loki-mode 10.5.16 → 10.5.19

package/docs/auto-claude-comparison.md

@@ -84,21 +84,6 @@ Loki Mode is built on peer-reviewed research:
 
 **Verdict: Loki Mode wins** - Academically grounded.
 
-### 2. Specialized Agent Types
-Loki Mode has 41 specialized agent roles across 8 domains - prompt-defined specifications the orchestrator adopts per phase, with parallel review (blind council) and optional worktree streams on Claude Code, sequential on other providers:
-- Engineering (8): frontend, backend, database, mobile, API, QA, perf, infra
-- Operations (8): DevOps, SRE, security, monitoring, incident, release, cost, compliance
-- Business (8): marketing, sales, finance, legal, support, HR, investor, partnerships
-- Data (3): ML, engineering, analytics
-- Product (3): PM, design, tech writer
-- Growth (4): hacker, community, success, lifecycle
-- Review (3): code, business, security
-- Orchestration (4): planner, sub-planner, judge, coordinator
-
-**Auto-Claude:** 4 agent types: planner, coder, memory_manager, QA
-
-**Verdict: Loki Mode wins** - 10x more specialized coverage.
-
 ### 3. Full SDLC Coverage
 Loki Mode covers:
 - Engineering (code, tests, deployment)
@@ -143,25 +128,10 @@ Beyond the gates, the SDLC pipeline covers these phases:
 
 **Verdict: Loki Mode wins** - Comprehensive quality vs single loop.
 
-### 6. Published Benchmarks
-Loki Mode:
-- HumanEval: 98.78% Pass@1 (162/164, reproducible harness)
-- SWE-bench: not measured. A patch-generation harness exists (299/300 diffs
-  produced), but the official evaluator was never run, so there is no resolve
-  rate. Patch generation is not a success metric and is not comparable to
-  other tools' resolution scores.
-
-**Auto-Claude:** No published benchmarks.
-
-**Verdict: Loki Mode wins on HumanEval transparency** - one real, reproducible
-number versus none. No SWE-bench performance claim is made.
-
 ### 7. Licensing
-- Loki Mode: MIT (free, no restrictions)
+- Loki Mode: BUSL-1.1 (source-available)
 - Auto-Claude: AGPL-3.0 (copyleft, requires open-sourcing modifications)
 
-**Verdict: Loki Mode wins** - More permissive for commercial use.
-
 ### 8. API Access
 - Loki Mode: Works with Claude API (any tier)
 - Auto-Claude: Requires Claude Pro/Max subscription
@@ -235,10 +205,8 @@ Loki Mode now incorporates proven patterns from Cursor's large-scale agent deplo
 ## What Auto-Claude Could Learn from Loki Mode
 
 1. **Research Foundation** - Document the science behind decisions
-2. **Specialized Agents** - More than 4 generic agent types
 3. **Anti-Sycophancy** - Blind review prevents false positives
 4. **Full SDLC** - Business, marketing, growth automation
-5. **Published Benchmarks** - Verify claims with reproducible tests
 6. **Source-available (BUSL-1.1)** - Inspect and self-host the full code
 
 ---

package/docs/certification/01-core-concepts/lab.md

@@ -8,7 +8,7 @@ Install Loki Mode, verify the installation, create a simple PRD, run it, and exa
 
 - Node.js 18+ installed
 - npm available on your PATH
-- A supported AI provider CLI installed (Claude Code, Codex CLI, or Gemini CLI)
+- A supported AI provider CLI installed (Claude Code, Cline, Codex CLI, Aider, or opencode)
 - An active API key for your chosen provider
 
 ## Step 1: Install Loki Mode
@@ -23,7 +23,7 @@ Verify the installation:
 loki version
 ```
 
-You should see output like `5.51.0` (or the current version).
+You should see the installed version number.
 
 ## Step 2: Run the Doctor Check
 
@@ -87,7 +87,7 @@ Alternatively, if using a specific provider:
 ```bash
 loki start --provider claude ./simple-prd.md    # Claude Code (default)
 loki start --provider codex ./simple-prd.md     # OpenAI Codex CLI
-loki start --provider gemini ./simple-prd.md    # Google Gemini CLI
+loki start --provider cline ./simple-prd.md     # Cline
 ```
 
 You can also force the simple complexity tier to limit the number of phases:

package/docs/certification/01-core-concepts/lesson.md

@@ -2,7 +2,7 @@
 
 ## What is Loki Mode?
 
-Loki Mode is a multi-agent autonomous system created by [Autonomi](https://www.autonomi.dev/). It takes a Product Requirements Document (PRD) and builds a fully deployed product with minimal human intervention. It supports three AI provider CLIs: Claude Code (full features), OpenAI Codex CLI (degraded mode), and Google Gemini CLI (degraded mode).
+Loki Mode is a multi-agent autonomous system created by [Autonomi](https://www.autonomi.dev/). It takes a Product Requirements Document (PRD) and builds a fully deployed product with minimal human intervention. It supports five provider CLIs: Claude Code (full features), Cline (reduced), OpenAI Codex CLI and Aider (degraded, sequential), and opencode (model-agnostic).
 
 Loki Mode is installed as an npm package and invoked through the `loki` CLI or directly within Claude Code as a skill.
 
@@ -70,18 +70,7 @@ Or via the environment variable `LOKI_COMPLEXITY=simple|standard|complex`.
 
 ## Agents
 
-Loki Mode defines **41 specialized agent types** organized into **8 domains**:
-
-| Domain | Agent Count | Examples |
-|-------|-------------|----------|
-| Engineering | 8 | frontend, backend, database, mobile, api, qa, perf, infra |
-| Operations | 8 | devops, sre, security, monitor, incident, release, cost, compliance |
-| Business | 8 | marketing, sales, finance, legal, support, hr, investor, partnerships |
-| Data | 3 | ml, eng, analytics |
-| Product | 3 | pm, design, techwriter |
-| Growth | 4 | hacker, community, success, lifecycle |
-| Review | 3 | code, business, security |
-| Orchestration | 4 | (internal system agents) |
+Loki Mode defines specialized agent roles across engineering, operations, product, data, growth and review domains.
 
 These agents are **roles defined through prompts**, not separate programs. They are implemented using the Claude Code Task tool with role-specific prompts:
 
@@ -99,7 +88,7 @@ Task(
 )
 ```
 
-A simple project typically uses 5-10 agents. Complex projects use more as needed. The orchestrator spawns only the agents required for the current task.
+The orchestrator spawns only the agents required for the current task.
 
 Full agent type definitions are in `references/agent-types.md`.
 
@@ -154,14 +143,14 @@ The memory engine is implemented in Python (`memory/engine.py`, `memory/retrieva
 
 Loki Mode maps tasks to model tiers rather than specific model names:
 
-| Task Type | Tier | Claude | Codex | Gemini |
-|-----------|------|--------|-------|--------|
-| Architecture, system design | planning | opus | effort=xhigh | thinking=high |
-| Feature implementation, bugs | development | opus | effort=high | thinking=medium |
-| Code review | development | opus (sonnet for reviewers) | effort=high | thinking=medium |
-| Unit tests, linting, docs | fast | sonnet | effort=low | thinking=low |
+| Task Type | Tier | Claude | Codex |
+|-----------|------|--------|-------|
+| Architecture, system design | planning | opus | effort=xhigh |
+| Feature implementation, bugs | development | opus | effort=high |
+| Code review | development | opus (sonnet for reviewers) | effort=high |
+| Unit tests, linting, docs | fast | sonnet | effort=low |
 
-Claude Code has full feature support: parallel agents (up to 10 simultaneous), the Task tool, and MCP integration. Codex and Gemini run in degraded mode: sequential execution only, no Task tool, no parallel agents.
+Claude Code has full feature support: parallel agents (up to 10 simultaneous), the Task tool, and MCP integration. Cline runs in reduced mode; Codex and Aider run in degraded mode (sequential execution only, no Task tool, no parallel agents); opencode is model-agnostic.
 
 ## Key Files
 
@@ -180,4 +169,4 @@ Every Loki Mode project uses these files in the `.loki/` directory:
 
 ## Summary
 
-Loki Mode is an autonomous multi-agent system that follows the RARV cycle to build software from PRDs. It uses 41 agent types organized into 8 domains, enforces quality through 8 gates with blind peer review, and maintains episodic/semantic/procedural memory for continuous learning. Projects are classified into simple, standard, or complex tiers that determine the number of phases executed.
+Loki Mode is an autonomous multi-agent system that follows the RARV cycle to build software from PRDs. It uses specialized agent roles, enforces quality through 8 gates with blind peer review, and maintains episodic/semantic/procedural memory for continuous learning. Projects are classified into simple, standard, or complex tiers that determine the number of phases executed.

package/docs/certification/01-core-concepts/quiz.md

@@ -13,12 +13,12 @@ D) Review, Approve, Release, Validate
 
 ---
 
-**Question 2:** How many specialized agent types does Loki Mode define?
+**Question 2:** How many active providers does Loki Mode support?
 
-A) 8
-B) 25
-C) 41
-D) 50
+A) 3
+B) 4
+C) 5
+D) 6
 
 ---
 
@@ -79,9 +79,9 @@ D) That cyclomatic complexity stays under 10
 **Question 9:** Which AI provider supports full Loki Mode features including parallel agents and the Task tool?
 
 A) OpenAI Codex CLI
-B) Google Gemini CLI
+B) Cline
 C) Claude Code
-D) All three providers equally
+D) All providers equally
 
 ---
 

package/docs/certification/02-enterprise-features/lab.md

@@ -43,8 +43,6 @@ Review the available audit environment variables:
 # LOKI_AUDIT_SYSLOG_HOST       -- Enable syslog forwarding
 # LOKI_AUDIT_SYSLOG_PORT       -- Syslog port (default: 514)
 # LOKI_AUDIT_SYSLOG_PROTO      -- Syslog protocol: udp or tcp
-# LOKI_AUDIT_LEVEL             -- Minimum severity to log
-# LOKI_AUDIT_EXCLUDE_EVENTS    -- Comma-separated events to skip
 ```
 
 To temporarily disable audit logging (not recommended for production):

package/docs/certification/02-enterprise-features/lesson.md

@@ -63,16 +63,6 @@ loki start ./prd.md
 | `LOKI_AUDIT_SYSLOG_PORT` | `514` | Syslog server port |
 | `LOKI_AUDIT_SYSLOG_PROTO` | `udp` | Protocol: `udp` or `tcp` |
 | `LOKI_SYSLOG_FACILITY` | `local0` | Syslog facility (local0-local7) |
-| `LOKI_SYSLOG_SEVERITY` | `info` | Minimum severity to forward |
-
-### Filtering
-
-Control audit log verbosity:
-
-```bash
-export LOKI_AUDIT_LEVEL=warning                           # Minimum severity
-export LOKI_AUDIT_EXCLUDE_EVENTS=api.request,api.response # Skip noisy events
-```
 
 ## OpenTelemetry (OTEL) Observability
 

package/docs/certification/04-production-deployment/lab.md

@@ -71,7 +71,7 @@ ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY docker-compose run \
   loki start ./prd.md
 ```
 
-**Note:** Replace `ANTHROPIC_API_KEY` with the appropriate key for your provider (`OPENAI_API_KEY` for Codex, `GOOGLE_API_KEY` for Gemini).
+**Note:** Replace `ANTHROPIC_API_KEY` with the appropriate key for your provider (for example `OPENAI_API_KEY` for Codex).
 
 ## Step 4: Verify the Dashboard
 

package/docs/certification/05-troubleshooting/lab.md

@@ -60,7 +60,7 @@ cat > .loki/state/circuit-breakers.json << 'EOF'
     "cooldown_until": "2026-02-20T14:30:42Z",
     "failure_window_start": "2026-02-20T14:24:50Z"
   },
-  "api/gemini": {
+  "api/opencode": {
     "state": "HALF_OPEN",
     "failure_count": 0,
     "success_count": 1,
@@ -137,7 +137,7 @@ cat .loki/state/circuit-breakers.json | jq 'to_entries[] | select(.value.state =
 cat .loki/state/circuit-breakers.json | jq '.["api/openai"].cooldown_until'
 
 # Check how many successes the HALF_OPEN circuit needs
-cat .loki/state/circuit-breakers.json | jq '.["api/gemini"].success_count'
+cat .loki/state/circuit-breakers.json | jq '.["api/opencode"].success_count'
 ```
 
 **Questions to answer:**

package/docs/certification/README.md

@@ -27,7 +27,7 @@ The Autonomi Certified Developer certification validates your knowledge of Loki
 - Familiarity with command-line tools (bash, git, npm)
 - Basic understanding of AI/LLM concepts (prompts, models, tokens)
 - Node.js 18+ installed
-- A supported AI provider CLI installed (Claude Code, Codex CLI, or Gemini CLI)
+- A supported AI provider CLI installed (Claude Code, Cline, Codex CLI, Aider, or opencode)
 
 ## How to Use This Program
 

package/docs/certification/answer-key.md

@@ -9,14 +9,14 @@ This file contains answers for all module quizzes and the final certification ex
 | Question | Answer | Explanation |
 |----------|--------|-------------|
 | 1 | B | RARV = Reason, Act, Reflect, Verify |
-| 2 | C | 41 agent types: 37 domain + 4 orchestration |
+| 2 | C | 5 active providers: claude, codex, cline, aider, opencode |
 | 3 | B | After 5 failures, the task moves to `.loki/queue/dead-letter.json` |
 | 4 | C | architecture-strategist is always one of the 3 selected reviewers |
 | 5 | D | 8 quality gates (Static Analysis through Magic Modules Debate); backward-compatibility is a conditional healing-mode auditor, not one of the 8 |
 | 6 | B | Episodic, semantic, and procedural memory |
 | 7 | B | Simple tier uses 3 phases |
 | 8 | C | Gate 7 (Documentation Coverage) checks README presence, docs freshness within 10 commits, and API docs for packages; coverage % is not measured this release |
-| 9 | C | Claude Code supports full features; Codex and Gemini run in degraded mode |
+| 9 | C | Claude Code supports full features; Cline runs reduced, Codex and Aider run degraded |
 | 10 | B | If all 3 reviewers unanimously approve, a Devil's Advocate reviewer runs |
 
 ---
@@ -25,16 +25,16 @@ This file contains answers for all module quizzes and the final certification ex
 
 | Question | Answer | Explanation |
 |----------|--------|-------------|
-| 1 | A | Audit logging is enabled by default since v5.38.0 |
+| 1 | B | Audit logging is enabled by default since v5.38.0 |
 | 2 | C | OTEL activates when `LOKI_OTEL_ENDPOINT` is set to an OTLP endpoint |
 | 3 | C | Default dashboard port is 57374 |
-| 4 | A | `LOKI_ENTERPRISE_AUTH=true` enables token authentication |
-| 5 | D | Syslog supports UDP and TCP via `LOKI_AUDIT_SYSLOG_PROTO` |
-| 6 | D | Without the endpoint set, OTEL uses no-op stubs with zero overhead |
-| 7 | A | `loki enterprise token generate my-token` creates an API token |
-| 8 | D | Set `LOKI_TLS_CERT` and `LOKI_TLS_KEY` to PEM file paths |
-| 9 | A | Controls whether `HUMAN_INPUT.md` can inject directives (disabled by default for security) |
-| 10 | A | `loki enterprise status` shows all enterprise feature states |
+| 4 | B | `LOKI_ENTERPRISE_AUTH=true` enables token authentication |
+| 5 | B | Syslog supports UDP and TCP via `LOKI_AUDIT_SYSLOG_PROTO` |
+| 6 | C | Without the endpoint set, OTEL uses no-op stubs with zero overhead |
+| 7 | B | `loki enterprise token generate my-token` creates an API token |
+| 8 | B | Set `LOKI_TLS_CERT` and `LOKI_TLS_KEY` to PEM file paths |
+| 9 | B | Controls whether `HUMAN_INPUT.md` can inject directives (disabled by default for security) |
+| 10 | B | `loki enterprise status` shows all enterprise feature states |
 
 ---
 
@@ -43,13 +43,13 @@ This file contains answers for all module quizzes and the final certification ex
 | Question | Answer | Explanation |
 |----------|--------|-------------|
 | 1 | B | Goal, Constraints, Context, Output (documented in `skills/agents.md`) |
-| 2 | D | Confidence below 0.40 triggers human decision flag |
+| 2 | C | Confidence below 0.40 triggers human decision flag |
 | 3 | C | Each verification runs independently without seeing the original response |
-| 4 | A | security-sentinel > test-coverage-auditor > performance-oracle > dependency-analyst |
-| 5 | A | Prevents "technically correct but wrong feature" by separating spec and quality reviews |
-| 6 | D | Extraction triggers on novel insights: bug fixes, non-obvious solutions, reusable patterns |
+| 4 | B | security-sentinel > test-coverage-auditor > performance-oracle > dependency-analyst |
+| 5 | B | Prevents "technically correct but wrong feature" by separating spec and quality reviews |
+| 6 | B | Extraction triggers on novel insights: bug fixes, non-obvious solutions, reusable patterns |
 | 7 | C | 4 parallel research agents enhance the plan in the deepen-plan phase |
-| 8 | D | on_file_write triggers lint, typecheck, and secrets scan immediately after writes |
+| 8 | B | on_file_write triggers lint, typecheck, and secrets scan immediately after writes |
 | 9 | C | Stage 1 failure returns to implementation; Stage 2 does not start |
 | 10 | B | Handoffs include completed work, files modified, decisions, open questions, and mistakes |
 
@@ -61,17 +61,34 @@ This file contains answers for all module quizzes and the final certification ex
 |----------|--------|-------------|
 | 1 | C | Dockerfile uses Ubuntu 24.04 as the base image |
 | 2 | B | `.:/workspace:rw` mounts the current directory read-write |
-| 3 | A | Staged autonomy requires human approval before execution |
+| 3 | B | Staged autonomy requires human approval before execution |
 | 4 | C | Default `LOKI_MAX_PARALLEL_AGENTS` is 10 |
-| 5 | A | `loki start --budget 10.00 ./prd.md` or `LOKI_BUDGET_LIMIT=10.00` |
-| 6 | D | The council votes on project completion to prevent premature termination |
-| 7 | B | The session auto-pauses when budget is exceeded |
+| 5 | B | `loki start --budget 10.00 ./prd.md` or `LOKI_BUDGET_LIMIT=10.00` |
+| 6 | B | The council votes on project completion to prevent premature termination |
+| 7 | C | Default dashboard port is 57374 |
 | 8 | B | `LOKI_TLS_CERT` and `LOKI_TLS_KEY` environment variables |
-| 9 | A | Stagnation limit flags when N iterations pass with no git changes |
+| 9 | B | Stagnation limit flags when N iterations pass with no git changes |
 | 10 | B | SANDBOX-SCOPED: `LOKI_ALLOWED_PATHS` restricts which host paths the Docker sandbox bind-mounts writable (`autonomy/sandbox.sh:1222`). It does NOT restrict agent writes inside the workspace. |
 
 ---
 
+## Module 5: Troubleshooting Quiz
+
+| Question | Answer | Explanation |
+|----------|--------|-------------|
+| 1 | B | Closed, Open, Half-Open |
+| 2 | C | 3 failures within 60 seconds open the breaker |
+| 3 | C | 300 seconds (5 minutes) cooldown |
+| 4 | B | After 5 failures the task moves to the dead-letter queue |
+| 5 | B | A context clear is triggered and state is reloaded |
+| 6 | B | `.loki/CONTINUITY.md` is read first |
+| 7 | C | `loki reset retries` resets retry counters only |
+| 8 | A | `LOKI_GATE_MOCK=false` disables the mock integrity gate (`autonomy/run.sh`) |
+| 9 | C | After 10+ total attempts, or the same error with 3 different approaches |
+| 10 | B | Hedging language ("probably", "should be fine") is the red flag |
+
+---
+
 ## Final Certification Exam
 
 | Question | Answer | Question | Answer |

package/docs/certification/certification-exam.md

@@ -17,12 +17,12 @@ D) Review, Approve, Release, Validate
 
 ---
 
-**Question 2:** How many specialized agent types does Loki Mode define?
+**Question 2:** How many active providers does Loki Mode support?
 
-A) 8
-B) 25
-C) 41
-D) 50
+A) 3
+B) 4
+C) 5
+D) 6
 
 ---
 
@@ -83,9 +83,9 @@ D) That cyclomatic complexity stays under 10
 **Question 9:** Which AI provider supports full Loki Mode features including parallel agents and the Task tool?
 
 A) OpenAI Codex CLI
-B) Google Gemini CLI
+B) Cline
 C) Claude Code
-D) All three providers equally
+D) All providers equally
 
 ---
 

package/docs/competitive/emergence-others-analysis.md

@@ -282,7 +282,7 @@ Developers who value open-source tooling, speed, and terminal-native workflows.
 | **Primary Focus** | Web automation | Mobile apps | Coding assistant | Coding assistant | PRD-to-deploy |
 | **Open Source / Source model** | Partial (Agent-E only) | No | Source-available | Yes (Apache-2.0) | Source-available (BUSL-1.1) |
 | **Multi-Provider** | Yes (OpenAI, Azure, Ollama) | Yes (Gemini, Claude) | Partial (Claude models via Bedrock/Vertex/Foundry) | No (GPT only) | Yes (5 providers, 3+ model families) |
-| **Multi-Agent** | Yes (2-agent model) | No | Yes (coordinated teams) | Yes (experimental) | Yes (41 agent types) |
+| **Multi-Agent** | Yes (2-agent model) | No | Yes (coordinated teams) | Yes (experimental) | Yes (agent types) |
 | **Autonomous Iteration** | No (task-level) | No | Partial (/loop, /schedule) | No (requires prompting) | Yes (RARV loop + completion council) |
 | **SDLC Pipeline** | No | No | No | No | Yes (9 phases) |
 | **Code Review** | No | No | Yes (single-pass) | Yes (single-pass) | Yes (3-reviewer blind) |
@@ -457,7 +457,7 @@ This positioning highlights three unique capabilities no competitor offers toget
 | Source model | Open source (Apache-2.0) | Source-available (BUSL-1.1) |
 | Speed | 240+ tokens/sec | Depends on provider |
 | Providers | OpenAI only | 5 providers |
-| Multi-agent | Experimental (isolated) | 41 agent types, 8 domains |
+| Multi-agent | Experimental (isolated) | Agent types across domains |
 | Quality | Single-pass review | 8-gate system |
 | **Loki Mode advantage:** | Autonomous pipeline, multi-provider, mature multi-agent |
 
@@ -542,7 +542,7 @@ Emergence AI charges enterprise contract rates for VPC deployment. Loki Mode is
 |--------|----------|----------|------------|
 | Claude Code Agent SDK enables custom SDLC pipelines | HIGH | 3-6 months | Deepen multi-provider advantage (Agent SDK is Claude-only); publish PRD-to-Deploy benchmarks that prove the integrated pipeline |
 | Claude Code adds autonomous mode natively | HIGH | 6-12 months | Quality gates, memory system, and multi-provider flexibility are structural advantages that cannot be replicated by adding a single feature |
-| Codex CLI adds orchestration layer | MEDIUM | 6-12 months | Codex is OpenAI-only; emphasize 41 agent types, cross-project memory, healing |
+| Codex CLI adds orchestration layer | MEDIUM | 6-12 months | Codex is OpenAI-only; emphasize agent types, cross-project memory, healing |
 | New entrant builds "Rork for full-stack" | MEDIUM | 6-12 months | Add mobile PRD template; Loki Mode's quality gates differentiate from naive generation |
 | Open-source Auto-Claude gains traction | LOW | Ongoing | Already adopted key patterns (v3.4.0); maintain feature lead |
 | Enterprise CI/CD platforms (GitHub, GitLab) add native AI SDLC | HIGH | 12-18 months | Self-hosted, provider-agnostic positioning; these will be vendor-locked |

package/docs/competitive/replit-lovable-analysis.md

@@ -312,7 +312,7 @@ Replit Agent has evolved rapidly through four major versions:
 |-----------|:-----------:|:-----------:|:---------:|
 | Natural language to app | Yes | Yes | Yes (via PRD) |
 | Autonomous execution | 200 min sessions | Per-prompt | Unlimited (budget-gated) |
-| Multi-agent orchestration | Parallel agents (Agent 4) | No | Yes (41 agent types, 8 domains) |
+| Multi-agent orchestration | Parallel agents (Agent 4) | No | Yes (agent types across domains) |
 | Self-testing loop | Yes | No | Yes (RARV cycle) |
 | Code review | No | No | Yes (3-reviewer blind review) |
 | Anti-sycophancy | No | No | Yes (devil's advocate) |
@@ -566,7 +566,7 @@ The term "vibe coding" (coined by Andrej Karpathy) has driven explosive growth i
 1. **Credit fatigue will drive users to transparent pricing** -- Loki Mode's BYOK model becomes more attractive as credit complaints mount
 2. **Production quality will become the differentiator** -- Prototype generation is commoditized; who can ship production code wins
 3. **Enterprise adoption requires compliance** -- SOC 2, on-premises, audit logs become table stakes
-4. **Multi-agent orchestration is the next frontier** -- Replit Agent 4's parallel agents are primitive; Loki Mode's 41 agent types are ahead
+4. **Multi-agent orchestration is the next frontier** -- Replit Agent 4's parallel agents are primitive; Loki Mode's agent types are ahead
 5. **Brownfield/legacy support is underserved** -- No competitor addresses existing codebases; Loki Mode's healing system is unique
 6. **Self-hosted alternatives are emerging** -- Projects like December (fully local Lovable/Replit alternative with own LLM) and Tinykit (self-hosted AI app builder) validate the demand for local-first, privacy-preserving alternatives. Loki Mode is already positioned here.
 

package/docs/cursor-comparison.md

@@ -164,7 +164,7 @@ For critical changes:
 - "Trillions of tokens" deployed
 - Hundreds of concurrent agents
 
-**Loki Mode:** Benchmarks only (SWE-bench, HumanEval). No 1M+ LoC projects demonstrated.
+**Loki Mode:** No 1M+ LoC projects demonstrated.
 
 ### 2. Empirical Iteration
 Cursor learned through failure:
@@ -177,7 +177,7 @@ Cursor learned through failure:
 ### 3. Simplicity Principle
 > "A surprising amount of the system's behavior comes down to how we prompt the agents. The harness and models matter, but the prompts matter more."
 
-**Loki Mode:** More elaborate infrastructure (8 gates, 41 agent types, memory systems). May be over-engineered for some use cases.
+**Loki Mode:** More elaborate infrastructure (8 gates, agent role prompts, memory systems). May be over-engineered for some use cases.
 
 ---
 

package/docs/dashboard-guide.md

@@ -35,7 +35,7 @@ The dashboard automatically syncs with Loki Mode when it's running, polling `das
 The sidebar provides navigation and system status at a glance.
 
 #### Logo & Version
-- Loki Mode branding with current version (v5.40.0)
+- Loki Mode branding with the running version
 - Version updates automatically from server state
 
 #### Theme Toggle
@@ -305,6 +305,18 @@ The dashboard includes several security measures:
 | **Python injection fix** | `completion-council.sh` sanitizes inputs to prevent code injection |
 | **CORS configuration** | Configurable via `LOKI_DASHBOARD_CORS` environment variable (default: localhost only) |
 
+### Host allowlist (DNS rebinding defense)
+
+The dashboard refuses any request whose `Host` header is not loopback (`127.0.0.1`, `localhost`, `[::1]`) with 403, except `GET /health` and `GET /metrics`, which are always exempt so container probes (Host = pod IP) and scrapers work.
+
+| Deployment | Behavior |
+|------------|----------|
+| Local default (bound to 127.0.0.1) | Only loopback Hosts; add others via `LOKI_DASHBOARD_ALLOWED_HOSTS` |
+| Non-loopback bind (`0.0.0.0` or `LOKI_DASHBOARD_HOST` set) with `LOKI_ENTERPRISE_AUTH` or OIDC enabled | Any Host accepted; authentication is the boundary |
+| Non-loopback bind without auth | Only loopback plus `LOKI_DASHBOARD_ALLOWED_HOSTS` |
+
+`LOKI_DASHBOARD_ALLOWED_HOSTS` is a comma-separated list of host names (no port), e.g. `loki.example.com,10.0.0.5`. Helm: `config.dashboardAllowedHosts`. Terraform (aws-ecs): `dashboard_allowed_hosts`.
+
 ---
 
 ## Technical Architecture
@@ -465,6 +477,6 @@ Useful for:
 ## Related Documentation
 
 - [Core Workflow](../references/core-workflow.md) - RARV cycle details
-- [Agent Types](../references/agent-types.md) - 41 agent definitions
+- [Agent Types](../references/agent-types.md) - Agent role definitions
 - [Quality Control](../references/quality-control.md) - Quality gates system
 - [Memory System](../references/memory-system.md) - Memory architecture

package/docs/dev/architecture-reference.md

@@ -21,9 +21,9 @@ regardless of vote count. D13 requires every council/review agent to pin its
 model explicitly rather than inherit the session model.
 
 ## Legacy System Healing (introduced v6.67.0)
-- **Current in v7.18.0**: Still active, no breaking changes since v6.67.0. Note: in v7.4.20 the `legacy-healing-auditor` reviewer was gated on healing-mode signals to avoid firing on non-healing changes.
+- **Status**: Still active, no breaking changes since v6.67.0. Note: in v7.4.20 the `legacy-healing-auditor` reviewer was gated on healing-mode signals to avoid firing on non-healing changes.
 - **Inspired by**: Amazon AGI Lab's "How Agentic AI Helps Heal Systems We Can't Replace"
-- **CLI**: `loki heal <path> [--phase archaeology|stabilize|isolate|modernize|validate]` (`autonomy/loki:9916`)
+- **CLI**: `loki heal <path> [--phase archaeology|stabilize|isolate|modernize|validate]` (`autonomy/loki` `cmd_heal`)
 - **Principles**: Friction-as-semantics, failure-first learning, universal adapters, incremental healing, institutional knowledge preservation
 - **Artifacts**: `.loki/healing/` (friction-map.json, failure-modes.json, institutional-knowledge.md)
 - **Review**: `legacy-healing-auditor` specialist added to code review pool (gated)

package/docs/dev/project-structure.md

@@ -33,7 +33,7 @@ skills/                     # On-demand skill modules (v3.0 architecture)
   testing.md                # Playwright, E2E, property-based testing
   production.md             # HN patterns, CI/CD, context management
   troubleshooting.md        # Common issues, red flags, fallbacks
-  agents.md                 # 41 agent types, structured prompting
+  agents.md                 # Agent types, structured prompting
   artifacts.md              # Generation, code transformation
   patterns-advanced.md      # OptiMind, k8s-valkey, Constitutional AI
   parallel-workflows.md     # Git worktrees, parallel streams, auto-merge

package/docs/enterprise/architecture.md

@@ -1,4 +1,4 @@
-# Loki Mode v5.51.0 -- Enterprise Architecture
+# Loki Mode Enterprise Architecture
 
 ## Overview
 
@@ -27,14 +27,14 @@ Loki Mode's enterprise layer is a non-breaking extension to the core autonomous
 |  | OTEL Bridge      |  | Policy Engine    |  | Audit Trail      |    |
 |  | (otel.js)        |  | (engine.js)      |  | (log.js)         |    |
 |  | Spans, Metrics   |  | YAML/JSON rules  |  | Hash-chained     |    |
-|  | OTLP/HTTP export |  | 4 enf. points    |  | JSONL, syslog    |    |
+|  | OTLP/HTTP export |  | policy checks    |  | JSONL, syslog    |    |
 |  +--------+---------+  +--------+---------+  +--------+---------+    |
 |           |                      |                      |            |
 |  +--------v---------+  +--------v---------+  +--------v---------+   |
 |  | Span Helpers      |  | Cost Controller  |  | Compliance       |   |
 |  | (spans.js)        |  | (cost.js)        |  | (compliance.js)  |   |
-|  | RARV, agents,     |  | Token budgets,   |  | SOC 2, ISO 27001 |   |
-|  | quality gates     |  | alerts           |  | GDPR reports     |   |
+|  | RARV, agents,     |  | Token budgets,   |  | Audit evidence   |   |
+|  | quality gates     |  | alerts           |  | report templates |   |
 |  +-------------------+  +--------+---------+  +--------+---------+   |
 |                                  |                      |            |
 |  +------------------+  +--------v---------+  +--------v---------+   |
@@ -115,9 +115,12 @@ The `OTLPExporter` class batches spans (flush at 100 or every 5 seconds) and sen
 
 **Source:** `src/policies/engine.js` (core), `src/policies/index.js` (public API), `src/policies/types.js` (validators)
 
-The policy engine provides governance-as-code through declarative policy files. Policies are loaded from `.loki/policies.json` or `.loki/policies.yaml` and evaluated synchronously at enforcement points.
+The policy engine provides governance-as-code through declarative policy files. Policies are loaded from `.loki/policies.json` or `.loki/policies.yaml` and evaluated synchronously.
 
-**Enforcement Points:**
+**Policy points.** The engine accepts the four points below, but at runtime
+`autonomy/run.sh` evaluates only `pre_execution`. The other three are not
+invoked by Loki itself; they are available only to code that calls the engine
+API directly.
 
 | Point | Purpose | Context Fields |
 |-------|---------|----------------|
@@ -223,14 +226,14 @@ All integrations extend `IntegrationAdapter`, which provides:
 **Jira Integration:** `src/integrations/jira/`
 
 - `api-client.js` -- Jira Cloud REST API v3 client with Basic Auth, rate limiting, 10MB response cap
-- `sync-manager.js` -- Bidirectional sync: epic-to-PRD import, RARV-to-Jira status mapping, quality report posting
+- `sync-manager.js` -- Epic-to-PRD import, RARV-to-Jira status mapping, quality report posting
 - `epic-converter.js` -- Converts Jira epics with children to PRD format
 - `webhook-handler.js` -- Handles inbound Jira webhooks
 
 **Linear Integration:** `src/integrations/linear/`
 
 - `client.js` -- Linear GraphQL API client with rate limit tracking, auto-retry
-- `sync.js` -- Bidirectional sync using the reusable adapter pattern
+- `sync.js` -- Linear sync using the reusable adapter pattern
 - `config.js` -- Config loader from `.loki/config.yaml` with minimal YAML parser
 
 **GitHub Integration:** `src/integrations/github/`