Prepare agentctl for public release

.agent/README.md

@@ -1,52 +1,58 @@
-# .agent/ — agentctl assets
+# .agent/
 
-This directory ships the prompts and config that `agentctl` uses.
+Prompts and config that `agentctl` ships with.
 
 ```
 .agent/
   prompts/
-    gpt55-orchestrator.md   # lead orchestrator: plan, route, final review, decide
-    claude-reviewer.md      # architecture reviewer: infra risk and design validation
-    deepseek-worker.md      # execution worker: token-efficient scout / executor
+    deepseek-worker.md          # retrieval worker: read-only tools, returns receipts
+    gpt55-reviewer.md           # optional second opinion from GPT-5.5
+    gpt55-orchestrator.md       # fallback template for the GPT-5.5 reviewer
+    claude-reviewer.md          # optional architecture and risk review
+    gpt55-operator-onboarding.md  # paste-in briefing for driving agentctl from a ChatGPT chat
   opencode/
-    deepseek-worker.md      # OpenCode execution worker definition
-  config.example.json       # copy to config.json to customize
-  config.json               # your local config (gitignored)
+    deepseek-worker.md          # OpenCode agent definition for the worker
+  config.example.json           # shipped defaults, copy to config.json to change them
+  config.json                   # your local config (gitignored)
 ```
 
-## Config resolution order
+## Config resolution
 
-`agentctl` loads the **first** file it finds, merged over built-in defaults:
+`agentctl` merges the first file it finds over the built-in defaults:
 
-1. `<cwd>/.agent/config.json`  — per-repo override
-2. `<install>/.agent/config.json` — your global override
-3. `<install>/.agent/config.example.json` — shipped example
+1. `<cwd>/.agent/config.json` for a per-repo override
+2. `<install>/.agent/config.json` for your global override
+3. `<install>/.agent/config.example.json`, the shipped example
 
-## Wiring DeepSeek (OpenCode)
+## DeepSeek via OpenCode
 
-The default deepseek command is:
+Default command:
 
 ```
 opencode run --agent deepseek-worker --format json "<prompt>"
 ```
 
-To use `--agent deepseek-worker`, install the agent definition:
+Install the agent definition once:
 
 ```
 mkdir -p ~/.config/opencode/agent
 cp .agent/opencode/deepseek-worker.md ~/.config/opencode/agent/
 ```
 
-If you don't install the agent, switch the deepseek args to the model form
-(already present as `fallback_args` in `config.example.json`):
+Without it, switch the deepseek `args` to the model form (already in `fallback_args` in the example):
 
 ```
 opencode run --model deepseek/deepseek-chat --format json "<prompt>"
 ```
 
-## Wiring GPT-5.5
+## GPT-5.5
 
-`config.example.json` ships `gpt55.command = "CONFIGURE_ME"`. Until you set a
-real command, the orchestrator falls back to a local Claude/heuristic planner
-and every run is flagged `GPT_LIMIT_ACTIVE`. Point `command`/`args` at your real
-GPT-5.5 CLI or a thin API wrapper to enable the Lead Orchestrator.
+The example config uses `"type": "http_openai"`: one HTTP request to the OpenAI API, key from `OPENAI_API_KEY`. Without
+a key the review is skipped and the run flags `GPT_LIMIT_ACTIVE`. The run itself still completes.
+
+No API key? Use `--route deepseek-chatgpt`. agentctl writes a prompt to paste into ChatGPT, and `agentctl ingest <run>`
+adds the answer to the report.
+
+## Claude
+
+`claude -p "<prompt>" --output-format json`, using the Claude Code login. Used for `--route deepseek-claude`.

.agent/config.example.json

@@ -1,19 +1,18 @@
 {
   "orchestrator": {
-    "primary": "gpt55",
-    "fallback": "claude",
+    "default_caller": "local",
     "default_route": "deepseek",
     "review_required": true,
     "review_policy": "best_effort_gpt55",
-    "default_caller": "gpt55",
     "run_timeout_seconds": 120,
     "agent_timeout_seconds": 90,
     "review_timeout_seconds": 45,
     "idle_timeout_seconds": 30,
     "role_policy": {
-      "gpt55": "Lead Orchestrator: planning, routing, final review, and final decision authority",
-      "claude": "Architecture Reviewer: infrastructure architecture, risk review, and lightweight code design validation",
-      "deepseek": "Execution Worker: token-efficient repository inspection, tests, logs, context packs, and low-risk implementation"
+      "deepseek": "Retrieval worker: read-only tools, returns raw tool receipts, no conclusions",
+      "local": "Logic: synthesizes the answer and the decision from the receipts",
+      "gpt55": "Optional best-effort reviewer, never on the critical path",
+      "claude": "Optional reviewer for architecture, design and risk"
     }
   },
   "agents": {
@@ -70,17 +69,11 @@
       ]
     }
   },
-  "limits": {
-    "detect_patterns": true,
-    "fallback_on_limit": true
-  },
   "worktrees": {
     "enabled": true,
     "base_dir": ".agent-runs/worktrees"
   },
   "safety": {
-    "block_dangerous_commands": true,
-    "deepseek_may_edit": false,
-    "claude_may_edit": true
+    "deepseek_may_edit": false
   }
-}
+}

.agent/prompts/gpt55-operator-onboarding.md

@@ -17,7 +17,7 @@ back the output, then you review and decide. Work in English, keep it concise.
 
 ## HOW THE USER DRIVES IT
 
-- Interactive console: `agentctl console`. In the console the user just types a task (no quotes). Meta commands:
+- Interactive console: `agentctl` (or `agentctl shell`). In the console the user just types a task (no quotes). Meta commands:
   `/cd <path>` — set target repo
   `/status`
   `/review`

.github/workflows/ci.yml

@@ -0,0 +1,21 @@
+name: tests
+
+on:
+  push:
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: macos-latest
+    strategy:
+      matrix:
+        python: ["3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+      - run: python -m unittest discover -s tests -v

AGENTS.md

@@ -10,9 +10,8 @@ This repository uses `agentctl` as the local multi-agent workflow. One rule:
   prose is not trusted. It carries the token-heavy file/log reading.
 - **Local / Claude — Logic.** The stable reasoner and decision-maker. Synthesizes the answer from
   DeepSeek's receipts. This is the default; it never depends on a flaky channel.
-- **GPT-5.5 — Optional best-effort review.** A second opinion only when it has a working route. The
-  free OpenCode/OAuth route is unreliable and a deterministic one needs OpenAI API billing, so GPT-5.5
-  is never load-bearing. If it returns empty, the run completes on local synthesis.
+- **GPT-5.5 — Optional best-effort review.** A second opinion when a route is configured (API key or the
+  ChatGPT handoff). It is never load-bearing. If it returns empty, the run completes on local synthesis.
 
 ## Default Workflow
 

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 nzrbits
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.