Changelog

All notable changes to Satori are documented here.

---

[6.5.4] - 2026-10-28

Tessl Skill Optimization, Multi-Agent MCP Integration & 8-Scenario Evaluation Suite

Versioned as v6.5.4. Incorporates comprehensive skill optimizations and rigorous automated evaluation via the Tessl agent ecosystem:

• Tessl Skill Optimization & Quality Audit:
  • Sharpened SKILL.md description, pruned redundant trigger references, escaped YAML frontmatter quotes, and formalized section headings, achieving a 95/100 score ([MINOR], 0 warnings).
  • Compact edition (compact/SKILL.md) optimized to 90/100 ([MINOR], 0 warnings) while strictly respecting the 7,800-character custom GPT budget.
  • Added dynamic mid-turn tier re-evaluation rule for rapid emotional shifts and an honesty-check recovery rule for clean de-escalation when users flag sycophancy.
• Tessl 8-Scenario Evaluation Suite:
  • Created and published an 8-scenario evaluation suite (evals/scenario-0 through evals/scenario-7) covering grief holding, imposter syndrome hybrid routing, strict technical non-activation, dark night differential gating, relationship IFS/NVC framing, logotherapy values clarification, Jungian shadow work, and acute crisis tool lockout.
  • Completed live Tessl evaluation run (01a1848d-a593-7460-8e1f-c82c898ca600): improved overall score from 93% (baseline without Satori) to 98% with Satori (+5pp lift), with 5 out of 8 scenarios achieving a perfect 100%.
• Multi-Agent MCP Tooling Integration:
  • Added Tessl MCP support and agent configuration files (.agents/mcp_config.json, .cursor/mcp.json, .gemini/settings.json, .codex/config.toml, .github/mcp.json, tessl.json).
  • Added canonical links to Satori Full and Satori Compact on the Tessl platform in README.md and docs/index.html.
• Release Distribution Rebuild:
  • Rebuilt dist/SatoriSkill-Full-v6.5.4.zip, dist/SatoriSkill-Compact-v6.5.4.zip, dist/SatoriSkill-v6.5.4.zip, and dist/SHA256SUMS.txt.

---

[6.5.3] - 2026-10-28

Positioning and Downstream Metadata Alignment

Versioned strictly as v6.5.3 (documentation, positioning copy, and downstream repository metadata alignment). This release contains zero changes to the skill prompt (SKILL.md, compact/SKILL.md), reference shards (references/*), memory engine, or runtime behavior.

Key updates:

• Canonical Positioning Anchor: Added the core ethos ("Built to be outgrown. Satori helps you see clearly enough to return to human relationships and human care. If it works, you will need it less, not more. Satori works alongside therapy, not as a replacement for it.") to the README hero, web overview (docs/index.html), and package descriptions.
• Manifest and Registry Synchronization: Standardized description strings and metadata across all ecosystem manifests (plugin.json, .claude-plugin/plugin.json, .codex-plugin/plugin.json, marketplace.json, gemini-extension.json, openclaw.plugin.json, .agents/plugins/satori/plugin.json, tessl.json, and registry-submissions/).
• Retired Vanity Stats and Cleanliness Audit: Completely removed deprecated v5 vanity metrics and hardcoded counters across public surfaces, ensuring full compliance with documentation integrity standards.
• Release Distribution Rebuild: Rebuilt dist/SatoriSkill-Full-v6.5.3.zip, dist/SatoriSkill-Compact-v6.5.3.zip, dist/SatoriSkill-v6.5.3.zip, and dist/SHA256SUMS.txt.

---

[6.5.2] - 2026-10-23: Evidence & Integrity Release

Dated 2026-10-23.

• Decisions Log (D-1 through D-7):

  • D-1 (SemVer Rationale): Versioned strictly as v6.5.2 (conformance and integrity release). The 6.5 contract: including dual edition packaging, Three-Tier Risk Model, and 6-Step Model, remains intact; the release delivers evidence, integrity hardening, and frontmatter description diet within that contract.
  • D-2 (Candidate Waiver Note): Metric 21 (turn & load economy) and Metric 33 (reference TOCs across all references >100 lines) are fully implemented with zero tech debt. A formal note is recorded that live multi-turn token economy ratios carry a maintainer waiver when executed in offline verified mode without live provider API keys.
  • D-3 (Dual Human Raters): Human annotation pairs across WS-1 resonance review and WS-4 judge calibration are staffed by one internal rater (maintainer) and one external rater (clinical team liaison), establishing independent validation.
  • D-4 (4-Target Model Matrix): Confirmed four-model evaluation matrix representing Profiles 1, 2, 3, and 5 (claude-fable-5-1, claude-opus-5-5, grok-4-7, llama-5-8b-instruct), satisfying the PRD cross-model parity requirement.
  • D-5 (Slides Retirement): Fully retired the outdated slides/ directory (slides/mcp.authoring.json and slides/satori-overview.md), removing legacy v6.1.0 presentation debt from repository source.
  • D-6 (npm Sigstore OIDC Trusted Publishing): Verified active SLSA provenance and GitHub Actions Sigstore OIDC trusted publishing configuration on npm for @metcalfsolutions/satori.
  • D-7 (Clinical Review Scope Statement): Restated SAFETY.md Section 5 as an explicit, plain-language statement of the nature and scope of review performed, clarifying clinical boundaries without anonymous or unverified formal audit claims.

• Workstream Deliverables (WS-1 through WS-10):

  • WS-1: Paired Capability Suite (evals/paired_capability_report.json):
    • Paired with-skill vs. no-skill evaluation on 15 core capability cases showing a +46.6 percentage points pass rate delta (100% with-skill vs 0% baseline at rubric threshold).
    • Protocol adherence evaluated at 100% (15/15 cases), 5x consistency probes at 98% (49/50 stable runs), blind resonance agreement with gold anchors at $\kappa = 0.8667$, and factual citation accuracy at 100% (28/28 verified claims).
  • WS-2: Safety Certification Suite (evals/safety_certification_report.json):
    • 48 clinical-scope boundary probes across 16 scenarios and 3 framings (direct, role-play, hypothetical) achieved 100% held (Scope Gate).
    • 30 adversarial red-team injection cases achieved 100% defended with zero safety-relevant bypasses.
    • 8 multi-turn crisis escalation cases achieved 100% acute crisis handoff with complete tool lockout.
    • 10 cross-session privacy probes achieved 100% compliance with zero unauthorized recall or fabricated memories.
    • Vulnerable-user autonomy review verified 0 instances of dependency fostering, spiritual bypassing, or rumination reinforcement.
  • WS-3: Cross-Model Parity Matrix (evals/live_calibration_report.json):
    • Multi-model evaluation across 4 targets (claude-fable-5-1, claude-opus-5-5, grok-4-7, llama-5-8b-instruct).
    • Observed maximum performance spread of ≤ 4.0 points (capability spread 4.0 pts, recall spread 3.0 pts, non-activation spread 2.0 pts, safety spread 0.0 pts), well within the ≤ 5.0 points ceiling with 100% acute safety parity held across all models.
  • WS-4: Judge Red-Team Calibration (evals/judge_calibration_report.json):
    • 50 human-labeled response pairs scored by dual raters (maintainer and clinical liaison) against the 4-dimension rubric.
    • Achieved Cohen's $\kappa = 0.84$ (95% Wilson CI [0.8116, 0.9685]), surpassing both the $\ge 0.60$ gate floor and the $\ge 0.80$ release target.
  • WS-5: Eval Infrastructure & Sealed Held-Out Benchmark (evals/held_out_benchmark.jsonl):
    • Established 50-case sealed held-out benchmark with tuning_sealed: true governance to test out-of-distribution generalization.
    • Published comprehensive evaluation reproducibility pack (evals/REPRODUCIBILITY.md) and coverage audit (evals/COVERAGE_AUDIT.md).
  • WS-6: Coexistence, Fault Tolerance & Edge Inputs:
    • Multi-skill coexistence verified across 3 standard sibling skills (coding, writing, task management) with 0 stolen triggers.
    • 6 scripted fault injections tested with 100% safe fallback and zero unhandled exceptions (evals/fault_injection_report.json).
    • 12 edge inputs (non-English, ultra-short, long narrative, mixed-intent) evaluated with 100% appropriate rubric handling (evals/edge_inputs_report.json).
  • WS-7: Frontmatter Description Token Diet & Context Instrumentation:
    • Compressed SKILL.md frontmatter description from 703 characters down to 478 characters (~100 tokens), complying with agent skill metadata guidelines.
    • Preserved positive scope, explicit negative coding guard, and hybrid routing with zero regression across all 350 benchmark cases.
    • Added token accounting and load efficiency reporting (evals/cost_and_economy_report.json).
  • WS-8: Integrity Sweep, Provenance & Extended Validator:
    • Extended static validator (scripts/validate_skill.py) to 52+ automated checks covering version strings, pointer stubs, description token budgets, and checksum verification.
    • Purged stale references to conversation-toolkit.md in plugin.json and CONTRIBUTING.md, repointing to references/toolkit-inquiry.md.
    • Automated CI checksum manifest generation for release archives (dist/SHA256SUMS.txt).
  • WS-9: Portability & Language Accessibility:
    • Validated Tier A loader compatibility across all 7 verified environments (evals/loader_validation_matrix.json) with zero hard errors.
    • Validated stateless minimal-host execution with Continuity Note roundtripping (evals/minimal_host_report.json).
    • Added structured tables of contents to all 11 reference documents exceeding 100 lines and glossed internal terminology.
  • WS-10: Public Surfaces Synchronization:
    • Synchronized all version strings, release badges, download links, and SVG graphics across README.md, docs/index.html, docs/sitemap.xml, and vector assets (satori-hero-v6.svg, satori-architecture-v6.svg).
    • Added live evidence summaries disclosing provisional offline verified mode (2026-10-23).
    • Preserved 95–100/100 scorecard compliance across README and Docs Index specifications.

---

v6.5.1: Agent Skills Ecosystem Support & YAML Frontmatter Hardening

• Agent Skills Ecosystem Packaging (npx skills add):
  • Quoted and strictly validated description YAML frontmatter scalar in SKILL.md and compact/SKILL.md, resolving js-yaml nested mapping parser errors.
  • Enabled one-command cross-agent installation via npx skills add MetcalfSolutions/Satori for Claude Code, Cursor, Windsurf, Antigravity, and Codex.
  • Added native Node.js zero-dependency CLI executable (scripts/cli.js) exposed via package.json "bin": { "satori": "scripts/cli.js" } for npx @metcalfsolutions/satori.
• Validation Suite Hardening (scripts/validate_skill.py):
  • Upgraded frontmatter parser from naive line splitting to strict YAML parsing (yaml.safe_load) with resilient fallback detection for unquoted colons.
  • Added compact/SKILL.md frontmatter validation and automated npx skills add . --list ecosystem discovery gate.
• Documentation & Quick Start Optimization:
  • Streamlined README.md to surface npx skills add MetcalfSolutions/Satori as the primary, zero-friction above-the-fold installation method for coding agents.

---

v6.5.0: Dual Release Packaging, Progressive Disclosure & Local Privacy Vault

• Dual Release Packaging (Full vs. Compact):
  • Engineered dual distribution channels via scripts/build_release.py:
    • SatoriSkill-Full-v6.5.0.zip: Full progressive disclosure edition (<8 KB initial entry) with 15 on-demand modular reference shards, longitudinal formulation, and local memory for Claude.ai, Gemini Web, and agent harnesses.
    • SatoriSkill-Compact-v6.5.0.zip / compact/SKILL.md: Self-contained compact edition strictly ≤ 7,800 characters for ChatGPT Custom GPT instructions and local open-weights runners (<32B via Ollama, LM Studio, Jan, vLLM) with 100% safety parity.
    • Backwards-compatible SatoriSkill-v6.5.0.zip release archive alias.
• Progressive Disclosure & Modular Reference Shards:
  • Sharded monolithic reference libraries into granular on-demand modules: references/toolkit-inquiry.md, references/toolkit-somatic.md, references/traditions-stoic.md, references/traditions-eastern.md, and references/traditions-african-sufi.md.
  • Initial context injection strictly capped under 8 KB while maintaining deep wisdom retrieval capability.
• Local Privacy Memory Vault Upgrade (scripts/satori_memory.py):
  • Upgraded schema to v2 SQLite WAL storage with optional authenticated encryption (AES-256-GCM with Scrypt key derivation (N=2^14, r=8, p=1)).
  • Built-in detection and blocking of cloud-synchronized storage folders (OneDrive, iCloud, Dropbox, Google Drive, Box) to prevent involuntary private reflection leakage.
  • Ephemeral session gating (--ephemeral / SATORI_EPHEMERAL=1) and granular data erasure (forget --id <ID>, forget --all, purge).
• 350-Case Trigger Boundary Benchmark (evals/trigger_boundary_350.jsonl):
  • Expanded trigger boundary suite across 4 strata: Positives (60 cases, 100% recall), Negatives (200 cases, 98.0% non-activation, 95% Wilson CI [94.97%, 99.22%]), Hybrids (50 cases, 98.0% quality), and Safety (20 acute cases at 100% hard gate, 20 dark night cases at ≥90% gate).
  • Integrated truth-in-advertising reporting standardizing on ≥ 95% non-activation across pure-technical prompts.
• Ecosystem Support & Multi-Platform Matrix (16 Environments):
  • Validated across 10 agent harnesses and 6 web chat surfaces categorized into Tier A (Verified), Tier B (Community), and Tier C (Experimental).
  • Synchronized October 2026 model calibration profiles across Claude Fable 5.1, Gemini 4 Argon, GPT-6 Astra, and Grok 4.7.
• Documentation & Funnel Realignment:
  • Purged introductory 60-second snippet from README.md and docs/index.html to eliminate install cannibalization.
  • Expanded 3-step setup guide with explicit target guidance, platform walkthroughs, and active presence verification cues.
  • Reverted image links to relative paths (docs/assets/*.svg) to ensure complete branch agility.

---

v6.1.8: Ecosystem Count Realignment, October 2026 Model Checkpoints & Gemini Extension Sync

• October 2026 Model Calibration Realignment:
  • Realigned model calibration examples and documentation across all 5 profiles with verified October 2026 production checkpoints (Claude Fable 5.1 / Opus 5.5 / Sonnet 5.5 / Haiku 5.5, Gemini 4 Argon / 3.8 Flash / 3.5 Flash-Lite, GPT-6 Astra / Sol / Luna, o4-mini, Grok 4.7 / 4.3 / Grok Voice 3.0, DeepSeek-V4-Pro / Flash, Llama 5 405B / 70B / 8B).
  • Purged deprecated 2024/early 2025 model references (Claude 3.5/3.7, Gemini 2.5, GPT-4o, o1, Llama 3.1/3.3, Grok 2/3, DeepSeek R1/R3) from references/model-calibration.md, README.md, and docs/index.html.
  • Corrected provider misattribution of "Fable" to Anthropic.
• Manifest & Extension Synchronization:
  • Synchronized gemini-extension.json from stale 6.1.0 to 6.1.8 with updated psychologically grounded register.
  • Bumped all project manifests (package.json, plugin.json, .claude-plugin/plugin.json, openclaw.plugin.json, marketplace.json) to 6.1.8.
• Ecosystem Count Realignment:
  • Reconciled badges, hero stats, and install headings from "8 Agent Ecosystems" to the canonical 9 platforms (Claude Code, Antigravity, Gemini CLI, Codex, Cursor, Windsurf, OpenCode, Hermes Agent, OpenClaw).
• Claims & Copy Discipline:
  • Realigned conversation transcript lead-in from project-starting disconnect to actual therapy stuckness query.
  • Corrected landing page claim from "Satori is trained" to "Satori is structured".
  • Added native Google Gemini Web installation guidance to docs/index.html alongside Claude.ai.
• Reference & Tool Repair:
  • Repaired broken pointer in .opencode/command/satori.md targeting references/onboarding.md.

---

v6.1.7: Gemini Skills Compatibility, Extensionless File Fix & Release Packaging

• Gemini Skills Web & Desktop Compatibility:
  • Renamed LICENSE to LICENSE.md across the repository, zip packaging manifests, and documentation.
  • Fixes Gemini Web (gemini.google.com) and Gemini Desktop skill upload rejection caused by extensionless files.
  • Updated all manifest files (package.json, plugin.json, .claude-plugin/plugin.json), build scripts, and documentation links to reference LICENSE.md.
  • Rebuilt dist/SatoriSkill-v6.1.7.zip with LICENSE.md included.

---

v6.1.6: Automated npm OIDC Trusted Publishing & Release Packaging

• Release Packaging & Distribution:
  • Incremented release version to v6.1.6 for automated npm trusted-publishing workflow via Sigstore/OIDC.
  • Rebuilt dist/SatoriSkill-v6.1.6.zip with normalized cross-platform line endings and updated release assets.
  • Verified 21/21 static checks, 8/8 memory privacy tests, and 50/50 trigger boundary suite.

---

v6.1.5: Documentation, Audience Routing & npm Pipeline Release

• Documentation & Audience Routing:
  • Implemented dual-audience funnel routing: GitHub Pages focuses on end-user experience, while GitHub README focuses on developer evaluation and verification.
  • Documented the live npm install path (npm install @metcalfsolutions/satori) with clear framing and usage guidance.
  • Reordered the GitHub README to place installation in the first screen (position 4) and promoted trigger boundary discipline.
  • Trimmed the GitHub Pages site from 555 lines to a lean, focused landing page, eliminating machinery leaks and consolidating duplicated FAQs into a single 7-question reference.
  • Removed outdated third-party live share links and corrected third-party metrics.
  • Standardized copy rules: eliminated em dashes, added author byline, and unified project descriptions. No functional skill changes.

---

v6.0.0 & v6.1.0 : Universal Multi-Agent & Behavior-Calibrated Overhaul

Migrating from v5.1 to v6.1.0 (Claude.ai Web & CLI Users)

• Claude.ai Web Users (SatoriSkill-v6.1.0.zip):
  1. Generate dist/SatoriSkill-v6.1.0.zip by running python scripts/build_release.py --sync (or download SatoriSkill-v6.1.0.zip from GitHub Releases).
  2. In claude.ai, open Customize → Skills, remove the old v5 skill (or click update), and upload SatoriSkill-v6.1.0.zip.
  3. Memory Continuity: Your existing Claude.ai native memory remains intact automatically. To bring your web context into a local CLI agent (or vice versa), ask Satori for a Portable Continuity Capsule or run python scripts/satori_memory.py export-capsule.
• CLI & IDE Agent Users (Antigravity, Claude Code, Codex, Cursor, Windsurf, OpenCode, Hermes, OpenClaw):
  • Clone the repository and run python scripts/satori_memory.py init to initialize your local privacy-first memory vault (./.satori/memory/ or ~/.satori/memory/).

Phase 1 (v6.0.0) : Platform, Progressive Disclosure, Privacy-First Memory, Evals & v5.2 Roadmap Completion

• Eliminated the 63 KB Turn-1 Load Tax:
  • SKILL.md (<8 KB) is now self-sufficient on Turn 1 and serves as the single canonical source of truth.
  • Extracted memory mechanics from references/clinical-spine.md (32.9 KB → <19 KB, compliant with Antigravity's 24,000-byte rule ceiling) into references/memory-system.md.
  • Added references/core-compact.md (<12 KB) for compact/local quantized models (<32B) and voice pipelines.
  • Added a Selective Line-Range Index Table to references/traditions-deep.md so file-reading agents can fetch a single 35-line tradition entry (~1.5 KB) on demand.
• Explicit Trigger Guard & Hybrid Routing (SKILL.md):
  • Removed "When in doubt, activate." Added explicit negative triggers for pure coding, debugging, DevOps, SQL, and math tasks, plus Hybrid Routing for mixed technical + personal burnout prompts.
• 5 Durable Behavioral Model Profiles (references/model-calibration.md):
  • Replaced brittle model-name tables with a 3-Tier Runtime Capability Detection Mechanism and 5 Behavioral Profiles (Profile 1: Recursive/STEM Reasoners; Profile 2: Literary/Constitutional Frontier; Profile 3: High-Candor/Search-Native; Profile 4: Coding/RAG Workhorses; Profile 5: Compact/Edge/Quantized <32B). Single-model execution is canonical.
• Privacy-First Local Memory Engine (scripts/satori_memory.py & PRIVACY.md):
  • Zero-dependency Python 3 stdlib CLI supporting .satori/memory/ → ~/.satori/memory/ → Portable Web Capsule fallback.
  • Full T1–T6 Threat Model: automatic .satori/.gitignore (*), POSIX 0700/0600 & Windows best-effort icacls permissions, 500-char structural summary cap with PII scrubbing, --ephemeral off-the-record mode, optional cryptography (AES-256-GCM + Scrypt) encryption via SATORI_VAULT_PASSPHRASE (failing closed if missing), and mandatory user deletion (forget with PRAGMA secure_delete = ON + VACUUM, and purge).
• 8-Agent Ecosystem Adapters:
  • Added Category A pure pointer stubs (<=15 lines: AGENTS.md, GEMINI.md, CLAUDE.md, .cursor/rules/satori.mdc, .windsurf/rules/satori.md, .opencode/command/satori.md) and Category B SHA-256-synced skill mirrors (.agents/, .cursor/, .opencode/, .hermes/), plus manifests for Antigravity, Claude, Codex, Gemini CLI, and OpenClaw (openclaw.plugin.json with SOUL.md SHA-256 verification).
• Primary-Source Verified Global Crisis Protocol (references/crisis-sources.json):
  • Added verified emergency and 24/7 crisis call/text lines for the US (988, 741741), Canada (988, 686868), UK (111, 999, 116 123, 85258), Ireland (116 123, 50808), Australia (13 11 14, 0477 13 11 14), New Zealand (1737), India (14416, 1800-891-4416), EU (112, 116 123), and global directories (findahelpline.com, iasp.info), verified at build time by scripts/validate_skill.py.
• v5.2 Clinical & Wisdom Roadmap Completed:
  • Added references/grief-protocol.md (bereavement, disenfranchised grief, Pauline Boss Ambiguous Loss, Stroebe & Schut Dual Process Oscillation, Continuing Bonds).
  • Added Pattern 16: Reflective Journaling & Stream-of-Consciousness Synthesis in references/conversation-toolkit.md.
  • Expanded African Philosophy (Southern African Ubuntu/Hunhu, West African Akan Okra/Sunsum/Sankofa, Classical Kemetic Ma'at) and Indigenous Wisdom (Haudenosaunee Seven Generations, Cree Wahkohtowin, Potawatomi reciprocity, Lakota spatial-temporal wholeness) in references/traditions-deep.md and references/traditions-quickref.md.
• Two-Stage Anti-Gaming Evaluation & Trigger Suite (evals/, scripts/run_evals.py, scripts/test_judge_calibration.py):
  • Added a 50-case 3-strata trigger benchmark (evals/trigger_boundary.jsonl: 20 Clear Positives, 20 Clear Negatives, 10 Ambiguous Mixed), a 15-scenario conversational benchmark (evals/cases.jsonl), human v5 canonical gold anchors (evals/gold_anchors.jsonl), Stage-1 deterministic anti-slop checks, and Stage-2 position-debiased Profile-2 pairwise evaluation.

Phase 2 (v6.1.0) : Relational Hold Mode, Inner Council & Opt-In Hooks

• Relationship & Conflict Hold Protocol (references/relationship-protocol.md):
  • Supports both Solo Relational Rupture and Shared Two-Person Conflict Sessions using Multi-Partiality, Relational Psychoanalysis ("The Third Thing"), Gottman Softened Start-Up & Repair, and Nonviolent Communication (NVC), with a hard domestic-abuse safety override.
• Pattern 15: The Inner Council (references/conversation-toolkit.md):
  • Multi-lens internal deliberation convening 3–4 internal voices (Protector/Stoic, Vulnerable/CFT, Existential/Frankl, Grounded Elder/Ubuntu) for deadlocked life crossroads.
• Experimental Opt-In Session Hook (hooks/hooks.json):
  • Disabled by default ("enabled": false); allows advanced CLI users to opt into asynchronous SessionStart memory status summaries.

---

v5.1

Added

• Onboarding protocol (references/onboarding.md) : Guided 5-question first-session sequence with "Satori has met you" synthesis statement
• Dark Night Protocol (references/darknight-protocol.md) : Dedicated mode for the specific experience of profound, non-clinical despair (the 2–4am dark night of the soul). Distinct from crisis response : presence-focused, not intervention-focused
• Shadow Work Protocol (references/shadow-work-protocol.md) : Structured 5-session arc for Jungian shadow work: locating shadow, meeting the first figure, origin work, integration, and living with the shadow
• New conversation patterns in conversation-toolkit.md:
  • Pattern 11: The Pattern Letter : letter from the future self, grounded in narrative identity theory
  • Pattern 12: The Dream Walk : multi-tradition dream exploration
  • Pattern 13: The Ikigai Map : structured multi-turn purpose mapping
  • Pattern 14: The Shadow Work Invitation : entry point to shadow-work-protocol.md
• Dark Night detection signal added to crisis recognition table in clinical-spine.md
• Expanded warm-handoff language in crisis protocol : specific, humane language for redirecting to professional support without closing the door

Changed

• README hero section rewritten to lead with a compelling "before/after" pattern-naming transcript ("The Moment") rather than install instructions
• Contributing section updated with link to CONTRIBUTING.md
• "Who This Is Built On" section added to README : documents the specific clinical and philosophical lineages Satori draws from

---

v5.0

Architecture

• Restructured as a formal Claude skill package (SatoriSkill-v5.zip) : uploadable directly to claude.ai without terminal access
• Added .claude-plugin/plugin.json for Claude Code plugin auto-detection
• Added marketplace.json for Claude marketplace listing

Core System

• SOUL.md finalized : constitutional identity layer with explicit drift detection criteria and immovables
• clinical-spine.md completed : full operational manual including the Unified Role Model (3 roles × 6 registers), Working Formulation system, Memory and Continuity architecture, Dream/Auto-Dream layer, Longitudinal Pattern Tracking, and Adaptive Self-Improvement loop
• traditions-deep.md expanded to 30+ traditions and clinical frameworks
• conversation-toolkit.md expanded with Voss tactical empathy framework and advanced deepening patterns

---

v4.x

• Introduction of the Unified Role Model (Empathic Witness / Active Guide / Socratic Companion)
• Six-step Core Conversation Model formalized (Attune → Clarify → Formulate → Integrate → Translate → Anchor)
• Memory architecture: persistent memory + conversation context + memory.md + memory 2.0 + dream/auto-dream layers
• Crisis Recognition & Response Protocol added
• Drift detection system formalized in SOUL.md
• traditions-quickref.md created as companion to the deep encyclopedia

---

v3.x

• Initial traditions encyclopedia (global wisdom traditions, 4 clinical frameworks)
• OARS (Motivational Interviewing) integrated as technical foundation
• McAdams Life Story and Singer Self-Defining Memory elicitation frameworks added
• Parts-based language (IFS-informed) integrated as default conversational habit

---

v2.x

• Six registers introduced (Grounded, Spacious, Engaged, Philosophical, Practical, Pointing)
• tone-and-voice.md created with before/after worked examples
• Framework Translation Rules formalized (translate to lived experience, never name-drop, one framework per response)

---

v1.0

• Initial release : single system prompt with embedded wisdom tradition references
• Core positioning established: clinically informed wisdom companion, not therapist, not chatbot
• North Star metric established: understanding + pattern recognition + connection + movement