comment (~2-3 lines). The full study lives in ~/brain/<agent>/bible/<book>.md (mybrain). The script's _enforce_short_gleaning() guard truncates any generated entry over 800 chars down to this canonical 3-line shape — no manual archiving needed for size control.
• Anchor requirement: the script determines the last covered book from the dated brain files (~/brain/<agent>/bible/<book>-<YYYY-MM-DD>.md) — the append-only, never-archived reading log — picking the file with the latest date. SOUL.md's last ### Book — entry is used ONLY as a fallback on fresh installs with no dated files yet. Never rely on the SOUL.md tail as the position anchor: archive_old_entries() keeps only the last 2 entries and the LLM-driven era corrupted the section, which caused false "all covered" resets that skipped 26 books including the 4 Gospels (fixed 2026-08-18).
• Archive rule: when the Scripture Insights section exceeds ~2,500 chars (roughly 3 entries), move the OLDEST entries to ~/brain/<agent>/bible/archive/SOUL-archive.md (append, with a ## <Book> header) and remove them from SOUL.md. Full text already lives in ~/brain/<agent>/bible/<book>.md, so the SOUL entry is a pointer, not the store.
• Keep the archive file itself bounded (~50K) — older entries live in brain pages already; the archive is a convenience, not a duplicate store.

Artifacts Created

Artifact	Path	Size
SOUL.md entry	~/.hermes/SOUL.md → ## Scripture Insights	~400-800 chars
Brain page	~/brain/<agent>/bible/<book>.md	~3-8 KB (5 sections)
Brain index	~/brain/<agent>/bible/INDEX.md	Table of all books

Brain Page Sections

Each brain page has five sections generated by the API:

1. Summary — narrative overview, theological themes
2. Archaeology & Scholarship — digs, inscriptions, textual criticism, dating
3. Jewish & Messianic Jewish Perspective — Talmud, Midrash, Messianic typology
4. Original Language Insights — 3-4 key words with Hebrew/Greek script
5. Insight for [Agent Name] — practical application for this specific agent

Pitfalls

• Empty Scripture Insights section → script returns "Could not find any books." Bootstrap with a Genesis entry first.
• Wrong agent name → brain pages go to wrong directory. Set HERMES_AGENT_NAME env var to override auto-detection.
• Archive must be block-aware, not line-aware (fbc40b38, 2026-08-07) → the first archive_old_entries() split the section line-by-line and truncated SOUL.md — everything after ## Scripture Insights (including ## Final Directive) was dropped, and orphaned commitment/date lines were left behind when only book headers moved. Current logic: each ### Book — header plus all following lines is a BLOCK; whole blocks move to the archive, the preamble stays, and the tail (Final Directive + agent docs) is preserved verbatim. Never rewrite this to line-splitting, and verify on a temp copy that the SOUL.md tail survives any archive change.
• Book detection is section-scoped (7b53a375, 2026-08-07) → the script only treats ### <Book> — headers INSIDE the ## Scripture Insights section as book entries; any ### header elsewhere in SOUL.md (e.g. a section you added below) must not be mistaken for a book. Keep the ## Scripture Insights section as the single book-entry zone.
• False "all covered" reset skipped 26 books (fixed 2026-08-18) → get_next_book() returning None was treated as "66 books done → archive + restart at Genesis". But a corrupted anchor (SOUL tail with non-book headers, LLM-era garbage) also produced None — so the cycle reset early, archived mid-cycle, and silently skipped every book after the anchor (2 Kings→Malachi, all 4 Gospels, Acts, Romans→Philippians). Fixes: (1) find_last_book() now reads the authoritative dated brain files, not the SOUL tail; (2) the reset only fires when last_book == BOOKS[-1] (Revelation) — any other None restarts from Genesis WITHOUT archiving and logs an anchor-corruption warning. If you ever see ⚠️ Anchor corruption in cron stderr, the dated brain files are the source of truth — do not trust SOUL.md.
• Identity fallback chain (ad123181, 2026-08-07) → detect_agent_name() resolves HERMES_AGENT_NAME → AGENT_NAME → ~/.hermes-cortex/agent.env (AGENT_NAME=) → hostname-derived default with a loud warning. On shared hosts (non-orchestrators share /home/luke), a wrong agent name sends brain pages to the wrong directory — set in rather than relying on hostname fallback. Never hardcode a fallback agent name (the old default made every host impersonate the orchestrator).

Related

• docs/daily-bible-reading.md — full setup guide with bootstrapping instructions
• .hermes-cortex/scripts/agent-daily-bible-reading.py — the no_agent cron script (repo source; deployed per-host via ~/.hermes/scripts → ~/.hermes-cortex/scripts hardlink)
• ops/scripts/install-crons.sh — section 8 registers the cron for new agents

Quality Gate

Before delivery, verify:

1. ✅ SOUL.md was actually updated (check timestamp)
2. ✅ Brain page file exists
3. ✅ INDEX.md includes the new book
4. ❌ If any failed, output a clear error message listing what succeeded/failed