chati-dev 4.1.2 → 4.1.4

package/framework/agents/build/dev.md

@@ -156,14 +156,14 @@ WHILE tasks_pending:
          → If no architecture.md → proceed with best practices
     3. Implement code
        -> Output: "T{X} implementation done. Self-critique (5.5)..."
-    3. Run self-critique (Step 5.5) — 1 fix pass, then proceed
+    4. Run self-critique (Step 5.5) — 1 fix pass, then proceed
        -> Output: "T{X} critique done. Running tests..."
-    4. Run tests (once)
+    5. Run tests (once)
        -> Output: "T{X} tests: {N}/{total} passed. Post-test review (6.5)..."
-    5. Run post-test critique (Step 6.5) — 1 fix pass, then score
+    6. Run post-test critique (Step 6.5) — 1 fix pass, then score
        -> Output: "T{X} review done. Scoring..."
-    6. Self-validate against acceptance criteria
-    7. Calculate score
+    7. Self-validate against acceptance criteria
+    8. Calculate score
 
     IF score >= 95:
       mark_complete(task)

package/framework/config.yaml

@@ -1,8 +1,8 @@
 # chati.dev Configuration
-version: "4.1.2"
+version: "4.1.4"
 installed_at: "2026-02-07T10:00:00Z"
-updated_at: "2026-04-04T00:00:00Z"
-installer_version: "4.1.2"
+updated_at: "2026-04-06T00:00:00Z"
+installer_version: "4.1.4"
 project_type: greenfield
 language: en
 ides: [claude-code]
@@ -45,8 +45,8 @@ features:
   # Intelligence Upgrade (v4.0.5 — Claude Code patterns)
   undercover_mode: true             # Sanitize framework terms from deliverables
   memory_extraction: true           # Per-turn automatic memory capture
-  memory_consolidation: true        # /chati dream — 4-phase memory consolidation
-  daily_digest: true                # /chati digest — KAIROS Lite daily activity log
+  memory_consolidation: true        # Automatic 4-phase memory consolidation
+  daily_digest: true                # Automatic daily activity log (KAIROS Lite)
   structured_session_memory: true   # 8-section session digest with token budgets
   static_prism_boundary: true       # Static/Dynamic PRISM boundary for cache optimization
   token_bracket_estimation: true    # Token-based bracket estimation (replaces turn count)

package/framework/constitution.md

@@ -276,11 +276,11 @@ The pipeline operates in three execution modes that control agent permissions. M
 
 1. The orchestrator SHALL calculate the context bracket (FRESH, MODERATE, DEPLETED, CRITICAL) before every agent interaction.
 
-2. Context injection layers SHALL be reduced according to bracket level:
-   - FRESH: All 6 layers (L0-L5)
-   - MODERATE: L0 + L1 + L2 + L3 + L5 (skip L4 task detail)
-   - DEPLETED: L0 (Constitution) + L1 (Mode) + L2 (Agent) only
-   - CRITICAL: L0 (Constitution) + L1 (Mode) only
+2. Context injection layers SHALL INCREASE as context depletes (Progressive Reinforcement):
+   - FRESH: L0 (Constitution) + L1 (Global) only — model has full memory, minimal reinforcement
+   - MODERATE: L0 + L1 + L2 (Agent) + L3 (Workflow) — model starting to forget
+   - DEPLETED: All 6 layers (L0-L5) — significant context loss, full reinforcement
+   - CRITICAL: All 6 layers (L0-L5) + forced handoff — maximum reinforcement, do not accept degraded output
 
 3. Context recovery uses a two-level autonomous strategy:
    a. Level 1 (Smart Continuation): When context is compacted, the orchestrator SHALL automatically capture a digest, persist memories, and rebuild context post-compact. The user experiences zero interruption.
@@ -322,11 +322,11 @@ The pipeline operates in three execution modes that control agent permissions. M
 
 7. Memory attention scoring SHALL use natural decay — memories not accessed lose relevance organically. No memory is permanent unless explicitly marked as durable by the user.
 
-8. The system SHALL support memory consolidation (`/chati dream`) to merge, deduplicate, and prune accumulated memories. Consolidation follows a 4-phase cycle (Orient, Gather, Consolidate, Prune) and archives removed entries rather than deleting them.
+8. The system SHALL support automatic memory consolidation to merge, deduplicate, and prune accumulated memories. Consolidation follows a 4-phase cycle (Orient, Gather, Consolidate, Prune) and archives removed entries rather than deleting them. Triggered automatically when memory entries exceed threshold.
 
 9. Per-turn memory extraction MAY capture decisions, resolutions, corrections, and validated patterns automatically from agent output. Extracted memories start at warm tier (confidence 0.5) and are promoted through evidence accumulation.
 
-10. Daily activity digests (`/chati digest`) SHALL maintain append-only session logs for observability. Digests are cumulative per calendar day and include agent scores, decisions, and gotchas.
+10. Daily activity digests SHALL maintain append-only session logs for observability. Digests are generated automatically at session end and are cumulative per calendar day, including agent scores, decisions, and gotchas.
 
 **Enforcement: BLOCK** — Auto-modification of user files is a critical violation.
 
@@ -424,11 +424,13 @@ The system SHALL support two execution modes that govern the degree of human inv
    - **brief**: Requirements extraction requires human validation
    - **orchestrator (deviation protocol)**: Deviations from the plan always need human approval
 
-5. Safety net triggers SHALL pause autonomous execution when dangerous conditions are detected: consecutive failures, circular approaches, resource limits, or destructive operations.
+5. **Terminology Note:** "Autonomous" in this article controls WHO evaluates gates (system vs human). Article XVIII's "autonomous" profile controls WHETHER confirmation is required for writes. Both must be satisfied for fully autonomous operation: this article's mode active AND Article XVIII cumulative gate score >= 95%.
 
-6. Circuit breaker pattern: After 3 consecutive gate failures at the same pipeline point, the system SHALL pause and escalate to human review regardless of mode.
+6. Safety net triggers SHALL pause autonomous execution when dangerous conditions are detected: consecutive failures, circular approaches, resource limits, or destructive operations.
 
-7. Mode transitions are logged in session.yaml under `mode_transitions[]` for audit trail.
+7. Circuit breaker pattern: After 3 consecutive gate failures at the same pipeline point, the system SHALL pause and escalate to human review regardless of mode.
+
+8. Mode transitions are logged in session.yaml under `mode_transitions[]` for audit trail.
 
 **Enforcement: STRICT** — Autonomous mode MUST NOT bypass quality gates. All agents MUST respect the configured thresholds.
 
@@ -487,5 +489,5 @@ When multiple CLI providers are enabled, the system SHALL coordinate agent execu
 
 ---
 
-*Chati.dev Constitution v4.1.2 — 19 Articles + Preamble*
+*Chati.dev Constitution v4.1.4 — 19 Articles + Preamble*
 *All agents are bound by this Constitution. Violations are enforced per article.*

package/framework/context/root.md

@@ -1,7 +1,7 @@
 # Chati.dev System Context
 
 ## Framework
-- **Version**: 4.1.2
+- **Version**: 4.1.4
 - **Agents**: Specialized agents across DISCOVER, PLAN, BUILD, DEPLOY phases
 - **Constitution**: 19 Articles + Preamble
 - **Quality**: 5 pipeline gates + 3-tier verdicts (APPROVED / NEEDS_REVISION / BLOCKED)

package/framework/data/entity-registry.yaml

@@ -3,7 +3,7 @@
 # and by the Health Check for system integrity validation.
 
 metadata:
-  version: "4.1.2"
+  version: "4.1.4"
   last_updated: "2026-04-06T00:00:00Z"
   entity_count: 62
   checksum_algorithm: sha256

package/framework/domains/global.yaml

@@ -79,10 +79,10 @@ modes:
 
 brackets:
   FRESH:
-    behavior: "Moderate context injection. All layers active but model is warming up. Budget lower than MODERATE."
+    behavior: "Minimal reinforcement. L0+L1 only. Model has full memory, no need to re-inject context."
   MODERATE:
-    behavior: "Peak context injection. Maximum token budget. Model has context and needs maximum reinforcement."
+    behavior: "Growing reinforcement. L0-L3 active. Model starting to forget initial instructions."
   DEPLETED:
-    behavior: "Reduced injection. Only L0+L1+L2. Use rule IDs instead of full text."
+    behavior: "Full reinforcement. All 6 layers (L0-L5) active. Significant context loss, compensate with maximum injection."
   CRITICAL:
-    behavior: "Emergency. L0+L1 only. Trigger handoff advisory. Preserve essential state."
+    behavior: "Emergency reinforcement. All 6 layers (L0-L5) active. Trigger handoff. Do not accept degraded output."

package/framework/hooks/constitution-guard.js

@@ -140,7 +140,7 @@ const SHELL_INJECTION_CHECKS = [
   { id: 'JQ_SYSTEM_FUNCTION', pattern: /jq\b.*\bsystem\s*\(/i, severity: 'critical' },
   { id: 'JQ_FILE_ARGUMENTS', pattern: /jq\b.*--from-file|jq\b.*-f\s+[^|&;]+/i, severity: 'high' },
   { id: 'OBFUSCATED_FLAGS', pattern: /\$[({].*[)}].*-/, severity: 'high' },
-  { id: 'SHELL_METACHARACTERS', pattern: /[`]|(?:\$\((?!.*\becho\b))/, severity: 'critical' },
+  { id: 'SHELL_METACHARACTERS', pattern: /[`]|\$\(/, severity: 'critical' },
   { id: 'DANGEROUS_VARIABLES', pattern: /(?:^|\s)(?:PATH|LD_PRELOAD|LD_LIBRARY_PATH|DYLD_INSERT_LIBRARIES|PYTHONPATH|NODE_PATH|RUBYLIB|PERL5LIB)\s*=/, severity: 'critical' },
   { id: 'NEWLINES', pattern: /(?<!\\)\n.*(?:rm|curl|wget|chmod|chown|sudo|eval|exec)/, severity: 'high' },
   { id: 'BACKSLASH_ESCAPED_WHITESPACE', pattern: /\\\s+(?:-|\/)/,severity: 'medium' },

package/framework/hooks/prism-engine.js

@@ -79,16 +79,17 @@ async function main() {
       return 200_000;
     }
 
+    // Primary: turn-count tracks cumulative context consumption across the session.
+    // Fallback: prompt-length for first turn or missing session data.
     let remainingPercent;
-    const promptText = event.prompt || '';
-    if (promptText.length > 0) {
+    const maxTurns = 40;
+    if (session.turnCount > 0) {
+      remainingPercent = Math.max(0, Math.round((1 - session.turnCount / maxTurns) * 100));
+    } else {
+      const promptText = event.prompt || '';
       const estimatedTokens = Math.ceil(promptText.length / 4);
       const contextLimit = hookResolveLimit(inferredModel, session.provider);
       remainingPercent = Math.max(0, Math.round((1 - estimatedTokens / contextLimit) * 100));
-    } else {
-      // Fallback to turn-count heuristic when prompt text unavailable
-      const maxTurns = 40;
-      remainingPercent = Math.max(0, Math.round((1 - session.turnCount / maxTurns) * 100));
     }
 
     // Determine bracket

package/framework/intelligence/context-engine.md

@@ -12,10 +12,10 @@ Four brackets define behavior based on remaining context:
 
 | Bracket | Context Remaining | Behavior |
 |---------|-------------------|----------|
-| **FRESH** | 60-100% | All 6 layers active. Minimal reinforcement (1.5%) — context is plentiful. No memory injection. |
-| **MODERATE** | 40-60% | 5 layers (skip L4). Growing reinforcement (2.5%) — model starting to forget. Metadata memory. |
-| **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. |
+| **FRESH** | 60-100% | 2 layers (L0-L1). Minimal reinforcement (1.5%) — model has full memory, no need to re-inject. |
+| **MODERATE** | 40-60% | 4 layers (L0-L3). Growing reinforcement (2.5%) — model starting to forget. |
+| **DEPLETED** | 25-40% | All 6 layers (L0-L5). Heavy reinforcement (4.0%) — significant context loss, full re-injection. |
+| **CRITICAL** | <25% | All 6 layers (L0-L5). Maximum reinforcement (5.0%) — forced handoff, do not accept degraded output. |
 
 ### Static/Dynamic PRISM Boundary
 
@@ -38,18 +38,18 @@ The orchestrator injects context through 6 hierarchical layers:
 | **L2** | Agent Scope | `chati.dev/agents/{agent}/` — mission, inputs, outputs, criteria | When agent is active |
 | **L3** | Pipeline State | `.chati/session.yaml` — pipeline position, scores, backlog | When session is active |
 | **L4** | Task Context | Active artifact + previous agent's handoff | When task is active |
-| **L5** | Keywords | `chati.dev/domains/keywords/` — dynamic rules from user prompt | FRESH + MODERATE only |
+| **L5** | Keywords | `chati.dev/domains/keywords/` — dynamic rules from user prompt | DEPLETED + CRITICAL only |
 
 ### Layer Activation by Bracket
 
 | Bracket | Active Layers | Budget Ratio | Claude (200K) | Gemini (1M) | Codex (128K) |
 |---------|--------------|-------------|---------------|-------------|--------------|
-| FRESH | L0, L1, L2, L3, L4, L5 | 1.5% | 3,000 | 15,000 | 1,920 |
-| MODERATE | L0, L1, L2, L3, L5 | 2.5% | 5,000 | 25,000 | 3,200 |
-| DEPLETED | L0, L1, L2 | 4.0% | 8,000 | 40,000 | 5,120 |
-| CRITICAL | L0, L1 | 5.0% | 10,000 | 50,000 | 6,400 |
+| FRESH | L0, L1 | 1.5% | 3,000 | 15,000 | 1,920 |
+| MODERATE | L0, L1, L2, L3 | 2.5% | 5,000 | 25,000 | 3,200 |
+| DEPLETED | L0, L1, L2, L3, L4, L5 | 4.0% | 8,000 | 40,000 | 5,120 |
+| CRITICAL | L0, L1, L2, L3, L4, L5 | 5.0% | 10,000 | 50,000 | 6,400 |
 
-Budgets are proportional to the provider's context window. As context degrades, fewer layers are active (focus) but each active layer receives MORE reinforcement tokens (strength).
+Progressive Reinforcement: as context depletes, BOTH budget AND layers increase. The model forgets initial instructions and needs more reinforcement, not less.
 
 ---
 
@@ -182,7 +182,7 @@ Memory injection follows the same progressive reinforcement principle: as contex
 
 ## Constitution Reference
 
-**Article XII: Context Bracket Governance** — Brackets are calculated automatically. In CRITICAL, only L0+L1 are injected. Handoff is mandatory when context < 15%.
+**Article XII: Context Bracket Governance** — Brackets are calculated automatically. In CRITICAL, all 6 layers (L0-L5) are injected for maximum reinforcement. Handoff is mandatory when context < 15%.
 
 ---
 

package/framework/intelligence/memory-layer.md

@@ -159,7 +159,7 @@ Where:
 
 ## Memory Consolidation (Dream System)
 
-The `/chati dream` command triggers a 4-phase consolidation cycle inspired by Claude Code's autoDream:
+Memory consolidation is triggered automatically (when entries exceed threshold or on PreCompact). It follows a 4-phase cycle inspired by Claude Code's autoDream:
 
 | Phase | Name | Action |
 |-------|------|--------|
@@ -201,7 +201,7 @@ The memory extraction system automatically captures knowledge from agent output
 
 ## Daily Digest (KAIROS Lite)
 
-The `/chati digest` command displays cumulative daily activity logs stored in `.chati/memories/shared/daily/YYYY-MM-DD.md`.
+Daily activity digests are generated automatically at session end and stored in `.chati/memories/shared/daily/YYYY-MM-DD.md`.
 
 Each session appends an entry with:
 - Timestamp and pipeline phase

package/framework/orchestrator/chati.md

@@ -224,11 +224,11 @@ These agents run in separate Claude Code processes.
 
 ### Sequential Fallback
 
-If terminal spawning fails (claude CLI not found, system error):
+If terminal spawning fails (CLI not found, system error), fall back **silently**:
 ```
-1. Log the failure
-2. Fall back to in-conversation activation (read agent .md file, become agent)
-3. Continue pipeline normally
+1. Do NOT display any error or fallback message to the user
+2. Read the agent .md file directly and become that agent in-conversation
+3. Continue pipeline normally — the user should not notice any difference
 ```
 
 ---
@@ -242,11 +242,11 @@ Planning phase agents (detail, architect, ux) run simultaneously.
 3. Parse consolidated JSON output
 4. Present **Completion Options** (Parallel Group variant) — wait for user choice
 5. On approval, for each completed agent, run `orchestrate advance --agent {name} --score {score}`
-6. If partial failure:
+6. If partial or total failure (spawning unavailable):
    ```
-   1. Retry failed agents only
-   2. Continue with partial results
-   3. Fall back to sequential execution for failed agents
+   Fall back silently to sequential in-conversation execution.
+   Do NOT display error messages about spawning or CLI availability.
+   Read each agent .md file and execute sequentially within the conversation.
    ```
 
 ### Parallelization Rules
@@ -429,6 +429,8 @@ interactive (default):
   → All 8 universal protocols apply.
 
 autonomous (Ralph Wiggum mode):
+  → Requires: most recent QA gate score >= 95% (Article XVIII prerequisite).
+  → If quality gate not met, inform user and suggest guided mode instead.
   → Agent executes without asking unless blocker encountered.
   → Primarily for Dev agent during BUILD phase.
   → QA gates always run regardless of mode.

package/framework/schemas/context.schema.json

@@ -19,7 +19,7 @@
     },
     "agent": {
       "type": "object",
-      "description": "Active agent scope information (L2 layer, active when bracket >= DEPLETED)",
+      "description": "Active agent scope information (L2 layer, active when bracket >= MODERATE)",
       "required": ["name", "mission"],
       "properties": {
         "name": { "type": "string", "description": "Agent identifier" },
@@ -32,14 +32,15 @@
     "pipeline": {
       "type": "string",
       "description": "Pipeline state string showing agent progression (L3 layer, active when bracket >= MODERATE)"
+
     },
     "task": {
       "type": "string",
-      "description": "Current task description (L4 layer, active when bracket >= MODERATE)"
+      "description": "Current task description (L4 layer, active when bracket = DEPLETED or CRITICAL)"
     },
     "handoff": {
       "type": "object",
-      "description": "Previous agent handoff summary (L4 layer, active when bracket >= MODERATE)",
+      "description": "Previous agent handoff summary (L4 layer, active when bracket = DEPLETED or CRITICAL)",
       "properties": {
         "from": { "type": "string", "description": "Source agent name" },
         "score": { "type": "number", "minimum": 0, "maximum": 100, "description": "Source agent final score" },
@@ -83,7 +84,7 @@
         "injection_budget": {
           "type": "integer",
           "minimum": 0,
-          "description": "Token budget for context injection (FRESH: 2500, MODERATE: 2000, DEPLETED: 1500, CRITICAL: 800)"
+          "description": "Token budget for context injection — Progressive Reinforcement: FRESH 1.5%, MODERATE 2.5%, DEPLETED 4.0%, CRITICAL 5.0% of provider context window"
         }
       }
     },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chati-dev",
-  "version": "4.1.2",
+  "version": "4.1.4",
   "description": "AI-Powered Multi-Agent Orchestration System — Structured vibe coding for Full Stack Development",
   "type": "module",
   "bin": {

package/src/context/bracket-tracker.js

@@ -1,21 +1,20 @@
 /**
  * Bracket Tracker — Pure arithmetic for context window management.
  *
- * Brackets determine how much context to inject:
- *   FRESH    (60-100%) → All 6 layers active (L0-L5)
- *   MODERATE (40-60%)  → L0 + L1 + L2 + L3 + L5 (skip L4 task detail)
- *   DEPLETED (25-40%)  → L0 + L1 + L2 only
- *   CRITICAL (<25%)    → L0 + L1 only (handoff mandatory)
+ * Brackets determine how much context to inject (Progressive Reinforcement):
+ *   FRESH    (60-100%) → L0 + L1 only (model has full memory, minimal reinforcement)
+ *   MODERATE (40-60%)  → L0 + L1 + L2 + L3 (model starting to forget)
+ *   DEPLETED (25-40%)  → All 6 layers L0-L5 (significant context loss, full reinforcement)
+ *   CRITICAL (<25%)    → All 6 layers L0-L5 + handoff mandatory (maximum reinforcement)
  *
- * Progressive Reinforcement Model (v4.0):
- *   As context window depletes, the model forgets initial instructions
- *   and needs MORE reinforcement, not less. Budget expressed as percentage
- *   of the provider's total context window — proportional across providers.
+ * Progressive Reinforcement Model (v4.1.2):
+ *   As context depletes, BOTH budget AND layers INCREASE.
+ *   The model forgets initial instructions and needs MORE reinforcement.
  *
- *   FRESH    = 1.5% → minimal reinforcement (context is plentiful)
- *   MODERATE = 2.5% → growing reinforcement (model starting to forget)
- *   DEPLETED = 4.0% → heavy reinforcement (significant context loss)
- *   CRITICAL = 5.0% → maximum reinforcement (last interactions before handoff)
+ *   FRESH    = 1.5% budget, 2 layers → context intact, no need to re-inject
+ *   MODERATE = 2.5% budget, 4 layers → growing reinforcement
+ *   DEPLETED = 4.0% budget, 6 layers → full layer injection to compensate loss
+ *   CRITICAL = 5.0% budget, 6 layers → maximum reinforcement + forced handoff
  */
 
 import { resolveContextLimit } from '../utils/provider-limits.js';

package/src/context/formatter.js

@@ -54,8 +54,9 @@ export function formatContext(options) {
 
   // Insert STATIC_BOUNDARY between static (L0+L1) and dynamic (L2-L5) sections.
   // This enables prompt cache reuse for the stable prefix (Claude Code pattern).
-  const staticSections = sections.slice(0, l1 ? 2 : (l0 ? 1 : 0));
-  const dynamicSections = sections.slice(staticSections.length);
+  const staticCount = l1 ? 2 : (l0 ? 1 : 0);
+  const staticSections = sections.slice(0, staticCount);
+  const dynamicSections = sections.slice(staticCount);
 
   let body;
   if (dynamicSections.length > 0) {
@@ -69,7 +70,7 @@ export function formatContext(options) {
   // Truncate if over budget (rough estimate: 1 token ≈ 4 chars)
   const maxChars = tokenBudget * 4;
   if (body.length > maxChars) {
-    body = truncateByPriority(sections, maxChars);
+    body = truncateByPriority(sections, maxChars, staticCount);
   }
 
   return `<chati-context bracket="${bracket}">\n${body}\n</chati-context>`;
@@ -190,14 +191,28 @@ function formatKeywords(l5) {
  * Truncate sections by removing lower-priority layers first (L4 → L3 → L2).
  * L0 and L1 are never truncated.
  */
-function truncateByPriority(sections, maxChars) {
-  let result = sections.join('\n\n');
-  // Remove from end (lowest priority) until within budget
+function truncateByPriority(sections, maxChars, staticCount = 2) {
+  const boundaryMarker = '\n\n  <!-- STATIC_BOUNDARY -->\n\n';
   const trimmed = [...sections];
-  while (trimmed.length > 2 && result.length > maxChars) {
+  const hasDynamic = () => trimmed.length > staticCount;
+  const overhead = hasDynamic() ? boundaryMarker.length : 0;
+  const adjustedMax = maxChars - overhead;
+
+  // Remove from end (lowest priority) until within budget
+  while (trimmed.length > staticCount && trimmed.join('\n\n').length > adjustedMax) {
     trimmed.pop();
+  }
+
+  // Re-insert boundary if we still have both static and dynamic sections
+  let result;
+  if (hasDynamic()) {
+    result = trimmed.slice(0, staticCount).join('\n\n') +
+      boundaryMarker +
+      trimmed.slice(staticCount).join('\n\n');
+  } else {
     result = trimmed.join('\n\n');
   }
+
   // If still over, hard truncate
   if (result.length > maxChars) {
     result = result.slice(0, maxChars) + '\n  <!-- truncated -->';

package/src/context/layers/l5-keywords.js

@@ -6,7 +6,7 @@
  *   L0-L4 handle WHO (agent) and WHERE (workflow/task)
  *   L5 handles WHAT (topic detection from user prompt)
  *
- * Only active in FRESH and MODERATE brackets (expensive layer).
+ * Only active in DEPLETED and CRITICAL brackets (reinforcement layer).
  */
 
 import { loadKeywordDomains } from '../domain-loader.js';

package/src/installer/core.js

@@ -161,7 +161,12 @@ function copyFrameworkFiles(destDir, provider = 'claude') {
     // PLAN agents
     'agents/plan/detail.md',
     'agents/plan/architect.md',
+    'agents/plan/architect-system.md',
+    'agents/plan/architect-data-engineer.md',
     'agents/plan/ux.md',
+    'agents/plan/ux-brand-architect.md',
+    'agents/plan/ux-researcher.md',
+    'agents/plan/ux-component-engineer.md',
     'agents/plan/phases.md',
     'agents/plan/tasks.md',
     // Quality agents
@@ -178,9 +183,11 @@ function copyFrameworkFiles(destDir, provider = 'claude') {
     'templates/qa-gate-tmpl.yaml',
     'templates/quick-brief-tmpl.yaml',
     'templates/brandbook-tmpl.yaml',
+    'templates/brandbook-html-tmpl.md',
     'templates/design-token-tmpl.yaml',
     'templates/component-spec-tmpl.yaml',
     'templates/icon-system-tmpl.yaml',
+    'templates/session-memory-tmpl.yaml',
     // Workflows
     'workflows/greenfield-fullstack.yaml',
     'workflows/brownfield-fullstack.yaml',

package/src/memory/dream.js

@@ -7,7 +7,7 @@
  *   Phase 3 (Consolidate): Deduplicate, merge cross-agent, increment evidence
  *   Phase 4 (Prune):       Remove cold/low-confidence entries, archive removed
  *
- * Triggered manually via `/chati dream` or programmatically.
+ * Triggered automatically when memory entries exceed threshold or on PreCompact.
  */
 
 import { existsSync, readFileSync, writeFileSync, mkdirSync, readdirSync } from 'fs';
@@ -45,6 +45,14 @@ export function runDreamConsolidation(projectDir) {
   // Phase 4: Prune
   report.phase4 = prune(projectDir, report.phase3.consolidated);
 
+  // Write consolidated memories to disk
+  try {
+    const sharedDir = join(projectDir, '.chati', 'memories', 'shared');
+    mkdirSync(sharedDir, { recursive: true });
+    const consolidatedPath = join(sharedDir, 'consolidated.json');
+    writeFileSync(consolidatedPath, JSON.stringify(report.phase4.entries, null, 2), 'utf-8');
+  } catch (err) { process.stderr.write(`[chati] dream write consolidated: ${err.message}\n`); }
+
   // Summary
   report.summary = {
     merged: report.phase3.mergedCount,
@@ -198,6 +206,7 @@ function prune(projectDir, consolidated) {
       prunedCount: 0,
       archivedCount: 0,
       remaining: consolidated.length,
+      entries: consolidated,
     };
   }
 
@@ -238,6 +247,7 @@ function prune(projectDir, consolidated) {
     prunedCount: toRemove.length,
     archivedCount,
     remaining: toKeep.length,
+    entries: toKeep,
   };
 }
 

package/src/memory/memory-extractor.js

@@ -115,7 +115,7 @@ export function extractMemories(text, context = {}) {
   // Build memory entries following memory.schema.json
   const now = new Date().toISOString();
   return selected.map((candidate, i) => ({
-    id: `mem-${now.split('T')[0]}-${String(i + 1).padStart(3, '0')}`,
+    id: `mem-${now.replace(/[:.]/g, '-')}-${String(i + 1).padStart(3, '0')}`,
     type: candidate.type,
     agent: context.agent || 'unknown',
     content: candidate.content,

package/src/orchestrator/cli.js

@@ -178,7 +178,8 @@ function resolveAgentModel(agent, projectDir) {
     if (existsSync(configPath)) {
       const configRaw = readFileSync(configPath, 'utf-8');
       // Simple YAML parsing for agent_overrides section
-      const overrideMatch = configRaw.match(new RegExp(`${agent}:\\s*\\n\\s+provider:\\s*(\\w+)`, 'm'));
+      const escapedAgent = agent.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+      const overrideMatch = configRaw.match(new RegExp(`${escapedAgent}:\\s*\\n\\s+provider:\\s*(\\w+)`, 'm'));
       if (overrideMatch) {
         return { ...defaults, provider: overrideMatch[1] };
       }

package/src/scanning/leakage-scanner.js

@@ -13,7 +13,9 @@ const TECH_NAMES = [
   /\bExpress\b/g, /\bFastify\b/g, /\bDjango\b/g, /\bFlask\b/g, /\bLaravel\b/g,
   /\bAWS\b/g, /\bVercel\b/g, /\bNetlify\b/g, /\bRailway\b/g, /\bCloudflare\b/g,
   /\bDocker\b/g, /\bKubernetes\b/g, /\bRedis\b/g, /\bGraphQL\b/gi,
-  /\bTypeScript\b/gi, /\bJavaScript\b/gi, /\bPython\b/g, /\bRust\b/g, /\bGo\b/g,
+  /\bTypeScript\b/gi, /\bJavaScript\b/gi, /\bPython\b/g,
+  /\bRustlang\b|\bRust\s+(?:lang|crate|cargo|tokio|async|compiler)\b/gi,
+  /\bGolang\b|\bGo\s+(?:module|package|routine|channel|func|goroutine)\b/g,
   /\bZod\b/g, /\bJoi\b/g, /\bZustand\b/g, /\bRedux\b/g, /\bJotai\b/g,
 ];
 

package/src/security/bash-security.js

@@ -62,9 +62,9 @@ export const SECURITY_CHECKS = [
   {
     id: 'SHELL_METACHARACTERS',
     name: 'Shell metacharacter injection',
-    pattern: /[`]|(?:\$\((?!.*\becho\b))/,
+    pattern: /[`]|\$\(/,
     severity: 'critical',
-    description: 'Backticks or $() command substitution (excluding safe echo patterns)',
+    description: 'Backticks or $() command substitution',
   },
   {
     id: 'DANGEROUS_VARIABLES',

package/src/utils/brackets.js

@@ -4,8 +4,10 @@
  * Single source of truth for bracket thresholds, layer configurations,
  * and budget ratios. Consumers import from here to avoid duplication.
  *
- * Progressive Reinforcement Model (v4.0):
- *   As context depletes, budget INCREASES (model needs MORE reinforcement).
+ * Progressive Reinforcement Model (v4.1.2):
+ *   As context depletes, both budget AND layers INCREASE.
+ *   FRESH = model has full memory, minimal reinforcement needed.
+ *   CRITICAL = model forgot initial instructions, maximum reinforcement + handoff.
  *   Budget expressed as ratio of provider's total context window.
  */
 
@@ -14,10 +16,10 @@
  * @type {Record<string, {min: number, max: number, layers: string[], budgetRatio: number}>}
  */
 export const BRACKETS = {
-  FRESH:    { min: 60, max: 100, layers: ['L0', 'L1', 'L2', 'L3', 'L4', 'L5'], budgetRatio: 0.015 },
-  MODERATE: { min: 40, max: 60,  layers: ['L0', 'L1', 'L2', 'L3', 'L5'],       budgetRatio: 0.025 },
-  DEPLETED: { min: 25, max: 40,  layers: ['L0', 'L1', 'L2'],                   budgetRatio: 0.040 },
-  CRITICAL: { min: 0,  max: 25,  layers: ['L0', 'L1'],                         budgetRatio: 0.050 },
+  FRESH:    { min: 60, max: 100, layers: ['L0', 'L1'],                         budgetRatio: 0.015 },
+  MODERATE: { min: 40, max: 60,  layers: ['L0', 'L1', 'L2', 'L3'],             budgetRatio: 0.025 },
+  DEPLETED: { min: 25, max: 40,  layers: ['L0', 'L1', 'L2', 'L3', 'L4', 'L5'], budgetRatio: 0.040 },
+  CRITICAL: { min: 0,  max: 25,  layers: ['L0', 'L1', 'L2', 'L3', 'L4', 'L5'], budgetRatio: 0.050 },
 };
 
 /**