@iinm/plain-agent 1.9.1 → 1.9.2

package/README.md

@@ -1,32 +1,71 @@
-<p align="center">
-  <img src="https://pub-0bb49aa929f242d49c89ed8c297932b5.r2.dev/plain-agent/plain-agent-logo.png" alt="plain-agent logo" width="320">
-</p>
-
 # Plain Agent
 
-A lightweight CLI-based coding agent.
+A lightweight, capable coding agent for the terminal.
 
 - **Multi-provider** — Use Claude, GPT, Gemini, or any OpenAI-compatible model.
-- **Compact system prompt** — Under 3.5 KB, keeping per-request overhead and cost low ([src/prompt.mjs](https://github.com/iinm/plain-agent/blob/main/src/prompt.mjs)).
-- **Fine-grained approval rules** — Auto-approve commands by name, arguments,
-  and file paths using regex patterns
-  ([config.predefined.json#autoApproval](https://github.com/iinm/plain-agent/blob/main/config/config.predefined.json)).
-- **Path validation** — Restricts access to the working directory.
-  Git-ignored and untracked files require explicit approval.
-- **Sandboxed execution** — Run the agent's shell commands inside a Docker
-  container with network access restricted to allowlisted destinations
-  (e.g., `registry.npmjs.org` only for `npm install`).
-- **Plain-text memory** — Task state is persisted as Markdown files under
-  `.plain-agent/memory/`, easy to review.
+  <details>
+  <summary>Predefined models</summary>
+  
+  ```
+  # Model+Variant (Platform+Variant)
+  claude-haiku-4-5+thinking-16k (platform: anthropic+default)
+  claude-haiku-4-5+thinking-32k (platform: anthropic+default)
+  claude-sonnet-4-6+thinking-high (platform: anthropic+default)
+  claude-sonnet-4-6+thinking-max (platform: anthropic+default)
+  claude-opus-4-7+thinking-high (platform: anthropic+default)
+  claude-opus-4-7+thinking-max (platform: anthropic+default)
+  claude-haiku-4-5+thinking-16k-bedrock (platform: bedrock+default)
+  claude-haiku-4-5+thinking-32k-bedrock (platform: bedrock+default)
+  claude-sonnet-4-6+thinking-high-bedrock (platform: bedrock+default)
+  claude-sonnet-4-6+thinking-max-bedrock (platform: bedrock+default)
+  claude-opus-4-7+thinking-high-bedrock (platform: bedrock+default)
+  claude-opus-4-7+thinking-max-bedrock (platform: bedrock+default)
+  gemini-3-flash-preview+thinking-medium (platform: gemini+default)
+  gemini-3-flash-preview+thinking-high (platform: gemini+default)
+  gemini-3.1-pro-preview+thinking-medium (platform: gemini+default)
+  gemini-3.1-pro-preview+thinking-high (platform: gemini+default)
+  gemini-3-flash-preview+thinking-medium-vertex-ai (platform: vertex-ai+default)
+  gemini-3-flash-preview+thinking-high-vertex-ai (platform: vertex-ai+default)
+  gemini-3.1-pro-preview+thinking-medium-vertex-ai (platform: vertex-ai+default)
+  gemini-3.1-pro-preview+thinking-high-vertex-ai (platform: vertex-ai+default)
+  gpt-5.4-mini+thinking-medium (platform: openai+default)
+  gpt-5.4-mini+thinking-high (platform: openai+default)
+  gpt-5.4-mini+thinking-xhigh (platform: openai+default)
+  gpt-5.5+thinking-medium (platform: openai+default)
+  gpt-5.5+thinking-high (platform: openai+default)
+  gpt-5.5+thinking-xhigh (platform: openai+default)
+  gpt-5.2-chat+thinking-medium-azure (platform: azure+openai)
+  gpt-oss-120b+fireworks (platform: openai-compatible+fireworks)
+  glm-5+vertex-ai (platform: vertex-ai+default)
+  glm-5.1+fireworks (platform: openai-compatible+fireworks)
+  glm-5.1+novita (platform: openai-compatible+novita)
+  kimi-k2.6+fireworks (platform: openai-compatible+fireworks)
+  kimi-k2.6+novita (platform: openai-compatible+novita)
+  deepseek-v4-pro+novita (platform: openai-compatible+novita)
+  deepseek-v4-pro+fireworks (platform: openai-compatible+fireworks)
+  minimax-m2.7+fireworks (platform: openai-compatible+fireworks)
+  minimax-m2.7+novita (platform: openai-compatible+novita)
+  qwen3.6-plus+fireworks (platform: openai-compatible+fireworks)
+  qwen3.6-27b+novita (platform: openai-compatible+novita)
+  nova-2-lite+bedrock (platform: bedrock+default)
+  claude-haiku-4-5+thinking-16k-bedrock-converse (platform: bedrock+default)
+  ```
+  </details>
+
+
+- **Approval rules & path validation** — Auto-approve tool uses by name and arguments using regex patterns ([config.predefined.json#autoApproval](https://github.com/iinm/plain-agent/blob/main/config/config.predefined.json)); restrict file access to the working directory — git-ignored and untracked files require explicit approval ([src/toolInputValidator.mjs](https://github.com/iinm/plain-agent/blob/main/src/toolInputValidator.mjs)).
+- **Sandboxed execution** — Run agent commands in a Docker container with a read-only project root and no network; writable mode and allowlisted network destinations can be enabled as needed.
+- **Plain-text memory** — Task state is saved as Markdown files under `.plain-agent/memory/` for easy review.
 - **Extensible** — Define prompts and subagents in Markdown. Connect MCP servers.
   Supports Claude Code plugins and `.claude/` commands, subagents, and skills.
 
 ## Limitations
 
+- **Path validation only covers tool arguments** — Path validation restricts only paths explicitly passed as tool-use arguments; it cannot control file access inside arbitrary scripts. Always use sandboxed execution when allowing arbitrary script execution.
+- **No session persistence** — Sessions are not persisted. Start a fresh session and use memory files (`.plain-agent/memory/`) instead.
 - **Sequential subagent execution** — Subagents run one at a time rather than
   in parallel. The trade-off is full visibility: every step is streamed to
   your terminal so you can follow exactly what each subagent is doing.
-- **No session persistence** — Sessions are not persisted. Start a fresh session and use memory files (`.plain-agent/memory/`) instead.
 
 ## Requirements
 
@@ -157,7 +196,7 @@ Create the configuration.
 ```
 
 ```js
-// OpenAI-compatible provider (Ollama) example with a custom model
+// Ollama example with a custom model
 {
   "platforms": [
     {
@@ -184,7 +223,6 @@ Create the configuration.
     }
   ]
 }
-
 ```
 </details>
 
@@ -327,9 +365,7 @@ Files are loaded in the following order. Settings in later files override earlie
         ├── (4) config.local.json  # Project-specific local configuration (including secrets)
         ├── prompts/               # Project-specific prompts
         ├── agents/                # Project-specific agent roles
-        ├── memory/                # Task-specific memory files
-        ├── sandbox/               # Sandbox runner scripts
-        └── tmp/                   # Agent scratch space
+        └── sandbox/               # Sandbox runner scripts
 ```
 
 ### Example

package/config/config.predefined.json

@@ -914,6 +914,30 @@
       }
     },
 
+    {
+      "name": "gpt-oss-120b",
+      "variant": "fireworks",
+      "platform": {
+        "name": "openai-compatible",
+        "variant": "fireworks"
+      },
+      "model": {
+        "format": "openai-messages",
+        "config": {
+          "model": "accounts/fireworks/models/gpt-oss-120b"
+        }
+      },
+      "cost": {
+        "currency": "USD",
+        "unit": "1M",
+        "costs": {
+          "prompt_tokens": 0.15,
+          "prompt_tokens_details.cached_tokens": -0.14,
+          "completion_tokens": 0.6 
+        }
+      }
+    },
+
     {
       "name": "glm-5",
       "variant": "vertex-ai",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@iinm/plain-agent",
-  "version": "1.9.1",
+  "version": "1.9.2",
   "description": "A lightweight CLI-based coding agent",
   "license": "MIT",
   "type": "module",