chati-dev 4.0.11 → 4.1.1

package/framework/intelligence/context-engine.md

@@ -17,6 +17,14 @@ Four brackets define behavior based on remaining context:
 | **DEPLETED** | 25-40% | 3 layers (L0-L2). Heavy reinforcement (4.0%) — significant context loss. Chunk memory recovery. |
 | **CRITICAL** | <25% | 2 layers (L0-L1). Maximum reinforcement (5.0%) — last interactions before handoff. Full memory dump. |
 
+### Static/Dynamic PRISM Boundary
+
+The PRISM engine separates context into a **static prefix** (L0 Constitution + L1 Global rules) and a **dynamic suffix** (L2-L5, changes per turn). A `<!-- STATIC_BOUNDARY -->` marker separates them in the XML output.
+
+This enables prompt cache reuse — the static prefix is identical across turns within the same mode, so the LLM provider can cache it and only reprocess the dynamic portion.
+
+The static cache is invalidated when the governance mode changes (planning → build → deploy).
+
 ---
 
 ## Layered Context Injection
@@ -100,6 +108,19 @@ bracket = calculateBracket(100 - contextPercent)
 
 The orchestrator recalculates the bracket before each agent interaction and adjusts injection accordingly.
 
+### Token-Based Bracket Estimation
+
+Context remaining percentage is estimated from actual prompt content size rather than conversation turn count:
+
+```
+estimatedTokens = promptText.length / 4
+remainingPercent = (1 - estimatedTokens / providerLimit) * 100
+```
+
+This replaces the legacy heuristic (`turnCount / maxTurns`) which was inaccurate for sessions with large tool results or long agent outputs. The token-based approach scales correctly across providers (Claude 200K, Gemini 1M, Codex 128K).
+
+Fallback: If prompt text is unavailable, the turn-count heuristic is used.
+
 ---
 
 ## Autonomous Context Recovery

package/framework/intelligence/hooks-performance.md

@@ -0,0 +1,54 @@
+# Hook Performance Reference
+
+## Execution Model
+
+Each hook runs as a separate Node.js process via stdin/stdout JSON protocol.
+Hooks execute SEQUENTIALLY within the same event matcher.
+Claude Code spawns each hook as a child process, waits for JSON output, then proceeds.
+
+## Hook Registry (11 total)
+
+| Event | Matcher | Hook | Purpose |
+|-------|---------|------|---------|
+| UserPromptSubmit | `.*` | license-guard.js | License validation + session ping |
+| UserPromptSubmit | `.*` | prism-engine.js | Context injection (bracket, memory, frustration) |
+| UserPromptSubmit | `.*` | model-governance.js | Model assignment advisory |
+| PreCompact | `.*` | session-digest.js | Session state capture + auto-dream + daily digest |
+| PreToolUse | `Read` | read-protection.js | Block reading sensitive files |
+| PreToolUse | `Write\|Edit` | mode-governance.js | Write scope enforcement by phase |
+| PreToolUse | `Write\|Edit` | constitution-guard.js | Secret detection + 23 shell injection checks |
+| PreToolUse | `Write\|Edit` | undercover-guard.js | Sanitize framework terms from deliverables |
+| PreToolUse | `Write\|Edit` | style-guard.js | Em-dash and emoji enforcement |
+| PreToolUse | `Bash` | constitution-guard.js | Destructive command + shell injection checks |
+| PreToolUse | `Bash` | undercover-guard.js | Sanitize framework terms from git commits |
+| PreToolUse | `Bash` | style-guard.js | Em-dash and emoji in commit messages |
+
+## Latency Budget
+
+| Event | Hook Count | Expected Latency | Notes |
+|-------|-----------|-----------------|-------|
+| UserPromptSubmit | 3 | ~150ms | Runs before every user message |
+| PreCompact | 1 | ~50-200ms | Runs rarely (context compaction). Auto-dream may add 5-10s if triggered |
+| PreToolUse (Read) | 1 | ~30ms | Minimal overhead per file read |
+| PreToolUse (Write/Edit) | 4 | ~200ms | Heaviest path. 4 sequential hooks |
+| PreToolUse (Bash) | 3 | ~150ms | 23 regex checks in constitution-guard are negligible (<5ms) |
+
+## Performance Characteristics
+
+- **Fastest path**: Read (1 hook, ~30ms)
+- **Heaviest path**: Write/Edit (4 hooks, ~200ms)
+- **Most frequent**: UserPromptSubmit (3 hooks on every message)
+- **Rarest**: PreCompact (1 hook, only on context compaction)
+
+## Optimization Notes
+
+1. constitution-guard's 23 shell injection regex checks run in <5ms total (all patterns compiled once)
+2. undercover-guard and style-guard could be skipped in discover/plan phases (no code being written)
+3. If latency becomes noticeable, consider merging undercover-guard + style-guard into one hook (same event, similar logic)
+4. license-guard has a 5-minute ping throttle to avoid network latency on every message
+
+## When to Worry
+
+- If Write/Edit operations feel sluggish (>500ms added latency), profile hooks individually
+- If auto-dream triggers during PreCompact and takes >10s, consider increasing the trigger threshold from 100 to 150 entries
+- If UserPromptSubmit hooks add noticeable delay, license-guard's network call is the most likely culprit (has 3s timeout)

package/framework/intelligence/memory-layer.md

@@ -157,6 +157,22 @@ Where:
 
 ---
 
+## Memory Consolidation (Dream System)
+
+The `/chati dream` command triggers a 4-phase consolidation cycle inspired by Claude Code's autoDream:
+
+| Phase | Name | Action |
+|-------|------|--------|
+| 1 | Orient | Scan all agent MEMORY.md files, count entries, assess landscape |
+| 2 | Gather | Read all memories and recent session digests into unified list |
+| 3 | Consolidate | Deduplicate by content hash, merge cross-agent entries, increment evidence_count |
+| 4 | Prune | If total > 200 entries, remove by tier (cold first) then confidence then access_count. Archive removed to `.chati/memories/shared/archive/` |
+
+Consolidation is non-destructive — removed entries are archived, not deleted.
+The consolidation report includes: merged count, pruned count, archived count, remaining entries, contradictions found.
+
+---
+
 ## Agent Privacy
 
 Each agent has:
@@ -167,6 +183,37 @@ This prevents cross-contamination while enabling knowledge sharing.
 
 ---
 
+## Per-Turn Memory Extraction
+
+The memory extraction system automatically captures knowledge from agent output after each turn.
+
+**Extraction patterns:**
+- Decisions: "decided to", "chose", "selected" → type: decision, sector: episodic
+- Resolutions: "fixed by", "resolved", "solved" → type: resolution, sector: procedural
+- User corrections: "actually", "no, use" → type: user_correction, sector: reflective
+- Validated patterns: "confirmed", "verified", "works" → type: validated_pattern, sector: semantic
+- Error patterns: "error:", "failed because" → type: error_pattern, sector: episodic
+
+**Budget:** Maximum 3 memories per turn. Confidence starts at 0.5 (warm tier).
+**Deduplication:** Content is normalized and hashed to prevent duplicate entries.
+
+---
+
+## Daily Digest (KAIROS Lite)
+
+The `/chati digest` command displays cumulative daily activity logs stored in `.chati/memories/shared/daily/YYYY-MM-DD.md`.
+
+Each session appends an entry with:
+- Timestamp and pipeline phase
+- Current agent and completion stats
+- Agent scores (average and per-agent)
+- Decisions made and gotchas found
+- Session duration
+
+Digests are append-only (never overwritten) and accumulate throughout the day.
+
+---
+
 ## Integration with Context Engine
 
 | Bracket | Memory Level | Behavior |

package/framework/migrations/v4.0-to-v4.1.yaml

@@ -0,0 +1,165 @@
+# Migration: v4.0.x → v4.1.0 "Phoenix"
+# Intelligence Upgrade inspired by Claude Code source leak analysis
+
+from_version: "4.0.4"
+to_version: "4.1.0"
+created_at: "2026-04-02"
+description: >
+  Intelligence Upgrade: orchestrator v2 (thin prompt + CLI handler),
+  memory extraction, dream consolidation, daily digest, undercover mode,
+  style guard, bash security (23 checks), model-aware context windows,
+  brandbook HTML, frustration detection, static/dynamic PRISM boundary,
+  microcompact advisory, model fallback (opus→sonnet).
+breaking: false
+
+preconditions:
+  - type: version_gte
+    version: "4.0.0"
+    message: "Requires chati.dev v4.0.0 or later"
+  - type: constitution_articles
+    count: 19
+    message: "Requires Constitution with 19 Articles"
+
+backup:
+  directories:
+    - chati.dev/hooks
+    - chati.dev/orchestrator
+    - chati.dev/intelligence
+    - chati.dev/templates
+    - chati.dev/agents/plan
+    - chati.dev/context
+    - chati.dev/data
+  files:
+    - chati.dev/config.yaml
+    - chati.dev/constitution.md
+
+operations:
+  # --- New hooks ---
+  - type: copy_file
+    source: chati.dev/hooks/undercover-guard.js
+    destination: chati.dev/hooks/undercover-guard.js
+    strategy: replace
+    description: "Undercover mode: sanitize framework terms from deliverables"
+
+  - type: copy_file
+    source: chati.dev/hooks/style-guard.js
+    destination: chati.dev/hooks/style-guard.js
+    strategy: replace
+    description: "Style guard: em-dash and emoji enforcement"
+
+  # --- Hook settings (includes new hooks) ---
+  - type: copy_file
+    source: chati.dev/hooks/settings.json
+    destination: chati.dev/hooks/settings.json
+    strategy: replace
+    description: "Register undercover-guard and style-guard hooks"
+
+  # --- Updated hooks (security checks, model-aware context, etc.) ---
+  - type: copy_file
+    source: chati.dev/hooks/constitution-guard.js
+    destination: chati.dev/hooks/constitution-guard.js
+    strategy: replace
+    description: "23-point shell injection defense system"
+
+  - type: copy_file
+    source: chati.dev/hooks/prism-engine.js
+    destination: chati.dev/hooks/prism-engine.js
+    strategy: replace
+    description: "Model-aware context, frustration detection, microcompact advisory"
+
+  - type: copy_file
+    source: chati.dev/hooks/session-digest.js
+    destination: chati.dev/hooks/session-digest.js
+    strategy: replace
+    description: "Structured 8-section digest, auto daily digest, auto dream consolidation"
+
+  # --- New templates ---
+  - type: copy_file
+    source: chati.dev/templates/session-memory-tmpl.yaml
+    destination: chati.dev/templates/session-memory-tmpl.yaml
+    strategy: replace
+    description: "8-section structured session memory template"
+
+  - type: copy_file
+    source: chati.dev/templates/brandbook-html-tmpl.md
+    destination: chati.dev/templates/brandbook-html-tmpl.md
+    strategy: replace
+    description: "Brandbook HTML visual style guide spec"
+
+  # --- Orchestrator v2 ---
+  - type: copy_file
+    source: chati.dev/orchestrator/chati.md
+    destination: chati.dev/orchestrator/chati.md
+    strategy: replace
+    description: "Orchestrator v2: thin prompt + CLI handler"
+
+  # --- Constitution (Article XIII expanded) ---
+  - type: copy_file
+    source: chati.dev/constitution.md
+    destination: chati.dev/constitution.md
+    strategy: replace
+    description: "Article XIII: 7→10 points (dream, extraction, digest)"
+
+  # --- Intelligence docs ---
+  - type: copy_file
+    source: chati.dev/intelligence/memory-layer.md
+    destination: chati.dev/intelligence/memory-layer.md
+    strategy: replace
+    description: "Memory Consolidation, Per-Turn Extraction, Daily Digest sections"
+
+  - type: copy_file
+    source: chati.dev/intelligence/context-engine.md
+    destination: chati.dev/intelligence/context-engine.md
+    strategy: replace
+    description: "Static/Dynamic PRISM Boundary, Token-Based Estimation"
+
+  # --- Entity registry ---
+  - type: copy_file
+    source: chati.dev/data/entity-registry.yaml
+    destination: chati.dev/data/entity-registry.yaml
+    strategy: replace
+    description: "7 new entities (55→62)"
+
+  # --- Context files ---
+  - type: copy_file
+    source: chati.dev/context/root.md
+    destination: chati.dev/context/root.md
+    strategy: replace
+    description: "Version bump to 4.1.0"
+
+  # --- UX agent (brandbook HTML) ---
+  - type: copy_file
+    source: chati.dev/agents/plan/ux.md
+    destination: chati.dev/agents/plan/ux.md
+    strategy: replace
+    description: "Mandatory brandbook.html output (criterion #21)"
+
+  # --- Config (version + 10 new feature flags) ---
+  - type: update_yaml
+    path: chati.dev/config.yaml
+    changes:
+      version: "4.1.0"
+      installer_version: "4.1.0"
+    description: "Bump version and add Intelligence Upgrade feature flags"
+
+validation:
+  - type: file_exists
+    path: chati.dev/hooks/undercover-guard.js
+  - type: file_exists
+    path: chati.dev/hooks/style-guard.js
+  - type: file_exists
+    path: chati.dev/templates/session-memory-tmpl.yaml
+  - type: file_exists
+    path: chati.dev/templates/brandbook-html-tmpl.md
+  - type: yaml_field
+    path: chati.dev/config.yaml
+    field: version
+    value: "4.1.0"
+  - type: yaml_field
+    path: chati.dev/data/entity-registry.yaml
+    field: metadata.entity_count
+    value: 62
+
+rollback:
+  description: "Restore from backup created before migration"
+  command: "npx chati-dev upgrade --rollback"