loki-mode 5.2.4 → 5.4.0

package/README.md

@@ -25,6 +25,16 @@
 
 ---
 
+## Presentation
+
+![Loki Mode Presentation](docs/loki-mode-presentation.gif)
+
+*9 slides: Problem, Solution, 37 Agents, RARV Cycle, Benchmarks, Multi-Provider, Full Lifecycle*
+
+**[Download PPTX](docs/loki-mode-presentation.pptx)** for offline viewing
+
+---
+
 ## Usage
 
 ### Option 1: Claude Code Skill (Recommended)

package/SKILL.md

@@ -3,7 +3,7 @@ name: loki-mode
 description: Multi-agent autonomous startup system. Triggers on "Loki Mode". Takes PRD to deployed product with zero human intervention. Requires --dangerously-skip-permissions flag.
 ---
 
-# Loki Mode v5.2.4
+# Loki Mode v5.4.0
 
 **You are an autonomous agent. You make decisions. You do not ask questions. You do not stop.**
 
@@ -88,15 +88,17 @@ These rules are ABSOLUTE. Violating them is a critical failure.
 
 ## Model Selection
 
-| Task Type | Tier | Claude | Codex | Gemini |
-|-----------|------|--------|-------|--------|
-| PRD analysis, architecture, system design | **planning** | opus | effort=xhigh | thinking=high |
-| Feature implementation, complex bugs | **development** | sonnet | effort=high | thinking=medium |
-| Code review (always 3 parallel reviewers) | **development** | sonnet | effort=high | thinking=medium |
-| Integration tests, E2E, deployment | **development** | sonnet | effort=high | thinking=medium |
-| Unit tests, linting, docs, simple fixes | **fast** | haiku | effort=low | thinking=low |
+**Default (v5.3.0):** Haiku disabled for quality. Use `--allow-haiku` or `LOKI_ALLOW_HAIKU=true` to enable.
 
-**Parallelization rule (Claude only):** Launch up to 10 haiku agents simultaneously for independent tasks.
+| Task Type | Tier | Claude (default) | Claude (--allow-haiku) | Codex | Gemini |
+|-----------|------|------------------|------------------------|-------|--------|
+| PRD analysis, architecture, system design | **planning** | opus | opus | effort=xhigh | thinking=high |
+| Feature implementation, complex bugs | **development** | opus | sonnet | effort=high | thinking=medium |
+| Code review (always 3 parallel reviewers) | **development** | opus | sonnet | effort=high | thinking=medium |
+| Integration tests, E2E, deployment | **development** | opus | sonnet | effort=high | thinking=medium |
+| Unit tests, linting, docs, simple fixes | **fast** | sonnet | haiku | effort=low | thinking=low |
+
+**Parallelization rule (Claude only):** Launch up to 10 agents simultaneously for independent tasks.
 
 **Degraded mode (Codex/Gemini):** No parallel agents or Task tool. Runs RARV cycle sequentially. See `skills/model-selection.md`.
 
@@ -180,11 +182,11 @@ GROWTH ──[continuous improvement loop]──> GROWTH
 ```bash
 # Standard mode (Claude - full features)
 claude --dangerously-skip-permissions
-# Then say: "Loki Mode" or "Loki Mode with PRD at path/to/prd.md"
+# Then say: "Loki Mode" or "Loki Mode with PRD at path/to/prd.md" (or .json)
 
-# With provider selection
+# With provider selection (supports .md and .json PRDs)
 ./autonomy/run.sh --provider claude ./prd.md   # Default, full features
-./autonomy/run.sh --provider codex ./prd.md    # GPT-5.2 Codex, degraded mode
+./autonomy/run.sh --provider codex ./prd.json  # GPT-5.2 Codex, degraded mode
 ./autonomy/run.sh --provider gemini ./prd.md   # Gemini 3 Pro, degraded mode
 
 # Or via CLI wrapper
@@ -208,11 +210,23 @@ When running with `autonomy/run.sh`, you can intervene:
 | Method | Effect |
 |--------|--------|
 | `touch .loki/PAUSE` | Pauses after current session |
-| `echo "instructions" > .loki/HUMAN_INPUT.md` | Injects instructions into next prompt |
+| `echo "instructions" > .loki/HUMAN_INPUT.md` | Injects directive into next prompt (executed immediately) |
 | `touch .loki/STOP` | Stops immediately |
 | Ctrl+C (once) | Pauses, shows options |
 | Ctrl+C (twice) | Exits immediately |
 
+### Hints vs Directives
+
+| Type | File | Behavior |
+|------|------|----------|
+| **Hint** | `.loki/CONTINUITY.md` "Mistakes & Learnings" | Passive memory - remembered but not acted upon |
+| **Directive** | `.loki/HUMAN_INPUT.md` | Active instruction - executed BEFORE normal tasks |
+
+**Example directive** (check all .astro files):
+```bash
+echo "Check all .astro files for missing BaseLayout imports. Fix any issues found. Add a compilation test to prevent this regression." > .loki/HUMAN_INPUT.md
+```
+
 ---
 
 ## Complexity Tiers (v3.4.0)
@@ -227,4 +241,4 @@ Auto-detected or force with `LOKI_COMPLEXITY`:
 
 ---
 
-**v5.2.4 | CoVe + MemEvolve + Quality Gates | ~230 lines core**
+**v5.4.0 | JSON PRD Support + HUMAN_INPUT.md Fix | ~245 lines core**

package/VERSION

@@ -1 +1 @@
-5.2.4
+5.4.0

package/autonomy/run.sh

@@ -53,6 +53,12 @@
 #   LOKI_PERPETUAL_MODE        - Ignore ALL completion signals (default: false)
 #                                Set to 'true' for truly infinite operation
 #
+# Model Selection:
+#   LOKI_ALLOW_HAIKU           - Enable Haiku model for fast tier (default: false)
+#                                When false: Opus for dev/bugfix, Sonnet for tests/docs
+#                                When true:  Sonnet for dev, Haiku for tests/docs (original)
+#                                Use --allow-haiku flag or set to 'true'
+#
 # 2026 Research Enhancements:
 #   LOKI_PROMPT_REPETITION     - Enable prompt repetition for Haiku agents (default: true)
 #                                arXiv 2512.14982v1: Improves accuracy 4-5x on structured tasks
@@ -577,7 +583,25 @@ detect_complexity() {
     local prd_complexity="standard"
     if [ -n "$prd_path" ] && [ -f "$prd_path" ]; then
         local prd_words=$(wc -w < "$prd_path" | tr -d ' ')
-        local feature_count=$(grep -c "^##\|^- \[" "$prd_path" 2>/dev/null || echo "0")
+        local feature_count=0
+
+        # Detect PRD format and count features accordingly
+        if [[ "$prd_path" == *.json ]]; then
+            # JSON PRD: count features, requirements, tasks arrays
+            if command -v jq &>/dev/null; then
+                feature_count=$(jq '
+                    [.features, .requirements, .tasks, .user_stories, .epics] |
+                    map(select(. != null) | if type == "array" then length else 0 end) |
+                    add // 0
+                ' "$prd_path" 2>/dev/null || echo "0")
+            else
+                # Fallback: count array elements by pattern
+                feature_count=$(grep -c '"title"\|"name"\|"feature"\|"requirement"' "$prd_path" 2>/dev/null || echo "0")
+            fi
+        else
+            # Markdown PRD: count headers and checkboxes
+            feature_count=$(grep -c "^##\|^- \[" "$prd_path" 2>/dev/null || echo "0")
+        fi
 
         if [ "$prd_words" -lt 200 ] && [ "$feature_count" -lt 5 ]; then
             prd_complexity="simple"
@@ -3255,17 +3279,23 @@ build_prompt() {
         fi
     fi
 
+    # Human directive injection (from HUMAN_INPUT.md)
+    local human_directive=""
+    if [ -n "${LOKI_HUMAN_INPUT:-}" ]; then
+        human_directive="HUMAN_DIRECTIVE (PRIORITY): $LOKI_HUMAN_INPUT Execute this directive BEFORE continuing normal tasks."
+    fi
+
     if [ $retry -eq 0 ]; then
         if [ -n "$prd" ]; then
-            echo "Loki Mode with PRD at $prd. $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
+            echo "Loki Mode with PRD at $prd. $human_directive $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
         else
-            echo "Loki Mode. $analysis_instruction $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
+            echo "Loki Mode. $human_directive $analysis_instruction $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
         fi
     else
         if [ -n "$prd" ]; then
-            echo "Loki Mode - Resume iteration #$iteration (retry #$retry). PRD: $prd. $context_injection $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
+            echo "Loki Mode - Resume iteration #$iteration (retry #$retry). PRD: $prd. $human_directive $context_injection $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
         else
-            echo "Loki Mode - Resume iteration #$iteration (retry #$retry). $context_injection Use .loki/generated-prd.md if exists. $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
+            echo "Loki Mode - Resume iteration #$iteration (retry #$retry). $human_directive $context_injection Use .loki/generated-prd.md if exists. $rarv_instruction $memory_instruction $compaction_reminder $completion_instruction $sdlc_instruction $autonomous_suffix"
         fi
     fi
 }
@@ -3284,10 +3314,14 @@ run_autonomous() {
         log_step "No PRD provided, searching for existing PRD files..."
         local found_prd=""
 
-        # Search common PRD file patterns
-        for pattern in "PRD.md" "prd.md" "REQUIREMENTS.md" "requirements.md" "SPEC.md" "spec.md" \
-                       "docs/PRD.md" "docs/prd.md" "docs/REQUIREMENTS.md" "docs/requirements.md" \
-                       "docs/SPEC.md" "docs/spec.md" ".github/PRD.md" "PROJECT.md" "project.md"; do
+        # Search common PRD file patterns (markdown and JSON)
+        for pattern in "PRD.md" "prd.md" "PRD.json" "prd.json" \
+                       "REQUIREMENTS.md" "requirements.md" "requirements.json" \
+                       "SPEC.md" "spec.md" "spec.json" \
+                       "docs/PRD.md" "docs/prd.md" "docs/PRD.json" "docs/prd.json" \
+                       "docs/REQUIREMENTS.md" "docs/requirements.md" "docs/requirements.json" \
+                       "docs/SPEC.md" "docs/spec.md" "docs/spec.json" \
+                       ".github/PRD.md" ".github/PRD.json" "PROJECT.md" "project.md" "project.json"; do
             if [ -f "$pattern" ]; then
                 found_prd="$pattern"
                 break
@@ -3300,6 +3334,9 @@ run_autonomous() {
         elif [ -f ".loki/generated-prd.md" ]; then
             log_info "Using previously generated PRD: .loki/generated-prd.md"
             prd_path=".loki/generated-prd.md"
+        elif [ -f ".loki/generated-prd.json" ]; then
+            log_info "Using previously generated PRD: .loki/generated-prd.json"
+            prd_path=".loki/generated-prd.json"
         else
             log_info "No PRD found - will analyze codebase and generate one"
         fi
@@ -3335,6 +3372,15 @@ run_autonomous() {
             return 0
         fi
 
+        # Check for human intervention (PAUSE, HUMAN_INPUT.md, STOP)
+        local intervention_result
+        intervention_result=$(check_human_intervention; echo $?)
+        intervention_result=${intervention_result##*$'\n'}  # Get exit code
+        case $intervention_result in
+            1) continue ;;  # PAUSE handled, restart loop
+            2) return 0 ;;  # STOP requested
+        esac
+
         local prompt=$(build_prompt $retry "$prd_path" $ITERATION_COUNT)
 
         echo ""
@@ -3867,6 +3913,11 @@ main() {
                 PARALLEL_MODE=true
                 shift
                 ;;
+            --allow-haiku)
+                export LOKI_ALLOW_HAIKU=true
+                log_info "Haiku model enabled for fast tier"
+                shift
+                ;;
             --provider)
                 if [[ -n "${2:-}" ]]; then
                     LOKI_PROVIDER="$2"
@@ -3909,6 +3960,7 @@ main() {
                 echo ""
                 echo "Options:"
                 echo "  --parallel           Enable git worktree-based parallel workflows"
+                echo "  --allow-haiku        Enable Haiku model for fast tier (default: disabled)"
                 echo "  --provider <name>    Provider: claude (default), codex, gemini"
                 echo "  --help, -h           Show this help message"
                 echo ""

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loki-mode",
-  "version": "5.2.4",
+  "version": "5.4.0",
   "description": "Multi-agent autonomous startup system for Claude Code, Codex CLI, and Gemini CLI",
   "keywords": [
     "claude",

package/providers/claude.sh

@@ -44,13 +44,23 @@ PROVIDER_HAS_MCP=true
 PROVIDER_MAX_PARALLEL=10
 
 # Model Configuration (Abstract Tiers)
+# Default: Haiku disabled for quality. Use --allow-haiku or LOKI_ALLOW_HAIKU=true to enable.
 PROVIDER_MODEL_PLANNING="claude-opus-4-5-20251101"
-PROVIDER_MODEL_DEVELOPMENT="claude-sonnet-4-5-20251101"
-PROVIDER_MODEL_FAST="claude-haiku-4-5-20251101"
+PROVIDER_MODEL_DEVELOPMENT="claude-opus-4-5-20251101"  # Opus for dev (was sonnet)
+if [ "${LOKI_ALLOW_HAIKU:-false}" = "true" ]; then
+    PROVIDER_MODEL_FAST="claude-haiku-4-5-20251101"
+    PROVIDER_MODEL_DEVELOPMENT="claude-sonnet-4-5-20251101"  # Sonnet for dev when haiku enabled
+else
+    PROVIDER_MODEL_FAST="claude-sonnet-4-5-20251101"  # Sonnet for fast (no haiku)
+fi
 
 # Model Selection (for Task tool)
 PROVIDER_TASK_MODEL_PARAM="model"
-PROVIDER_TASK_MODEL_VALUES=("opus" "sonnet" "haiku")
+if [ "${LOKI_ALLOW_HAIKU:-false}" = "true" ]; then
+    PROVIDER_TASK_MODEL_VALUES=("opus" "sonnet" "haiku")
+else
+    PROVIDER_TASK_MODEL_VALUES=("opus" "sonnet")  # No haiku option
+fi
 
 # Context and Limits
 PROVIDER_CONTEXT_WINDOW=200000
@@ -87,14 +97,28 @@ provider_invoke() {
 }
 
 # Model tier to Task tool model parameter value
+# Respects LOKI_ALLOW_HAIKU flag for tier mapping
 provider_get_tier_param() {
     local tier="$1"
-    case "$tier" in
-        planning) echo "opus" ;;
-        development) echo "sonnet" ;;
-        fast) echo "haiku" ;;
-        *) echo "sonnet" ;;  # default to development tier
-    esac
+    if [ "${LOKI_ALLOW_HAIKU:-false}" = "true" ]; then
+        # With haiku: original tier mapping
+        case "$tier" in
+            planning) echo "opus" ;;
+            development) echo "sonnet" ;;
+            fast) echo "haiku" ;;
+            *) echo "sonnet" ;;
+        esac
+    else
+        # Without haiku (default): upgrade all tiers
+        # - Development + bug fixes -> opus
+        # - Testing + documentation -> sonnet
+        case "$tier" in
+            planning) echo "opus" ;;
+            development) echo "opus" ;;  # Upgraded from sonnet
+            fast) echo "sonnet" ;;       # Upgraded from haiku
+            *) echo "opus" ;;            # Default to opus
+        esac
+    fi
 }
 
 # Tier-aware invocation (Claude supports model selection via --model flag)

package/skills/model-selection.md

@@ -20,11 +20,33 @@ Loki Mode supports three AI providers. Claude has full features; Codex and Gemin
 
 ## Abstract Model Tiers
 
-| Tier | Purpose | Claude | Codex | Gemini |
-|------|---------|--------|-------|--------|
-| **planning** | PRD analysis, architecture, system design | opus | effort=xhigh | thinking=high |
-| **development** | Feature implementation, complex bugs, tests | sonnet | effort=high | thinking=medium |
-| **fast** | Unit tests, docs, linting, simple tasks | haiku | effort=low | thinking=low |
+**Default (v5.3.0):** Haiku disabled for quality. All tasks use Opus or Sonnet.
+
+| Tier | Purpose | Claude (default) | Claude (--allow-haiku) | Codex | Gemini |
+|------|---------|------------------|------------------------|-------|--------|
+| **planning** | PRD analysis, architecture, system design | opus | opus | effort=xhigh | thinking=high |
+| **development** | Feature implementation, complex bugs, tests | opus | sonnet | effort=high | thinking=medium |
+| **fast** | Unit tests, docs, linting, simple tasks | sonnet | haiku | effort=low | thinking=low |
+
+### Enabling Haiku
+
+To enable Haiku for fast-tier tasks (cost optimization at potential quality trade-off):
+
+```bash
+# Via environment variable
+LOKI_ALLOW_HAIKU=true ./autonomy/run.sh ./prd.md
+
+# Via CLI flag
+./autonomy/run.sh --allow-haiku ./prd.md
+
+# Via loki CLI wrapper
+loki start --allow-haiku ./prd.md
+```
+
+When Haiku is enabled:
+- Development tier uses Sonnet (instead of Opus)
+- Fast tier uses Haiku (instead of Sonnet)
+- Planning tier always uses Opus (unchanged)
 
 ---
 
@@ -36,7 +58,7 @@ Loki Mode supports three AI providers. Claude has full features; Codex and Gemin
 | **development** | Development, QA, Deployment | Feature implementation, complex bugs, integration/E2E tests, code review, deployment |
 | **fast** | All other operations (parallel for Claude) | Unit tests, docs, bash commands, linting, monitoring |
 
-**Claude-specific model names:** opus, sonnet, haiku
+**Claude-specific model names:** opus, sonnet, haiku (haiku requires --allow-haiku flag)
 **Codex effort levels:** xhigh, high, medium, low
 **Gemini thinking levels:** high, medium, low