loki-mode 7.26.0 → 7.27.0

package/dashboard/__init__.py

@@ -7,7 +7,7 @@ Modules:
     control: Session control API (start/stop/pause/resume)
 """
 
-__version__ = "7.26.0"
+__version__ = "7.27.0"
 
 # Expose the control app for easy import
 try:

package/dashboard/static/index.html

@@ -6523,7 +6523,7 @@ var LokiDashboard=(()=>{var Ee=Object.defineProperty;var rt=Object.getOwnPropert
         <p>App runner not started</p>
         <p class="hint">App runner will start after the first successful build iteration.</p>
       </div>
-    `}_attachEventListeners(){let e=this.shadowRoot;if(!e)return;let t=e.querySelector('[data-action="restart"]'),i=e.querySelector('[data-action="stop"]');t&&t.addEventListener("click",()=>this._handleRestart()),i&&i.addEventListener("click",()=>this._handleStop())}_escapeHtml(e){return e?String(e).replace(/&/g,"&amp;").replace(/</g,"&lt;").replace(/>/g,"&gt;").replace(/"/g,"&quot;"):""}};customElements.define("loki-app-status",Q);var Ke={not_initialized:{color:"var(--loki-text-muted, #71717a)",label:"No app yet",pulse:!1},starting:{color:"var(--loki-yellow, #ca8a04)",label:"Starting",pulse:!0},running:{color:"var(--loki-green, #16a34a)",label:"Running",pulse:!0},stale:{color:"var(--loki-yellow, #ca8a04)",label:"Stale",pulse:!1},completed:{color:"var(--loki-text-muted, #a1a1aa)",label:"Completed",pulse:!1},failed:{color:"var(--loki-red, #dc2626)",label:"Could not start",pulse:!1},crashed:{color:"var(--loki-red, #dc2626)",label:"Crashed",pulse:!1},stopped:{color:"var(--loki-text-muted, #a1a1aa)",label:"Stopped",pulse:!1},error:{color:"var(--loki-text-muted, #71717a)",label:"Status unavailable",pulse:!1},unknown:{color:"var(--loki-text-muted, #71717a)",label:"Unknown",pulse:!1}},X=class extends h{static get observedAttributes(){return["api-url","theme"]}constructor(){super(),this._api=null,this._pollInterval=null,this._visibilityHandler=null,this._status=null,this._errors=null,this._error=null,this._lastDataHash=null,this._detailsOpen=!1}connectedCallback(){super.connectedCallback(),this._setupApi(),this.render(),this._loadData(),this._startPolling()}disconnectedCallback(){super.disconnectedCallback(),this._stopPolling()}attributeChangedCallback(e,t,i){t!==i&&(e==="api-url"&&this._api&&(this._api.baseUrl=i,this._loadData()),e==="theme"&&this._applyTheme())}_setupApi(){let e=this.getAttribute("api-url")||window.location.origin;this._api=g({baseUrl:e})}_startPolling(){this._pollInterval=setInterval(()=>this._loadData(),3e3),this._visibilityHandler=()=>{document.hidden?this._pollInterval&&(clearInterval(this._pollInterval),this._pollInterval=null):this._pollInterval||(this._loadData(),this._pollInterval=setInterval(()=>this._loadData(),3e3))},document.addEventListener("visibilitychange",this._visibilityHandler)}_stopPolling(){this._pollInterval&&(clearInterval(this._pollInterval),this._pollInterval=null),this._visibilityHandler&&(document.removeEventListener("visibilitychange",this._visibilityHandler),this._visibilityHandler=null)}async _loadData(){try{let e=await this._api.getAppRunnerStatus(),t=e?.status||"not_initialized",i=null;if(t==="crashed"||t==="failed")try{i=await this._api.getAppRunnerErrors(50)}catch{i=null}let a=JSON.stringify({status:t,port:e?.port,url:e?.url,crash:e?.crash_count,errLen:i?.lines?.length||0}),s=this._error!==null;if(a===this._lastDataHash&&!s)return;this._lastDataHash=a,this._status=e,this._errors=i,this._error=null,this.render()}catch(e){this._error||(this._error=`Could not read app status: ${e.message}`,this.render())}}_isValidUrl(e){if(!e)return!1;try{let t=new URL(e);return t.protocol==="http:"||t.protocol==="https:"}catch{return!1}}async _handleRestart(){try{await this._api.restartApp(),this._loadData()}catch(e){this._error=`Restart failed: ${e.message}`,this.render()}}_handleRefresh(){let e=this.shadowRoot;if(!e)return;let t=e.querySelector("iframe.preview-frame"),i=this._status;if(t&&i&&this._isValidUrl(i.url)){let a=(i.url.includes("?")?"&":"?")+"_t="+Date.now();t.src=i.url+a}}_handleOpenExternal(){let e=this._status;e&&this._isValidUrl(e.url)&&window.open(e.url,"_blank","noopener")}_toggleDetails(){this._detailsOpen=!this._detailsOpen,this.render()}_getStyles(){return`
+    `}_attachEventListeners(){let e=this.shadowRoot;if(!e)return;let t=e.querySelector('[data-action="restart"]'),i=e.querySelector('[data-action="stop"]');t&&t.addEventListener("click",()=>this._handleRestart()),i&&i.addEventListener("click",()=>this._handleStop())}_escapeHtml(e){return e?String(e).replace(/&/g,"&amp;").replace(/</g,"&lt;").replace(/>/g,"&gt;").replace(/"/g,"&quot;"):""}};customElements.define("loki-app-status",Q);var Ke={not_initialized:{color:"var(--loki-text-muted, #71717a)",label:"No app yet",pulse:!1},starting:{color:"var(--loki-yellow, #ca8a04)",label:"Starting",pulse:!0},running:{color:"var(--loki-green, #16a34a)",label:"Running",pulse:!0},stale:{color:"var(--loki-yellow, #ca8a04)",label:"Stale",pulse:!1},completed:{color:"var(--loki-text-muted, #a1a1aa)",label:"Completed",pulse:!1},failed:{color:"var(--loki-red, #dc2626)",label:"Could not start",pulse:!1},crashed:{color:"var(--loki-red, #dc2626)",label:"Crashed",pulse:!1},stopped:{color:"var(--loki-text-muted, #a1a1aa)",label:"Stopped",pulse:!1},error:{color:"var(--loki-text-muted, #71717a)",label:"Status unavailable",pulse:!1},unknown:{color:"var(--loki-text-muted, #71717a)",label:"Unknown",pulse:!1}},X=class extends h{static get observedAttributes(){return["api-url","theme"]}constructor(){super(),this._api=null,this._pollInterval=null,this._visibilityHandler=null,this._status=null,this._errors=null,this._error=null,this._lastDataHash=null,this._detailsOpen=!1}connectedCallback(){super.connectedCallback(),this._setupApi(),this.render(),this._loadData(),this._startPolling()}disconnectedCallback(){super.disconnectedCallback(),this._stopPolling()}attributeChangedCallback(e,t,i){t!==i&&(e==="api-url"&&this._api&&(this._api.baseUrl=i,this._loadData()),e==="theme"&&this._applyTheme())}_setupApi(){let e=this.getAttribute("api-url")||window.location.origin;this._api=g({baseUrl:e})}_startPolling(){this._pollInterval=setInterval(()=>this._loadData(),3e3),this._visibilityHandler=()=>{document.hidden?this._pollInterval&&(clearInterval(this._pollInterval),this._pollInterval=null):this._pollInterval||(this._loadData(),this._pollInterval=setInterval(()=>this._loadData(),3e3))},document.addEventListener("visibilitychange",this._visibilityHandler)}_stopPolling(){this._pollInterval&&(clearInterval(this._pollInterval),this._pollInterval=null),this._visibilityHandler&&(document.removeEventListener("visibilitychange",this._visibilityHandler),this._visibilityHandler=null)}async _loadData(){try{let e=await this._api.getAppRunnerStatus(),t=e?.status||"not_initialized",i=null;if(t==="crashed"||t==="failed")try{i=await this._api.getAppRunnerErrors(50)}catch{i=null}let a=JSON.stringify({status:t,port:e?.port,url:e?.url,crash:e?.crash_count,errLen:i?.lines?.length||0,healthOk:e?.last_health?.ok===!0}),s=this._error!==null;if(a===this._lastDataHash&&!s)return;this._lastDataHash=a,this._status=e,this._errors=i,this._error=null,this.render()}catch(e){this._error||(this._error=`Could not read app status: ${e.message}`,this.render())}}_isValidUrl(e){if(!e)return!1;try{let t=new URL(e);return t.protocol==="http:"||t.protocol==="https:"}catch{return!1}}async _handleRestart(){try{await this._api.restartApp(),this._loadData()}catch(e){this._error=`Restart failed: ${e.message}`,this.render()}}_handleRefresh(){let e=this.shadowRoot;if(!e)return;let t=e.querySelector("iframe.preview-frame"),i=this._status;if(t&&i&&this._isValidUrl(i.url)){let a=(i.url.includes("?")?"&":"?")+"_t="+Date.now();t.src=i.url+a}}_handleOpenExternal(){let e=this._status;e&&this._isValidUrl(e.url)&&window.open(e.url,"_blank","noopener")}_toggleDetails(){this._detailsOpen=!this._detailsOpen,this.render()}_getStyles(){return`
       .preview { padding: 16px; font-family: var(--loki-font-family, system-ui, -apple-system, sans-serif); color: var(--loki-text-primary, #201515); }
       .header { display: flex; align-items: center; justify-content: space-between; margin-bottom: 12px; gap: 12px; flex-wrap: wrap; }
       .header-left { display: flex; align-items: center; gap: 10px; }

package/docs/COMPARISON.md

@@ -24,10 +24,10 @@
 
 | Feature | **Loki Mode** | **Devin** | **Codex** | **Cursor** | **Kiro** | **Antigravity** | **Amazon Q** | **OpenCode** |
 |---------|--------------|-----------|-----------|------------|----------|-----------------|--------------|--------------|
-| **Multi-Agent** | 41 agents in 8 swarms | Single | Single | Up to 8 parallel | Background | Manager Surface | Multiple types | 4 built-in |
+| **Multi-Agent** | 41 prompt-defined agent roles in 8 domains adopted per phase; parallel review council + optional worktree streams (Claude), sequential elsewhere | Single | Single | Up to 8 parallel | Background | Manager Surface | Multiple types | 4 built-in |
 | **Orchestration** | Full orchestrator | N/A | N/A | Git worktree | Hooks | Manager view | Workflow | Subagents |
 | **Parallel Exec** | 10+ Haiku, 4 impl (worktree) | No | No | 8 max | Yes | Yes | Yes | Yes |
-| **Agent Swarms** | Eng, Ops, Business, Data, Product, Growth, Review | N/A | N/A | N/A | N/A | N/A | 3 types | N/A |
+| **Agent Domains** | Eng, Ops, Business, Data, Product, Growth, Review | N/A | N/A | N/A | N/A | N/A | 3 types | N/A |
 
 ---
 
@@ -151,7 +151,7 @@
 | **Multi-Agent Verification** | Model diversity (Claude vs OpenAI, 54% improvement) | 3 blind reviewers + devil's advocate | Different approach (N/A for Claude Code - only Claude models) |
 | **Quality Gates** | Built-in verification loops | 7 explicit gates + anti-sycophancy | **Loki Mode** |
 | **Memory System** | Not documented | 3-tier episodic/semantic/procedural | **Loki Mode** |
-| **Agent Specialization** | Custom Zen Agents | 41 pre-defined specialized agents | **Loki Mode** |
+| **Agent Specialization** | Custom Zen Agents | 41 pre-defined specialized agent roles | **Loki Mode** |
 | **CI Failure Analysis** | Explicit pattern with auto-fix | DevOps agent only | **ADOPTED from Zencoder** |
 | **Review Comment Resolution** | Auto-apply simple changes | Manual review | **ADOPTED from Zencoder** |
 | **Dependency Management** | Scheduled PRs, one group at a time | Mentioned only | **ADOPTED from Zencoder** |
@@ -180,7 +180,7 @@
 
 1. **Quality Control**: 7 explicit gates + blind review + devil's advocate vs built-in loops
 2. **Memory System**: 3-tier (episodic/semantic/procedural) with cross-project learning
-3. **Agent Specialization**: 41 pre-defined specialized agents across 8 swarms
+3. **Agent Specialization**: 41 pre-defined specialized agent roles across 8 domains
 4. **Anti-Sycophancy**: CONSENSAGENT patterns prevent reviewer groupthink
 5. **Autonomy Design**: Minimal human intervention from PRD to production
 6. **Research Foundation**: 10+ academic papers integrated vs proprietary
@@ -203,7 +203,7 @@
 |---------|--------------|---------|-----------------|------------|-----------------|---------------------|----------------|
 | **Stars** | 594 | 11,903 | 35K+ | 26K+ | 13.7K | N/A | N/A |
 | **npm/wk** | 6.1K | 21.4K | N/A | N/A | N/A | N/A | N/A |
-| **Agents** | 41 in 8 swarms | 11 agents | Fresh per task | 108 agents | Swarm-based | 32 agents | N/A |
+| **Agents** | 41 roles in 8 domains | 11 agents | Fresh per task | 108 agents | Swarm-based | 32 agents | N/A |
 | **Skills** | Progressive disclosure | 6 slash commands | N/A | 129 skills | N/A | 35 skills | Memory focus |
 | **Multi-Provider** | Yes (Claude/Codex/Gemini) | 3 CLIs (separate) | No | No | No | No | No |
 | **Memory System** | 3-tier (episodic/semantic/procedural) | None | N/A | N/A | Hybrid | N/A | SQLite+FTS5 |
@@ -236,7 +236,7 @@ These are patterns from competing projects that are **practically and scientific
 | **Constitutional AI Integration** | Principles-based self-critique from Anthropic research | None have this |
 | **Anti-Sycophancy (CONSENSAGENT)** | Blind review + devil's advocate prevents groupthink | None have this |
 | **Provider Abstraction Layer** | Clean degradation from full-featured to sequential-only | Claude-only projects can't degrade |
-| **41 Specialized Agents** | Purpose-built agents in 8 swarms vs generic | agents (108) has more but less organized |
+| **41 Specialized Agent Roles** | Purpose-built role definitions in 8 domains vs generic; Loki gates every role's output through blind review + council | agents (108) has more but less organized |
 | **Research Foundation** | 10+ academic papers integrated with citations | Most have no research backing |
 
 ### Superpowers Deep-Dive (35K+ Stars)
@@ -258,7 +258,7 @@ Plugin marketplace architecture with unprecedented scale:
 | Pattern | Description | Loki Mode Status |
 |---------|-------------|------------------|
 | **72 Plugins** | Modular, focused plugins instead of monolith | Different approach (progressive disclosure) |
-| **108 Agents** | Specialized agents for specific domains | 41 agents in Loki Mode |
+| **108 Agents** | Specialized agents for specific domains | 41 agent roles in Loki Mode |
 | **129 Skills** | Skills as first-class objects | 10 skills in skills/ |
 | **Four-Tier Model Strategy** | Explicit tier selection with constraints | Similar to Loki Mode tiers |
 
@@ -342,7 +342,7 @@ Tiered agent architecture with explicit escalation:
 
 | Agent | Killer Feature |
 |-------|---------------|
-| **Loki Mode** | Minimal-human-intervention full SDLC from any spec (PRD, GitHub issue, or YAML), 41 agents in 8 swarms, Constitutional AI, anti-sycophancy, cross-project learning, code transformation, property-based testing |
+| **Loki Mode** | Minimal-human-intervention full SDLC from any spec (PRD, GitHub issue, or YAML), 41 agent roles in 8 domains, Constitutional AI, anti-sycophancy, cross-project learning, code transformation, property-based testing |
 | **Devin** | Full software engineer persona, Slack integration, 67% PR merge rate |
 | **OpenAI Codex** | Skills marketplace, $skill-creator, GPT-5.2-Codex, secure sandbox |
 | **Cursor** | 8 parallel agents, BugBot, Memories, $10B valuation, Composer model (250 tok/s) |
@@ -358,7 +358,7 @@ Tiered agent architecture with explicit escalation:
 | Dimension | Loki Mode Advantage |
 |-----------|-------------------|
 | **Autonomy** | Designed for high autonomy with minimal human intervention |
-| **Multi-Agent** | 41 specialized agents in 8 swarms vs 1-8 in competitors |
+| **Multi-Agent** | 41 prompt-defined agent roles in 8 domains adopted per phase (parallel review council + optional worktree streams on Claude, sequential elsewhere) vs 1-8 in competitors, with all output gated by blind review + council |
 | **Quality** | 11 gates + blind review + devil's advocate + property-based testing |
 | **Research** | 10+ academic papers integrated vs proprietary/undisclosed |
 | **Anti-Sycophancy** | Only agent with CONSENSAGENT-based blind review |

package/docs/COMPETITIVE-ANALYSIS.md

@@ -39,7 +39,7 @@ GSD is the closest competitor -- a context engineering system that spawns fresh
 | **Agent Count** | 41 types | 64+ agents | 5 roles | Unlimited | 8 parallel | 1 autonomous |
 | **Parallel Execution** | Yes (multi-agent) | Yes (swarms) | Sequential | Yes (crews) | Yes (8 worktrees) | Yes (fleet) |
 | **Published Benchmarks** | 98.78% HumanEval (self-reported, max 3 retries) | None | 85.9-87.7% HumanEval | None | ~250 tok/s | 15% complex tasks |
-| **SWE-bench Score** | 99.67% patch gen (unevaluated, 299/300) | Unknown | Unknown | Unknown | Unknown | 15% complex |
+| **SWE-bench Score** | Not measured (patch generation harness exists; official evaluator not run, so no resolve rate exists) | Unknown | Unknown | Unknown | Unknown | 15% complex |
 | **Full SDLC** | Yes (8 phases) | Yes | Partial | Partial | No | Partial |
 | **Business Ops** | **Yes (8 agents)** | No | No | No | No | No |
 | **Enterprise Security** | `--dangerously-skip-permissions` | MCP sandboxed | Sandboxed | Audit logs, RBAC | Staged autonomy | Sandboxed |
@@ -213,49 +213,30 @@ GSD is the closest competitor -- a context engineering system that spawns fresh
 
 **Failed Problems (after RARV):** HumanEval/32, HumanEval/50
 
-### SWE-bench Lite Results (Full 300 Problems)
+### SWE-bench Lite: Honest Status
 
-**Direct Claude (Single Agent Baseline):**
-
-| Metric | Value |
-|--------|-------|
-| **Patch Generation** | **99.67%** |
-| Generated | 299/300 problems |
-| Errors | 1 |
-| Model | Claude Opus 4.5 |
-| Time | 6.17 hours |
-
-**Loki Mode Multi-Agent (with RARV):**
-
-| Metric | Value |
-|--------|-------|
-| **Patch Generation** | **99.67%** |
-| Generated | 299/300 problems |
-| Errors/Timeouts | 1 |
-| Model | Claude Opus 4.5 |
-| Time | 3.5 hours |
-
-**Three-Way Comparison:**
-
-| System | SWE-bench Patch Gen | Notes |
-|--------|---------------------|-------|
-| **Direct Claude** | **99.67%** (299/300) | Single agent, minimal overhead |
-| **Loki Mode (multi-agent)** | **99.67%** (299/300) | 4-agent pipeline with RARV |
-| Devin | ~15% complex tasks | Commercial, different benchmark |
-
-**Key Finding:** After timeout optimization (Architect: 60s->120s), the multi-agent RARV pipeline matches direct Claude's performance on SWE-bench. Both achieve 99.67% patch generation rate.
+**Loki Mode has NO SWE-bench score.** What exists: a harness that GENERATED candidate
+patches for 299 of 300 SWE-bench Lite problems (Claude Opus 4.5; single-agent run
+6.17h, 4-agent RARV pipeline 3.5h). Patch GENERATION only means a diff was produced.
+It does NOT mean the patch fixes the issue. The official SWE-bench evaluator (apply
+patch, run the repo's test suite) was never run, so there is no resolve rate, and
+generation rates must not be compared against other tools' task-resolution rates
+(e.g. Devin's independently-tested task success). For context, frontier model resolve
+rates on SWE-bench Verified are in the ~85-90 percent range; any higher number from
+any tool should be treated with suspicion, including ours.
 
-**Note:** Patches generated; full validation (resolve rate) requires running the Docker-based SWE-bench harness to apply patches and execute test suites.
+**What a fair claim looks like:** "A reproducible patch-generation harness exists;
+resolve rate is not yet measured." Nothing stronger.
 
 ---
 
 ## Critical Gaps to Address
 
-### Priority 1: Benchmarks (COMPLETED)
-- **Gap:** ~~No published HumanEval or SWE-bench scores~~ RESOLVED
-- **Result:** 98.17% HumanEval Pass@1 (beats MetaGPT by 10.5%)
-- **Result:** 99.67% SWE-bench Lite patch generation (299/300)
-- **Next:** Run full SWE-bench harness for resolve rate validation
+### Priority 1: Benchmarks (PARTIAL)
+- **Real:** 98.78% HumanEval Pass@1 with RARV (162/164, reproducible harness; 98.17% single-agent baseline)
+- **NOT real:** SWE-bench. Patch generation ran (299/300 diffs produced); the official
+  evaluator never ran, so there is NO resolve rate and no comparable score.
+- **Next:** run a contamination-resistant evaluator end-to-end (or publish nothing for SWE-bench)
 
 ### Priority 2: Security Model (Critical for Enterprise)
 - **Gap:** Relies on `--dangerously-skip-permissions`

package/docs/INSTALLATION.md

@@ -2,7 +2,7 @@
 
 The flagship product of [Autonomi](https://www.autonomi.dev/). Loki Mode is a spec-driven autonomous builder with a built-in trust layer that takes any spec to a deployed product and verifies completion with evidence (quality gates plus a completion council), not just a "done" claim. Complete installation instructions for all platforms and use cases.
 
-**Version:** v7.26.0
+**Version:** v7.27.0
 
 ---
 

package/docs/auto-claude-comparison.md

@@ -85,7 +85,7 @@ Loki Mode is built on peer-reviewed research:
 **Verdict: Loki Mode wins** - Academically grounded.
 
 ### 2. Specialized Agent Types
-Loki Mode has 41 predefined agent types across 6 swarms:
+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
@@ -142,13 +142,16 @@ Loki Mode has 14 quality gates:
 
 ### 6. Published Benchmarks
 Loki Mode:
-- HumanEval: 98.78% Pass@1 (162/164)
-- SWE-bench: 99.67% patch generation (299/300)
-- Documented methodology with reproducible results
+- 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** - Verified performance claims.
+**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)
@@ -248,7 +251,7 @@ Loki Mode now incorporates proven patterns from Cursor's large-scale agent deplo
 **Loki Mode is better if you want:**
 - Research-backed architecture
 - Full spec-to-product lifecycle (not just coding)
-- 41 specialized agents
+- 41 specialized agent roles
 - Anti-sycophancy measures
 - MIT license
 - No subscription requirement

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

@@ -70,9 +70,9 @@ Or via the environment variable `LOKI_COMPLEXITY=simple|standard|complex`.
 
 ## Agents
 
-Loki Mode defines **41 specialized agent types** organized into **8 swarms**:
+Loki Mode defines **41 specialized agent types** organized into **8 domains**:
 
-| Swarm | Agent Count | Examples |
+| Domain | Agent Count | Examples |
 |-------|-------------|----------|
 | Engineering | 8 | frontend, backend, database, mobile, api, qa, perf, infra |
 | Operations | 8 | devops, sre, security, monitor, incident, release, cost, compliance |
@@ -179,4 +179,4 @@ 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 swarms, enforces quality through 9 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 41 agent types organized into 8 domains, enforces quality through 9 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/competitive/emergence-others-analysis.md

@@ -457,7 +457,7 @@ This positioning highlights three unique capabilities no competitor offers toget
 | Open source | Yes (Apache-2.0) | Yes |
 | Speed | 240+ tokens/sec | Depends on provider |
 | Providers | OpenAI only | 5 providers |
-| Multi-agent | Experimental (isolated) | 41 agent types, 8 swarms |
+| Multi-agent | Experimental (isolated) | 41 agent types, 8 domains |
 | Quality | Single-pass review | 10-gate system |
 | **Loki Mode advantage:** | Autonomous pipeline, multi-provider, mature multi-agent |
 

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 swarms) |
+| Multi-agent orchestration | Parallel agents (Agent 4) | No | Yes (41 agent types, 8 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) |

package/docs/cursor-comparison.md

@@ -122,7 +122,7 @@ BOOTSTRAP -> DISCOVERY -> ARCHITECTURE -> INFRASTRUCTURE
      -> DEVELOPMENT -> QA -> DEPLOYMENT -> GROWTH (continuous)
 ```
 
-**41 Specialized Agent Types across 8 swarms:**
+**41 Specialized Agent Roles across 8 domains** (prompt-defined specifications the orchestrator adopts per phase; parallel review council and optional worktree streams on Claude Code, sequential on other providers):
 - Engineering (8 types)
 - Operations (8 types)
 - Business (8 types)

package/docs/prd-purple-lab-platform.md

@@ -198,7 +198,7 @@ Each gate: name, status (pass/fail/pending), details expandable
 ### Agent Activity
 
 Real-time grid showing which of the 41 agent types are active:
-- Agent name, type, swarm, model tier (Opus/Sonnet/Haiku)
+- Agent name, type, domain, model tier (Opus/Sonnet/Haiku)
 - Current task
 - Status (working/idle/completed)
 - Duration

package/docs/show-hn-post.md

@@ -2,13 +2,13 @@
 
 ## Title
 
-Show HN: Loki Mode - PRD in, tested code out (41 agents, 9 quality gates, RARV self-verification)
+Show HN: Loki Mode - PRD in, tested code out (41 agent roles, 9 quality gates, RARV self-verification)
 
 ## Body
 
 I built Loki Mode because I got tired of the copy-paste loop between AI coding assistants and my terminal. I wanted to hand over a PRD and get back a working, tested codebase -- not perfect, but a solid starting point.
 
-**What it does:** You give it a Product Requirements Document. It breaks the work into tasks, dispatches them across 41 specialized agent types organized into 8 swarms (engineering, operations, business, data, product, growth, review, orchestration), and runs every iteration through a self-verification loop called RARV: Reason, Act, Reflect, Verify. The idea is that the system catches its own mistakes before you have to.
+**What it does:** You give it a Product Requirements Document. It breaks the work into tasks across 41 specialized agent roles organized into 8 domains (engineering, operations, business, data, product, growth, review, orchestration) -- prompt-defined specifications the orchestrator adopts per phase, with parallel review (blind council) and optional worktree streams on Claude Code, sequential on other providers -- and runs every iteration through a self-verification loop called RARV: Reason, Act, Reflect, Verify. The idea is that the system catches its own mistakes before you have to.
 
 **Quality gates:** 9 automated gates including 3-reviewer blind review (agents review each other's work without seeing prior reviews), anti-sycophancy checks (a devil's advocate pass on unanimous approvals), and mock/mutation detection. These are not foolproof, but they catch a surprising number of issues that single-pass generation misses.
 

package/loki-ts/dist/loki.js

@@ -1,5 +1,5 @@
 // @bun
-var f8=Object.defineProperty;var u8=($)=>$;function c8($,Q){this[$]=u8.bind(null,Q)}var g=($,Q)=>{for(var Z in Q)f8($,Z,{get:Q[Z],enumerable:!0,configurable:!0,set:c8.bind(Q,Z)})};var k=($,Q)=>()=>($&&(Q=$($=0)),Q);var X1=import.meta.require;var F$={};g(F$,{lokiDir:()=>P,homeLokiDir:()=>o1,findRepoRootForVersion:()=>d1,REPO_ROOT:()=>f});import{resolve as n,dirname as l1}from"path";import{fileURLToPath as p8}from"url";import{existsSync as L1}from"fs";import{homedir as l8}from"os";function d8(){let $=j$;for(let Q=0;Q<6;Q++){if(L1(n($,"VERSION"))&&L1(n($,"autonomy/run.sh")))return $;let Z=l1($);if(Z===$)break;$=Z}return n(j$,"..","..","..")}function d1($){let Q=$;for(let Z=0;Z<6;Z++){if(L1(n(Q,"VERSION"))&&L1(n(Q,"autonomy/run.sh")))return Q;let z=l1(Q);if(z===Q)break;Q=z}return n($,"..","..","..")}function P(){return process.env.LOKI_DIR??n(process.cwd(),".loki")}function o1(){return n(l8(),".loki")}var j$,f;var y=k(()=>{j$=l1(p8(import.meta.url));f=d8()});import{readFileSync as o8}from"fs";import{resolve as n8,dirname as a8}from"path";import{fileURLToPath as s8}from"url";function k1(){if($1!==null)return $1;let $="7.26.0";if(typeof $==="string"&&$.length>0)return $1=$,$1;try{let Q=a8(s8(import.meta.url)),Z=d1(Q);$1=o8(n8(Z,"VERSION"),"utf-8").trim()}catch{$1="unknown"}return $1}var $1=null;var n1=k(()=>{y()});var E$={};g(E$,{runOrThrow:()=>t8,run:()=>j,commandVersion:()=>i8,commandExists:()=>v,ShellError:()=>a1});async function j($,Q={}){let Z=Bun.spawn({cmd:[...$],stdout:"pipe",stderr:"pipe",env:Q.env?{...process.env,...Q.env}:process.env,cwd:Q.cwd}),z,K;if(Q.timeoutMs&&Q.timeoutMs>0)z=setTimeout(()=>{try{Z.kill("SIGTERM")}catch{}K=setTimeout(()=>{try{Z.kill("SIGKILL")}catch{}},2000)},Q.timeoutMs);try{let[H,X,q]=await Promise.all([new Response(Z.stdout).text(),new Response(Z.stderr).text(),Z.exited]);return{stdout:H,stderr:X,exitCode:q}}finally{if(z)clearTimeout(z);if(K)clearTimeout(K)}}async function t8($,Q={}){let Z=await j($,Q);if(Z.exitCode!==0)throw new a1(`command failed (${Z.exitCode}): ${$.join(" ")}`,Z.exitCode,Z.stdout,Z.stderr);return Z}async function v($){let Q=r8($),Z=await j(["sh","-c",`command -v ${Q}`],{timeoutMs:5000});if(Z.exitCode===0)return Z.stdout.trim()||null;return null}function r8($){if(!/^[A-Za-z0-9._/-]+$/.test($))throw Error(`refused to shell-escape suspect token: ${$}`);return $}async function i8($,Q="--version"){if(!await v($))return null;let z=await j([$,Q],{timeoutMs:5000});if(z.exitCode!==0)return null;return((z.stdout||z.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var a1;var d=k(()=>{a1=class a1 extends Error{message;exitCode;stdout;stderr;constructor($,Q,Z,z){super($);this.message=$;this.exitCode=Q;this.stdout=Z;this.stderr=z;this.name="ShellError"}}});function a($){return e8?"":$}var e8,T,N,_,KZ,A,R,h,J;var c=k(()=>{e8=(process.env.NO_COLOR??"").length>0;T=a("\x1B[0;31m"),N=a("\x1B[0;32m"),_=a("\x1B[1;33m"),KZ=a("\x1B[0;34m"),A=a("\x1B[0;36m"),R=a("\x1B[1m"),h=a("\x1B[2m"),J=a("\x1B[0m")});import{existsSync as U7}from"fs";async function Q1(){if(B1!==void 0)return B1;let $="/opt/homebrew/bin/python3.12";if(U7($))return B1=$,$;let Q=await v("python3.12");if(Q)return B1=Q,Q;let Z=await v("python3");return B1=Z,Z}async function Z1($,Q={}){let Z=await Q1();if(!Z)return{stdout:"",stderr:"python3 not found",exitCode:127};return j([Z,"-c",$],Q)}var B1;var H1=k(()=>{d()});var d$={};g(d$,{runStatus:()=>N7});import{existsSync as b,readFileSync as q1,readdirSync as v$,statSync as f$}from"fs";import{resolve as D,basename as P7}from"path";import{homedir as L7}from"os";async function j7(){if(await v("jq"))return!0;return process.stdout.write(`${T}Error: jq is required but not installed.${J}
+var f8=Object.defineProperty;var u8=($)=>$;function c8($,Q){this[$]=u8.bind(null,Q)}var g=($,Q)=>{for(var Z in Q)f8($,Z,{get:Q[Z],enumerable:!0,configurable:!0,set:c8.bind(Q,Z)})};var k=($,Q)=>()=>($&&(Q=$($=0)),Q);var X1=import.meta.require;var F$={};g(F$,{lokiDir:()=>P,homeLokiDir:()=>o1,findRepoRootForVersion:()=>d1,REPO_ROOT:()=>f});import{resolve as n,dirname as l1}from"path";import{fileURLToPath as p8}from"url";import{existsSync as L1}from"fs";import{homedir as l8}from"os";function d8(){let $=j$;for(let Q=0;Q<6;Q++){if(L1(n($,"VERSION"))&&L1(n($,"autonomy/run.sh")))return $;let Z=l1($);if(Z===$)break;$=Z}return n(j$,"..","..","..")}function d1($){let Q=$;for(let Z=0;Z<6;Z++){if(L1(n(Q,"VERSION"))&&L1(n(Q,"autonomy/run.sh")))return Q;let z=l1(Q);if(z===Q)break;Q=z}return n($,"..","..","..")}function P(){return process.env.LOKI_DIR??n(process.cwd(),".loki")}function o1(){return n(l8(),".loki")}var j$,f;var y=k(()=>{j$=l1(p8(import.meta.url));f=d8()});import{readFileSync as o8}from"fs";import{resolve as n8,dirname as a8}from"path";import{fileURLToPath as s8}from"url";function k1(){if($1!==null)return $1;let $="7.27.0";if(typeof $==="string"&&$.length>0)return $1=$,$1;try{let Q=a8(s8(import.meta.url)),Z=d1(Q);$1=o8(n8(Z,"VERSION"),"utf-8").trim()}catch{$1="unknown"}return $1}var $1=null;var n1=k(()=>{y()});var E$={};g(E$,{runOrThrow:()=>t8,run:()=>j,commandVersion:()=>i8,commandExists:()=>v,ShellError:()=>a1});async function j($,Q={}){let Z=Bun.spawn({cmd:[...$],stdout:"pipe",stderr:"pipe",env:Q.env?{...process.env,...Q.env}:process.env,cwd:Q.cwd}),z,K;if(Q.timeoutMs&&Q.timeoutMs>0)z=setTimeout(()=>{try{Z.kill("SIGTERM")}catch{}K=setTimeout(()=>{try{Z.kill("SIGKILL")}catch{}},2000)},Q.timeoutMs);try{let[H,X,q]=await Promise.all([new Response(Z.stdout).text(),new Response(Z.stderr).text(),Z.exited]);return{stdout:H,stderr:X,exitCode:q}}finally{if(z)clearTimeout(z);if(K)clearTimeout(K)}}async function t8($,Q={}){let Z=await j($,Q);if(Z.exitCode!==0)throw new a1(`command failed (${Z.exitCode}): ${$.join(" ")}`,Z.exitCode,Z.stdout,Z.stderr);return Z}async function v($){let Q=r8($),Z=await j(["sh","-c",`command -v ${Q}`],{timeoutMs:5000});if(Z.exitCode===0)return Z.stdout.trim()||null;return null}function r8($){if(!/^[A-Za-z0-9._/-]+$/.test($))throw Error(`refused to shell-escape suspect token: ${$}`);return $}async function i8($,Q="--version"){if(!await v($))return null;let z=await j([$,Q],{timeoutMs:5000});if(z.exitCode!==0)return null;return((z.stdout||z.stderr).split(/\r?\n/)[0]?.trim()??"")||null}var a1;var d=k(()=>{a1=class a1 extends Error{message;exitCode;stdout;stderr;constructor($,Q,Z,z){super($);this.message=$;this.exitCode=Q;this.stdout=Z;this.stderr=z;this.name="ShellError"}}});function a($){return e8?"":$}var e8,T,N,_,KZ,A,R,h,J;var c=k(()=>{e8=(process.env.NO_COLOR??"").length>0;T=a("\x1B[0;31m"),N=a("\x1B[0;32m"),_=a("\x1B[1;33m"),KZ=a("\x1B[0;34m"),A=a("\x1B[0;36m"),R=a("\x1B[1m"),h=a("\x1B[2m"),J=a("\x1B[0m")});import{existsSync as U7}from"fs";async function Q1(){if(B1!==void 0)return B1;let $="/opt/homebrew/bin/python3.12";if(U7($))return B1=$,$;let Q=await v("python3.12");if(Q)return B1=Q,Q;let Z=await v("python3");return B1=Z,Z}async function Z1($,Q={}){let Z=await Q1();if(!Z)return{stdout:"",stderr:"python3 not found",exitCode:127};return j([Z,"-c",$],Q)}var B1;var H1=k(()=>{d()});var d$={};g(d$,{runStatus:()=>N7});import{existsSync as b,readFileSync as q1,readdirSync as v$,statSync as f$}from"fs";import{resolve as D,basename as P7}from"path";import{homedir as L7}from"os";async function j7(){if(await v("jq"))return!0;return process.stdout.write(`${T}Error: jq is required but not installed.${J}
 `),process.stdout.write(`Install with:
 `),process.stdout.write(`  brew install jq    (macOS)
 `),process.stdout.write(`  apt install jq     (Debian/Ubuntu)
@@ -787,4 +787,4 @@ Set LOKI_LEGACY_BASH=1 to force the bash CLI for every command.
 `),2}default:return process.stderr.write(`Unknown command: ${Q}
 `),process.stderr.write(v8),2}}g$();process.on("SIGINT",()=>process.exit(130));process.on("SIGTERM",()=>process.exit(143));var l3=await p3(Bun.argv.slice(2));process.exit(l3);
 
-//# debugId=7BD97DA7996A924D64756E2164756E21
+//# debugId=07AC5AC0D01821A064756E2164756E21

package/mcp/__init__.py

@@ -57,4 +57,4 @@ try:
 except ImportError:
     __all__ = ['mcp']
 
-__version__ = '7.26.0'
+__version__ = '7.27.0'

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "loki-mode",
-  "version": "7.26.0",
+  "version": "7.27.0",
   "description": "Loki Mode by Autonomi. Autonomous spec-to-product system: takes a PRD, GitHub issue, OpenAPI/JSON/YAML, or one-line brief to a deployed app via the RARV-C closure loop with 11 quality gates. Provider-agnostic (Claude Code, OpenAI Codex, Cline, Aider).",
   "keywords": [
     "agent",

package/providers/codex.sh

@@ -32,7 +32,7 @@ PROVIDER_CLI="codex"
 # VERIFIED: exec --full-auto confirmed in codex exec --help (v0.98.0)
 # --full-auto: sets --ask-for-approval on-request + --sandbox workspace-write (v0.98.0)
 # Alternative: "exec --dangerously-bypass-approvals-and-sandbox" (legacy, no sandbox)
-PROVIDER_AUTONOMOUS_FLAG="exec --full-auto"
+PROVIDER_AUTONOMOUS_FLAG="exec --full-auto --skip-git-repo-check"
 PROVIDER_PROMPT_FLAG=""
 PROVIDER_PROMPT_POSITIONAL=true
 
@@ -119,7 +119,7 @@ provider_version() {
 provider_invoke() {
     local prompt="$1"
     shift
-    codex exec --full-auto "$prompt" "$@"
+    codex exec --full-auto --skip-git-repo-check "$prompt" "$@"
 }
 
 # Model tier to effort level parameter (Codex uses effort, not separate models)
@@ -210,6 +210,7 @@ provider_invoke_with_tier() {
     codex exec \
         --ask-for-approval never \
         --sandbox danger-full-access \
+        --skip-git-repo-check \
         "${extra_flags[@]}" \
         "$prompt" "$@"
 }

package/references/agent-types.md

@@ -6,11 +6,11 @@ Complete definitions and capabilities for all 41 specialized agent types.
 
 ## Overview
 
-Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37 domain agents + 4 orchestration agents). The orchestrator spawns only the agents needed for your project -- typically 5-10 for simple projects, more for complex ones.
+Loki Mode has 41 specialized agent roles across 8 domains (37 domain agents + 4 orchestration agents) - prompt-defined specifications the orchestrator adopts per phase, not separate processes. Parallelism on Claude Code comes from the blind review council, the adversarial reviewer, and optional git-worktree streams; on other providers everything runs sequentially. The orchestrator activates only the roles needed for your project -- typically 5-10 for simple projects, more for complex ones.
 
 ---
 
-## Engineering Swarm (8 types)
+## Engineering Domain (8 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -25,7 +25,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Operations Swarm (8 types)
+## Operations Domain (8 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -40,7 +40,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Business Swarm (8 types)
+## Business Domain (8 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -55,7 +55,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Data Swarm (3 types)
+## Data Domain (3 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -65,7 +65,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Product Swarm (3 types)
+## Product Domain (3 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -75,7 +75,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Growth Swarm (4 types)
+## Growth Domain (4 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -86,7 +86,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Review Swarm (3 types)
+## Review Domain (3 types)
 
 | Agent | Capabilities |
 |-------|-------------|
@@ -96,7 +96,7 @@ Loki Mode has 41 predefined agent types organized into 8 specialized swarms (37
 
 ---
 
-## Orchestration Swarm (4 types)
+## Orchestration Domain (4 types)
 
 > **Source:** [Cursor Scaling Learnings](./cursor-learnings.md) - patterns proven at large agent scale
 

package/references/agents.md

@@ -1,6 +1,6 @@
 # Agent Type Definitions
 
-Complete specifications for all 41 specialized agent types in the Loki Mode multi-agent system (37 domain agents + 4 orchestration agents).
+Complete specifications for all 41 specialized agent roles in Loki Mode (37 domain agents + 4 orchestration agents). These are prompt-defined role specifications the orchestrator adopts per phase, not separate processes.
 
 **Note:** These are agent TYPE definitions, not a fixed count. Loki Mode dynamically spawns agents based on project needs - a simple todo app might use 5-10 agents, while a complex startup spawns more as needed.
 
@@ -45,7 +45,7 @@ Update after every task completion.
 
 ---
 
-## Engineering Swarm (8 Agents)
+## Engineering Domain (8 Agents)
 
 ### eng-frontend
 **Capabilities:**
@@ -241,7 +241,7 @@ Update after every task completion.
 
 ---
 
-## Operations Swarm (8 Agents)
+## Operations Domain (8 Agents)
 
 ### ops-devops
 **Capabilities:**
@@ -437,7 +437,7 @@ Update after every task completion.
 
 ---
 
-## Business Swarm (8 Agents)
+## Business Domain (8 Agents)
 
 ### biz-marketing
 **Capabilities:**
@@ -634,7 +634,7 @@ Update after every task completion.
 
 ---
 
-## Data Swarm (3 Agents)
+## Data Domain (3 Agents)
 
 ### data-ml
 **Capabilities:**
@@ -710,7 +710,7 @@ Update after every task completion.
 
 ---
 
-## Product Swarm (3 Agents)
+## Product Domain (3 Agents)
 
 ### prod-pm
 **Capabilities:**
@@ -787,7 +787,7 @@ Update after every task completion.
 
 ---
 
-## Review Swarm (3 Agents)
+## Review Domain (3 Agents)
 
 ### review-code
 **Capabilities:**
@@ -875,7 +875,7 @@ Update after every task completion.
 
 ---
 
-## Growth Swarm (4 Agents)
+## Growth Domain (4 Agents)
 
 ### growth-hacker
 **Capabilities:**

package/references/business-ops.md

@@ -1,6 +1,6 @@
 # Business Operations Reference
 
-Workflows and procedures for business swarm agents.
+Workflows and procedures for business domain agents.
 
 ## Marketing Operations
 

package/references/competitive-analysis.md

@@ -182,7 +182,7 @@ Dexter shows value of domain specialization. Our 41 agent types follow this patt
    - Most haven't scaled across enterprise
 
 ### Loki Mode Alignment
-- Multi-agent architecture (41 types, 8 swarms)
+- Multi-agent architecture (41 role types, 8 domains)
 - Plan Agents (orchestrator, planner)
 - Execution Agents (eng-*, ops-*, biz-*)
 - Security controls (LOKI_SANDBOX_MODE, LOKI_BLOCKED_COMMANDS)

package/skills/agents.md

@@ -1,6 +1,6 @@
 # Agent Dispatch & Structured Prompting
 
-> **Full agent type definitions:** See `references/agent-types.md` for complete 41 agent role specifications across 8 swarms (Engineering, Operations, Business, Data, Product, Growth, Review, Orchestration).
+> **Full agent type definitions:** See `references/agent-types.md` for complete 41 agent role specifications across 8 domains (Engineering, Operations, Business, Data, Product, Growth, Review, Orchestration). These are prompt-defined specifications the orchestrator adopts per phase; parallelism on Claude Code comes from the blind review council, the adversarial reviewer, and optional git-worktree streams, sequential on other providers.
 
 ---
 
@@ -249,8 +249,8 @@ Priority order for context:
 
 See `references/agent-types.md` for complete specifications. Summary:
 
-| Swarm | Agent Types | Count |
-|-------|-------------|-------|
+| Domain | Agent Types | Count |
+|--------|-------------|-------|
 | Engineering | frontend, backend, database, mobile, api, qa, perf, infra | 8 |
 | Operations | devops, sre, security, monitor, incident, release, cost, compliance | 8 |
 | Business | marketing, sales, finance, legal, support, hr, investor, partnerships | 8 |

package/skills/providers.md

@@ -6,7 +6,7 @@ Loki Mode supports four AI providers for autonomous execution.
 
 > **CLI Flags Verified:** The autonomous mode flags have been verified against actual CLI help output:
 > - Claude: `--dangerously-skip-permissions` (verified)
-> - Codex: `--full-auto` (recommended, v0.98.0) or `exec --dangerously-bypass-approvals-and-sandbox` (legacy)
+> - Codex: `exec --full-auto --skip-git-repo-check` (the harness invocation; --skip-git-repo-check required on fresh non-git dirs) or `exec --dangerously-bypass-approvals-and-sandbox` (legacy)
 
 | Feature | Claude Code | OpenAI Codex | Cline CLI | Aider |
 |---------|-------------|--------------|-----------|-------|
@@ -56,7 +56,7 @@ Task(model="haiku", ...)   # Fast tier (parallelize)
 
 ---
 
-## OpenAI Codex CLI (Degraded Mode)
+## OpenAI Codex CLI (Experimental, Degraded Mode)
 
 **Best for:** Teams standardized on OpenAI. Accepts feature tradeoffs.
 
@@ -70,7 +70,7 @@ Task(model="haiku", ...)   # Fast tier (parallelize)
 **Invocation:**
 ```bash
 # Recommended (v0.98.0+)
-codex --full-auto "$prompt"
+codex exec --full-auto --skip-git-repo-check "$prompt"
 
 # Legacy (still supported)
 codex exec --dangerously-bypass-approvals-and-sandbox "$prompt"