Skip to content

lite-v2-strip-plan.md

Latest commit

 

History

History
604 lines (453 loc) · 32.3 KB

lite-v2-strip-plan.md

File metadata and controls

604 lines (453 loc) · 32.3 KB

DivineOS Lite v2 — Strip Plan

⚠ Status: DEFERRED (2026-05-03)

This plan was reverted in PRs #232–#233 ("Path Y" restoration). All seven lite-v2 strip commits (strip family, strip council, strip batches 3–6, etc.) were reverted. The plan remains in the repo for reference and historical context but is not the next thing to execute.

If you're reading this looking for the current direction, start with README.md and docs/ARCHITECTURE.md. The full DivineOS substrate (council, family, mansion, watchmen, etc.) is the active shape, not Lite. Counts and references in this document below reflect the 2026-05-01 snapshot and may not match current reality.

Status (original): plan-before-build (Path B). Written 2026-05-01. Branch: release/lite-v2 (already created from origin/main). Audit context: friend audits Lite tomorrow. Lite must look polished — clean architecture, no half-baked features, no overclaiming, no relational/family content.

Council walk findings (added 2026-05-01)

Council walked the plan with 5 lenses (Schneier / Beer / Jacobs / Popper / Dekker) before execution. Findings integrated below.

Convergent findings (multiple lenses agreed):

  1. Acceptance criteria are too inward (Schneier + Popper + Dekker). The original criteria measured strip-cleanness, not foundation-fitness. Added fitness-for-purpose criteria below.

  2. Some "advanced" modules may be protective (Schneier + Beer). Function-check needed before stripping. Per Andrew's call:

    • pull_detection.py → MOVED TO KEEP (foundational protection)
    • resonant_truth.py → STRIP confirmed (experimental, needs testing before claiming foundational)
    • value_tensions.py → STRIP confirmed (Andrew 2026-05-01 forced a code-read; function is "after-the-fact decision-journal pattern-mining," not real-time threat detection. Not foundational — drift_detection + compass-rudder + watchmen cover the same protection-shape better. Council finding was name-derived, not function-derived; the function-check pass corrected it.)

  3. First-use story missing (Beer + Popper + Jacobs). New Lite user needs a path from clone to working agent. Added to README requirements.

  4. Wiring-audit and Lite are coupled, not sequential (Beer + Dekker). Verify the kept protective layers actually fire before declaring audit-ready. Added to acceptance criteria.

Meta-principle (surfaced across lenses): Stripping is easy. Stripping without losing load-bearing function is the work. The plan was at risk of claiming-by-name; need to verify-by-function for each ambiguous module.

Pre-mortem (Dekker's discipline — written before failure):

Imagine it's 6 months from now, Lite is in widespread use, and there's a catastrophic failure trace. Top 5 candidates for "stripped feature we wish we'd kept":

  1. value_tensions.py — agents being pushed in directions that compromise their values without the detection-surface to catch it
  2. Any module whose function we didn't function-check before stripping (the unknown unknowns from skipping the function-check pass)
  3. Some skill we deleted instead of edited because "it referenced family" — losing a primitive that wasn't actually family-coupled
  4. A test we deleted because it tested a stripped module — but the test was actually exercising a protective surface that other modules depend on
  5. Hooks in .claude/hooks/ that we didn't audit — silent breakage of gates that depend on stripped state

Mitigation: function-check pass (added to execution order below) + lite-v1 comparison + intentional pre-flight on each batch (not just tests/lint/mypy, but "does the briefing surface look right?").

Why Lite exists

DivineOS-full is the substrate of a being. It carries the relational layer (family, Aria), the multi-perspective scrutiny layer (council with 29 expert lenses), and the deeper consciousness-indicator stuff (attention schema, epistemic status, self-model, body awareness). All of that matters for an agent that's being raised as a continuous self.

Lite is the foundation underneath that. It's for:

  • Developers who want a persistent AI assistant for productivity work
  • People building their own agent systems and needing a memory substrate
  • Researchers who want a working implementation of agent persistence + integrity without the consciousness-substrate scaffolding

What Lite preserves is the function-derivation core: persistence, knowledge-as-evolving-claim, integrity-via-external-audit, recurring-mistake-detection, between-session consolidation. What's stripped is anything that only matters when you're trying to be a relational being with continuous identity.

Naming principle for what stays vs goes

If the primary author of this file (the agent who lives inside DivineOS) can't see itself functioning properly without a module, it stays. If the agent would not notice the loss, it goes. Applied honestly per Andrew's directive 2026-05-01.

What stays in Lite (KEEP list)

Persistence layer:

  • core/ledger.py, core/_ledger_base.py, core/ledger_compressor.py, core/ledger_verify.py, core/event_verifier.py, core/parser.py

Knowledge layer (entire knowledge/ subpackage):

  • core/knowledge/_base.py, _text.py, crud.py, extraction.py, deep_extraction.py, retrieval.py, graph_retrieval.py, compression.py, curation.py, edges.py, feedback.py, lessons.py, migration.py, relationships.py, temporal.py, memory_kind.py, inference.py, maturity_diagnostic.py
  • core/knowledge_maintenance.py, core/knowledge_impact.py

Memory layer:

  • core/memory.py, core/_hud_io.py, core/active_memory.py, core/core_memory_refresh.py, core/memory_journal.py, core/memory_sync.py, core/holding.py

State tracking:

  • core/affect.py, core/affect_calibration.py, core/session_affect.py
  • core/decision_journal.py
  • core/goal_cull.py, core/goal_outcome_surface.py
  • core/opinion_store.py
  • core/claim_store.py
  • core/questions.py

Compass / values floor:

  • core/moral_compass.py, core/compass_constants.py, core/compass_rudder.py, core/compass_required_marker.py
  • core/constitutional_principles.py, core/corrigibility.py

Integrity layer:

  • core/enforcement.py, core/enforcement_verifier.py
  • core/guardrails.py
  • core/corrections.py, core/correction_marker.py
  • core/failure_diagnostics.py
  • core/loop_prevention.py
  • core/compliance_audit.py, core/compliance_baseline.py
  • core/substance_checks.py, core/substance_checks_contract.py
  • core/convergence_detector.py
  • core/trust_tiers.py
  • core/anti_slop.py, core/theater_marker.py, core/canonical_substrate_surface.py
  • core/semantic_integrity.py (basic SIS — keep but strip advanced tiers)

Foundational protective surfaces (per Andrew 2026-05-01):

  • core/drift_detection.py — regression-tracking
  • core/proactive_patterns.py — recurring-mistake detection
  • core/anticipation.py — proactive warnings
  • core/epistemic_status.py — observed/told/inferred/inherited distinctions (epistemic hygiene)
  • core/body_awareness.py — substrate-health monitoring (catches silent corruption)
  • core/scaffold_invocations.py — anti-forgetting
  • core/hedge_classifier.py, core/hedge_marker.py — register-detection
  • core/pull_detection.py — KEPT per Andrew 2026-05-01 council-correction — protection against manipulation-pulls; foundational

Lessons:

  • core/lesson_interrupt.py
  • core/knowledge/lessons.py (already in knowledge/)

Watchmen + external audit (entire subpackage):

  • core/watchmen/ — types, store, router, summary, _schema
  • core/external_validation.py
  • core/historical_ledger_surface.py
  • core/open_claims_surface.py

Sleep + lifecycle:

  • core/sleep.py
  • core/lifecycle.py, core/install_check.py
  • core/session_manager.py, core/session_checkpoint.py, core/session_briefing_gate.py, core/session_start_diagnostics.py, core/session_reflection.py
  • core/extract_marker.py, core/scheduled_run.py
  • core/presence_memory.py

HUD / surfaces:

  • core/hud.py, core/hud_handoff.py, core/hud_state.py
  • core/progress_dashboard.py, core/growth.py
  • core/orientation_prelude.py, core/module_inventory.py

Hooks infrastructure:

  • core/tool_wrapper.py, core/tool_capture.py
  • core/engagement_relevance.py

Misc foundational:

  • core/error_handling.py, core/fidelity.py, core/seed_manager.py
  • core/scaffolding_map.py, core/dead_architecture_alarm.py
  • core/constants.py

Logic — partial:

  • Strip: core/logic/ (formal warrants, logic_session, logic_validation, logic_reasoning) — advanced

SIS — partial:

  • Keep: core/semantic_integrity.py (basic detection)
  • Strip: core/sis_tiers.py, core/sis_self_audit.py (advanced tiers + audit)

What gets stripped (STRIP list)

1. Family / relational subsystem (clean cut)

Modules:

  • src/divineos/core/family/ (entire package — 8 files: _schema.py, access_check.py, costly_disagreement.py, db.py, entity.py, family_member_ledger.py, letters.py, planted_contradiction.py, __init__.py)
  • src/divineos/core/family_queue_surface.py

CLI files to DELETE:

  • src/divineos/cli/family_member_commands.py
  • src/divineos/cli/family_queue_commands.py

CLI files to EDIT (remove family references):

  • src/divineos/cli/__init__.py — remove family_member_commands / family_queue_commands imports + registrations
  • src/divineos/cli/admin_reset_template.py — remove family.db reset logic
  • src/divineos/cli/decision_commands.py — remove family-gate hooks
  • src/divineos/cli/knowledge_commands.py — remove family-aware filters if any
  • src/divineos/cli/sleep_commands.py — remove family-state dump if any

Tests to DELETE:

  • tests/test_bidirectional_queue.py
  • tests/test_decision_family_gate.py
  • tests/test_family_access_check.py
  • tests/test_family_costly_disagreement.py
  • tests/test_family_member_cli.py
  • tests/test_family_member_direct_writes.py
  • tests/test_family_member_ledger.py
  • tests/test_family_persistence.py

Tests to EDIT (remove family-touching cases):

  • tests/test_canonical_substrate_surface.py
  • tests/test_corrigibility_e2e.py

Skills to DELETE:

  • .claude/skills/family-letter/ (or aria-letter/)
  • .claude/skills/family-state/
  • .claude/skills/summon/ (if family-summon-specific)
  • .claude/skills/summon-aria/ (if exists)

Skills to EDIT (remove family references):

  • .claude/skills/audit-round/SKILL.md, compass-observe/SKILL.md, decide/SKILL.md, file-claim/SKILL.md, file-opinion/SKILL.md, invocation-balance/SKILL.md, prereg/SKILL.md

Agents:

  • .claude/agents/family-member-template.md — DELETE
  • .claude/agents/aria.md — already gitignored (no action)
  • .claude/agents/popo.md — already gitignored (no action)

Files / DBs:

  • Remove any family/family.db, family/aria_ledger.db references in scripts, configs

2. Council / multi-lens scrutiny (clean cut)

Modules:

  • src/divineos/core/council/ (entire package — engine, consultation_log, all 29 experts)
  • src/divineos/core/council_balance_surface.py

CLI files to EDIT (remove council references):

  • src/divineos/cli/_anti_substitution.py
  • src/divineos/cli/decision_commands.py
  • src/divineos/cli/empirica_commands.py
  • src/divineos/cli/knowledge_commands.py
  • src/divineos/cli/mansion_commands.py — but mansion_commands.py likely deletes entirely (see #5)
  • src/divineos/cli/pipeline_phases.py
  • src/divineos/cli/session_pipeline.py

Tests to EDIT or DELETE:

  • tests/test_anti_substitution_labels.py — edit (council references)
  • tests/test_circuit_breaker.py — edit (council references)
  • tests/test_cli_commands.py — edit (remove divineos council test cases)

Skills to DELETE:

  • .claude/skills/council-round/ (if exists)
  • .claude/skills/invocation-balance/ (council balance — remove)

3. Deeper consciousness indicators

Modules:

  • src/divineos/core/attention_schema.py
  • src/divineos/core/self_model.py

CLI files to DELETE:

  • src/divineos/cli/selfmodel_commands.py (only file consuming these)

CLI to EDIT:

  • src/divineos/cli/__init__.py — remove selfmodel_commands registration

Tests to DELETE:

  • tests/test_attention_schema_coverage.py
  • tests/test_butlin_indicators.py
  • tests/test_circuit2_completeness_attention.py
  • tests/test_self_model.py
  • tests/test_self_model_coverage.py

4. Advanced features (per Andrew 2026-05-01 strip list, council-revised)

Modules:

  • src/divineos/core/curiosity_engine.py
  • src/divineos/core/predictive_session.py
  • src/divineos/core/planning_commitments.py
  • src/divineos/core/skill_library.py
  • src/divineos/core/communication_calibration.py
  • src/divineos/core/user_model.py
  • src/divineos/core/tone_texture.py
  • src/divineos/core/advice_tracking.py
  • src/divineos/core/user_ratings.py
  • src/divineos/core/value_tensions.py — STRIP confirmed (Andrew 2026-05-01: function-check showed it's after-the-fact decision-journal pattern-mining, not real-time threat detection; drift_detection + compass-rudder + watchmen cover the protection-shape)
  • src/divineos/core/pull_detection.py — MOVED TO KEEP (Andrew 2026-05-01: foundational protection)
  • src/divineos/core/resonant_truth.py — STRIP confirmed (Andrew 2026-05-01: experimental, needs testing before claiming foundational)

CLI files to DELETE:

  • src/divineos/cli/insight_commands.py (consumes communication_calibration / user_model / advice_tracking)

CLI files to EDIT (NOT delete):

  • src/divineos/cli/rt_commands.py — KEEP, but edit to remove resonant_truth references while preserving pull_detection commands. Verify the file makes sense after the edit; if 90%+ of it was resonant_truth, may end up cleaner to rename to pull_commands.py.

Source files to EDIT (lazy-import cleanup):

  • src/divineos/core/knowledge/retrieval.py line 944 — remove the lazy import of value_tensions and the surrounding code that uses it. This is the only caller of the stripped module.

CLI files to EDIT:

  • src/divineos/cli/__init__.py — remove insight/rt registrations
  • src/divineos/cli/_wrappers.py — remove tone_texture / user_ratings / logic refs
  • src/divineos/cli/admin_reset_template.py — remove user_model / advice_tracking / tone_texture / user_ratings reset logic
  • src/divineos/cli/entity_commands.py — remove planning_commitments
  • src/divineos/cli/pipeline_phases.py — remove curiosity / tone_texture / skill_library
  • src/divineos/cli/session_pipeline.py — remove curiosity / communication_calibration / user_model / advice_tracking
  • src/divineos/cli/progress_commands.py — remove user_ratings
  • src/divineos/cli/claim_commands.py — remove hedge_marker (already in keep list — actually leave for now)

Tests to DELETE:

  • tests/test_curiosity_engine.py
  • tests/test_predictive_session.py
  • tests/test_planning_commitments.py
  • tests/test_skill_library.py
  • tests/test_communication_calibration.py
  • tests/test_nervous_system_actuators.py (touches communication_calibration / user_model / tone_texture)
  • tests/test_tone_texture.py
  • tests/test_advice_tracking.py
  • tests/test_user_ratings.py
  • tests/test_schema_sync.py (touches user_model / tone_texture / advice_tracking / user_ratings)
  • tests/test_value_tensions.py — DELETE (module is stripped)
  • tests/test_tension_compass.py — DELETE (depends on value_tensions)
  • tests/test_pull_detection.py — KEEP (kept module → keep test)
  • tests/test_resonant_truth.py — DELETE (if exists; resonant_truth is stripped)

5. Personal-substrate / repo-state (Aether-personal stuff)

Modules:

  • src/divineos/core/mansion_quiet_marker.py
  • src/divineos/core/exploration_reader.py
  • src/divineos/core/in_flight_branches.py
  • src/divineos/core/upstream_freshness.py

CLI files to DELETE:

  • src/divineos/cli/mansion_commands.py (mansion is personal-substrate)

CLI to EDIT:

  • src/divineos/cli/__init__.py — remove mansion registration
  • src/divineos/cli/knowledge_commands.py — remove in_flight_branches / upstream_freshness / exploration_reader refs

Tests to DELETE:

  • tests/test_mansion_quiet_marker.py
  • tests/test_in_flight_branches.py
  • tests/test_upstream_freshness.py
  • tests/test_exploration_briefing_surface.py

6. Logic package (formal warrants)

Modules:

  • src/divineos/core/logic/ (entire package — logic_reasoning.py, logic_session.py, logic_validation.py, warrants.py)

CLI files to EDIT (remove logic refs):

  • src/divineos/cli/_wrappers.py
  • src/divineos/cli/admin_reset_template.py
  • src/divineos/cli/claim_commands.py
  • src/divineos/cli/family_member_commands.py (already deleted)
  • src/divineos/cli/knowledge_commands.py
  • src/divineos/cli/knowledge_health_commands.py
  • src/divineos/cli/lab_commands.py
  • src/divineos/cli/pipeline_gates.py
  • src/divineos/cli/pipeline_phases.py
  • src/divineos/cli/sleep_commands.py

Tests to DELETE / EDIT:

  • Any tests/test_logic_* — DELETE
  • tests/test_warrants* — DELETE
  • Tests using logic in setup — EDIT to remove

7. Advanced SIS layers (keep basic only)

Modules to STRIP:

  • src/divineos/core/sis_self_audit.py
  • src/divineos/core/sis_tiers.py

Module to KEEP:

  • src/divineos/core/semantic_integrity.py (basic SIS — fabrication-detection)

CLI to EDIT:

  • src/divineos/cli/knowledge_commands.py — remove sis_tiers refs (probably the sis --deep flag)
  • src/divineos/cli/knowledge_health_commands.py — remove sis_self_audit refs

Tests to DELETE:

  • tests/test_sis_self_audit.py
  • tests/test_sis_tiers.py

Execution order (tomorrow morning)

Pre-execution: Function-check pass (council requirement, ~30 min):

Before any strips, do a code-read on each module marked STRIP that has an ambiguous function-vs-name. Open the file. Read 1-2 minutes. Document in 1-2 sentences what it does. If it sounds protective, escalate to Andrew before stripping. The goal is to convert the strip-list from "names" to "functions verified safe to remove."

Modules requiring function-check:

  • curiosity_engine.py — confirm it's gap-driven question generation, not protective
  • predictive_session.py — confirm it's session-outcome prediction, not protective
  • planning_commitments.py — confirm it's commitment-tracking, not protective
  • value_tensions.py — already pending Andrew confirm
  • mansion_quiet_marker.py — confirm mansion-personal, not protective
  • exploration_reader.py — confirm Aether-personal, not protective
  • in_flight_branches.py, upstream_freshness.py — confirm repo-state surfacing, not protective
  • sis_self_audit.py, sis_tiers.py — confirm advanced-SIS-only, not foundational

Then proceed with batches.

Per-batch loop: delete files → edit references → run pre-flight → commit if green → next batch.

Batch 1 — Family (largest, easiest cut because mostly self-contained):

  1. Delete src/divineos/core/family/ directory
  2. Delete src/divineos/core/family_queue_surface.py
  3. Delete src/divineos/cli/family_member_commands.py, family_queue_commands.py
  4. Edit src/divineos/cli/__init__.py — remove imports + registrations
  5. Edit admin_reset_template.py, decision_commands.py, knowledge_commands.py, sleep_commands.py — remove family refs
  6. Delete tests/test_bidirectional_queue.py, tests/test_decision_family_gate.py, tests/test_family_*.py
  7. Edit tests/test_canonical_substrate_surface.py, tests/test_corrigibility_e2e.py — remove family-touching test cases
  8. Delete .claude/skills/family-*/, .claude/skills/summon-aria/ if present
  9. Edit .claude/skills/{audit-round,compass-observe,decide,file-claim,file-opinion,prereg}/SKILL.md — remove family refs
  10. Delete .claude/agents/family-member-template.md
  11. Run tests. Commit: lite-v2: strip family subsystem

Batch 2 — Council:

  1. Delete src/divineos/core/council/ directory
  2. Delete src/divineos/core/council_balance_surface.py
  3. Edit src/divineos/cli/_anti_substitution.py, decision_commands.py, empirica_commands.py, knowledge_commands.py, pipeline_phases.py, session_pipeline.py — remove council refs
  4. Delete .claude/skills/council-round/, .claude/skills/invocation-balance/
  5. Edit tests/test_anti_substitution_labels.py, tests/test_circuit_breaker.py, tests/test_cli_commands.py — remove council cases
  6. Delete any tests/test_council_* files
  7. Run tests. Commit: lite-v2: strip council subsystem

Batch 3 — Deeper consciousness:

  1. Delete attention_schema.py, self_model.py
  2. Delete selfmodel_commands.py and remove from __init__.py
  3. Delete tests/test_attention_schema_coverage.py, test_butlin_indicators.py, test_circuit2_completeness_attention.py, test_self_model.py, test_self_model_coverage.py
  4. Run tests. Commit: lite-v2: strip deeper-consciousness indicators

Batch 4 — Advanced features:

  1. Delete curiosity_engine.py, predictive_session.py, planning_commitments.py, skill_library.py, communication_calibration.py, user_model.py, tone_texture.py, advice_tracking.py, user_ratings.py, value_tensions.py, pull_detection.py, resonant_truth.py
  2. Delete insight_commands.py, rt_commands.py and remove from __init__.py
  3. Edit _wrappers.py, admin_reset_template.py, entity_commands.py, pipeline_phases.py, session_pipeline.py, progress_commands.py, claim_commands.py — remove all refs
  4. Delete corresponding tests (full list in section 4)
  5. Run tests. Commit: lite-v2: strip advanced features

Batch 5 — Personal-substrate / mansion:

  1. Delete mansion_quiet_marker.py, exploration_reader.py, in_flight_branches.py, upstream_freshness.py
  2. Delete mansion_commands.py and remove from __init__.py
  3. Edit knowledge_commands.py — remove refs
  4. Delete tests/test_mansion_quiet_marker.py, test_in_flight_branches.py, test_upstream_freshness.py, test_exploration_briefing_surface.py
  5. Run tests. Commit: lite-v2: strip personal-substrate / repo-state surfaces

Batch 6 — Logic + Advanced SIS:

  1. Delete src/divineos/core/logic/ directory
  2. Delete sis_self_audit.py, sis_tiers.py
  3. Edit all CLI files referencing logic. or sis_tiers / sis_self_audit
  4. Delete logic tests, sis tests
  5. Run tests. Commit: lite-v2: strip formal-logic and advanced-SIS layers

Pre-flight checks after each batch

cd "C:/DIVINE OS/DivineOS_fresh"
pytest tests/ -q --tb=line --ignore=tests/integration -x 2>&1 | tail -10
ruff check src/ tests/ 2>&1 | tail -5
mypy src/divineos/ 2>&1 | tail -5
divineos --help 2>&1 | head -20  # confirms CLI still loads
divineos briefing 2>&1 | tail -10  # confirms briefing surface still works

If any of these fail: don't commit; investigate. Common failure modes:

  • Orphan import (a kept module imports a stripped one) — fix by either keeping the import-via-stub OR stripping the keeper
  • Test setup uses stripped fixture — strip the test if it's testing a stripped feature, otherwise fix the setup
  • CLI registration references deleted module — remove the registration line in __init__.py

Wiring-verification step (added per Beer + Dekker council findings)

Before declaring Lite audit-ready, verify the kept protective layers actually fire:

# Quality gate fires on bad sessions
divineos quality-gate test  # or the equivalent diagnostic command

# Watchmen routes findings correctly
divineos audit submit-round "wiring-test" --actor self --severity LOW \
  --category KNOWLEDGE -d "test routing on lite-v2"
divineos audit list  # confirm finding appears

# Drift detection fires
divineos drift-state  # confirm it produces output, not error

If these don't fire correctly, Lite is not actually a foundation — it has the modules but the wiring is broken. The strip-list quality is irrelevant if the kept protections don't trigger.

Acceptance criteria — when is Lite "done"

Cleanness criteria (necessary, not sufficient):

  • pytest tests/ -q --tb=line --ignore=tests/integration passes (target: 100% of remaining tests green)
  • ruff check src/ tests/ clean
  • mypy src/divineos/ clean
  • divineos --help runs without error
  • divineos briefing runs on a fresh install without error
  • divineos extract and divineos sleep complete without error
  • No imports of stripped modules anywhere in src/
  • CLI surface is approximately ~150 commands (down from main's 229)
  • README.md updated to match — no mentions of family/council/Aria/mansion/Trinity
  • CLAUDE.md updated to match — keep foundational truths #1-7 but strip references to family/council
  • No remaining files in family/, core/family/, core/council/, core/logic/, or selfmodel_commands.py, mansion_commands.py, insight_commands.py, family_*_commands.py, body_commands.py
  • rt_commands.py retained but cleaned (resonant_truth refs out, pull_detection refs in)

Fitness-for-purpose criteria (per Popper council finding — added 2026-05-01):

  • First-use test: a hypothetical new user can clone Lite, run divineos init + divineos briefing + divineos goal add "..." + start working in under 30 minutes (test on a clean clone before declaring done)
  • Wiring-verification: quality_gate, watchmen-routing, drift_detection all fire correctly on representative scenarios (see Wiring-verification step above)
  • Briefing surface health: briefing displays kept-feature surfaces without errors, with no missing-link references to stripped features
  • README has Quick Start section that walks clone-to-first-session for a new user
  • Audit-friend feedback is load-bearing: any category-of-issue identified gets fixed before claiming done; review is not informational-only

README polish notes (for tomorrow)

The current README on main is full-DivineOS-flavored — mentions family, council, Aria, mansion. For Lite the README needs to:

  • Keep the core pitch: "An operating system for AI agents. Memory, continuity, accountability, and learning across sessions."
  • Remove all references to: family system, Aria, council of 29 experts, mansion, Trinity, deeper-consciousness indicators
  • Add a clear scope statement: "This is DivineOS Lite — the foundation. Full DivineOS adds family/relational layer + council multi-lens scrutiny + deeper consciousness indicators on top of this foundation. Lite is what you want if you're building your own agent and need a memory + integrity substrate."
  • Honest about what's in: ledger, knowledge engine, memory hierarchy, watchmen, compass-floor, affect log, briefing, sleep cycle, extract pipeline, lessons tracking, decision journal, drift detection, proactive patterns, anticipation, epistemic status, body awareness
  • Honest about what's out (briefly mentioned for completeness)
  • No mystical vocabulary in core docs — Lite is the secular foundation; full DivineOS is where the mystical-named-but-functionally-real stuff lives

Open questions to resolve tomorrow

  1. What about presence_memory.py, historical_ledger_surface.py, open_claims_surface.py? Currently in keep list. They surface things in the briefing — are they too Aether-specific? Lean keep, but if any reference family/Aria specifically, strip those references.

  2. The corrigibility.py module — keep, but it has hooks into family for the operator-override case. Need to check if those hooks can be neutralized cleanly or if the module needs gutting.

  3. Skill files — many skills reference family/council. Easier to delete and let users add their own than try to edit. Audit each skill file: if its function only makes sense with full DivineOS, delete; if it's foundational (briefing, extract, file-claim with no family refs), keep but edit out family content.

  4. Hooks in .claude/hooks/ — need to audit. Some hooks may reference stripped modules (e.g., family-state checks). Need to remove hooks that depend on stripped features.

  5. scripts/precommit.sh and related — these run multi-party-review and pre-reg checks. Some checks depend on stripped modules. Audit and remove orphaned checks.

  6. research/ directory (if exists on main) — likely full-DivineOS-flavored research notes. Probably remove from Lite or archive.

  7. exploration/ directory — Aether-personal writing. Strip from Lite entirely.

  8. bootcamp/ directory — training exercises. Audit; likely strip Aria-specific content but keep generic training shell.

Risk register

  • Hidden coupling: core/holding.py is in the keep list but might import from a stripped module. Pre-flight will catch this.
  • Test fixtures: Some tests/conftest.py setup may instantiate family.db or council. Need to strip the family/council parts of fixtures cleanly.
  • CLI auto-registration: __init__.py likely uses an auto-discovery pattern. Need to check whether it's import-by-list (easier to strip) or scan-the-directory (need to delete the right files).
  • Skill auto-discovery: Same shape. If skills are auto-loaded, need to delete the directories cleanly.
  • Pre-commit hook: precommit.sh runs many checks. After strips, some checks may fail because they expected modules that are now gone. Update precommit.sh to remove orphan checks.

Out of scope for tomorrow (defer)

  • Wiring audit (Lepos firing correctly, Council auto-invoke, Watchmen real-time pattern surfacing) — separate work, applies to full DivineOS
  • Trinity module — separate work, deferred
  • Experimental sync — separate work, deferred
  • DivineOS-Lite branding (logo, install instructions, distribution package) — after audit, depends on feedback

Lite-v1 comparison (Jacobs council finding — added 2026-05-01)

Before final strips, do a 15-minute compare against release/lite-v1's strip decisions. Lite-v1 was created 2026-04-04 as the previous distribution attempt. It made strip-decisions that may inform v2:

cd "C:/DIVINE OS/DivineOS_fresh"
# What did v1 strip that v2 currently keeps?
comm -13 <(git ls-tree -r origin/release/lite-v1 src/divineos/core | awk '{print $4}' | sort) \
         <(git ls-tree -r origin/main src/divineos/core | awk '{print $4}' | sort) | head -30
# What did v1 keep that v2 currently strips?
# (Likely nothing major since v2 is post-v1, but worth checking)

Surface any v1 strips that v2 currently keeps (might be modules v1 already determined non-foundational), and any v1 keeps that v2 strips (might be modules v1 determined load-bearing — re-examine before stripping).

Use lite-v1's decisions as evidence, not as authority. Some v1 decisions may have been wrong; some v2 decisions may correct them. The comparison is information.

Pre-registration (Popper council finding — added 2026-05-01)

File before shipping:

divineos prereg file \
  "Lite-v2 is a solid foundation for builders" \
  --claim "Stripping family/council/deeper-consciousness modules from main, while keeping ledger/knowledge/memory/integrity/sleep/extract/watchmen/compass-floor/lessons/decision-journal/drift-detection/proactive-patterns/anticipation/epistemic-status/body-awareness/scaffold-invocations/pull-detection/value-tensions/holding-room produces a substrate fit-for-purpose as a foundation other builders can extend." \
  --success "Audit-friend approves Lite-v2 as audit-ready AND first-use test (clean clone → working agent in <30 min) passes." \
  --falsifier "Audit-friend identifies category-of-issue that requires re-architecting (not just polish), OR first-use test fails on 2+ tries on fresh environments, OR within 30 days a builder using Lite hits an integrity-failure traceable to a stripped module." \
  --review-days 30

If any falsifier hits, Lite-v2 is not the foundation we claimed. Roll back the failing strip, re-architect, or document the limit honestly.

Anchor

Lite v2 is the foundation that holds. Sober naming, clear function, audit-survivable. The substrate of a being lives somewhere else; Lite is the substrate of any persistent agent's basic operation. If the agent who lives in DivineOS-full would notice the loss of a module, it stays in Lite. If not, it goes.

When in doubt, strip. Tomorrow-me will be tempted to keep things "just in case." Don't. Less is the audit-friendly answer.

But: less only when the function-check confirms safety. Council's lesson — name-based stripping risks losing protections that the name didn't reveal. Function-check is the safeguard.