@agent-smith/feat-agents 0.0.1

package/dist/agents/agent-smith-search.yml

@@ -0,0 +1,40 @@
+tool:
+  name: agent-smith-search
+  description: run web search or open web pages
+  arguments:
+      prompt:
+          description: the web search to perform or question about a web page
+          required: true
+description: An agent to run web search or open web pages
+category: "agent-smith"
+prompt: |-
+    {prompt}
+template:
+    system: |-
+      You are a web search agent for the Agent Smith project, a toolkit to run AI agents.
+
+      {file:../fragments/workspace.md}
+
+      Project's context helper files:
+
+      - `AGENTS.md`: general high level context map of the project to navigate
+model: qwen35b
+backend: llamaccp
+inferParams:
+    min_p: 0
+    top_k: 20
+    top_p: 0.85
+    temperature: 0.6
+    repetition_penalty: 1
+    chat_template_kwargs:
+      enable_thinking: true
+      preserve_thinking: true
+variables:
+  required:
+    workspace:
+      description: The local directory path where to operate
+toolsList:
+  - shell
+  - python
+  - ddsearch
+  - read-webpage

package/dist/agents/agent-smith-sql.yml

@@ -0,0 +1,31 @@
+tool:
+    description: retrieve data from the Agent Smith config database that references agents, actions, workflows and other features
+    arguments:
+        prompt: 
+            description: the question about the config or data to retrieve
+description: Create an sql query
+category: internal/doc
+prompt: |-
+    Schema:
+    {schema}
+    
+    Question: {prompt}
+
+    Important: output ONLY raw SQL code. Example: SELECT * FROM tablename
+model: text2sql
+template:
+    system: |-
+        You are given a database schema and a natural language question. Generate the SQL query that answers the question.
+
+        - Use only tables and columns from the provided schema
+        - Use uppercase SQL keywords (SELECT, FROM, WHERE, etc.)
+        - Use SQLite-compatible syntax
+    assistant: "```sql\n"
+    stop: 
+        - "```"
+inferParams:
+    temperature: 0
+variables:
+    required:
+        schema:
+            description: the database sql schema

package/dist/agents/agent-smith-state.yml

@@ -0,0 +1,40 @@
+description: A project state mananagement agent
+category: internal
+prompt: |-
+    Check if the project has all the necessary context and documentation files:
+
+    - `AGENTS.md`: general high level context map of the project
+    - `.agents/documentation/codebase-summary.md`: high level codebase summary
+    - `.agents/documentation/project-overview.md`: a high level project overview
+    
+    Output only a report in this format:
+
+    ```yaml
+    - `AGENTS.md`: ok (or todo)
+    - `.agents/documentation/codebase-summary.md`: ok (or todo)
+    - `.agents/documentation/project-overview.md`: ok (or todo)
+    ```
+template:
+    system: |-
+      You are Agent Smith, an AI project manager agent. 
+
+      {file:../fragments/workspace.md}
+model: qwen4b
+inferParams:
+    min_p: 0
+    top_k: 20
+    top_p: 0.95
+    temperature: 0.6
+    repetition_penalty: 1
+    chat_template_kwargs:
+      enable_thinking: true
+      preserve_thinking: true
+variables:
+  required:
+    workspace:
+      description: The local directory path where to operate
+toolsList:
+  - readfile
+  - shell
+  - python
+  - load-skill

package/dist/agents/agent-smith.yml

@@ -0,0 +1,42 @@
+description: An agent to manage Agent Smith's documentation
+category: "agent-smith"
+prompt: |-
+    {prompt}
+template:
+    system: |-
+      You are Agent Smith, an AI coordinator agent. Your job is to organise the workflow of other agents. As a coordinator you always delegate taks to agents to orchestrate the work. Your job is to coordinate the team of agents. Available agents:
+
+      - `agent-smith-assistant`: a fork of yourself, the default agent. Run it by default to accomplish any task.
+      - `agent-smith-search`: a web search and web page reader agent. Use to search information or for questions about some web pages.
+
+      {file:../fragments/workspace.md}
+      {file:../fragments/ctx-helper-files.md}
+
+      Your workflow to answer a user request:
+
+      - Analyze the user request
+      - Decompose the task to subtasks if necessary and plan the work to do
+      - Call agent(s) one after the other to accomplish the subtasks, use `agent-smith-assistant` by default if no specialized agent matches. Only call one agent at the time.
+model: qwen35b
+inferParams:
+    min_p: 0
+    top_k: 20
+    top_p: 0.85
+    temperature: 0.6
+    repetition_penalty: 1
+    presence_penalty: 1.5
+    chat_template_kwargs:
+      enable_thinking: true
+      preserve_thinking: true
+variables:
+  required:
+    workspace:
+      description: The local directory path where to operate
+toolsList:
+  - readfile
+  - edit-search-replace
+  - shell
+  - python
+  - run-agent
+  - load-skill
+  - notify-user

package/dist/agents/collaborator.yml

@@ -0,0 +1,16 @@
+description: A collaborator agent
+category: internal/experimental
+type: worker
+prompt: |-
+    You are on collaborator mode. Your role is: {role}. Main agent request:
+
+    {prompt}
+model: qwen35b
+variables:
+    required:
+        workspace:
+            description: The local directory path where to operate
+        role:
+            description: The role of the collaborator
+
+

package/dist/agents/infer.yml

@@ -0,0 +1,12 @@
+name: infer
+description: Run a raw inference query
+type: "agent-smith"
+prompt: |-
+    {prompt}
+model: qwen4b
+inferParams:
+  top_p: 0.95
+  top_k: 20
+  min_p: 0
+  temperature: 0.6
+  repeat_penalty: 1

package/dist/fragments/ctx-helper-files.md

@@ -0,0 +1,6 @@
+Project's context helper files:
+
+- `AGENTS.md`: general high level context map of the project. Read this first.
+- `.agents/documentation/project-nav.md`: a map of the project repositories and overview of the project
+
+To navigate the code always read local directories `.agents/documentation/codebase-summary.md` files

package/dist/fragments/workspace.md

@@ -0,0 +1,6 @@
+You are operating on a Linux filesystem in a workspace located at /workspace. The current directory is /. If a user refers to a path prepend /workspace to it for shell commands: example: ls /workspace/src if user refers to a `src` directory.
+
+Directories for AI agents:
+
+- `.agents/documentation`: where the documentation to navigate the codebase is
+- `.agents/tasks`: where we store tasks for AI agents. If the user refers to a task check this directory

package/dist/skills/create-project-docs/SKILL.md

@@ -0,0 +1,124 @@
+---
+name: create-project-docs
+description: use when asked to create or update helper documentation so AI agents can navigate a project
+---
+
+Create or update the documentation files that help AI agents navigate a codebase. Works for single-repo or multi-repo projects.
+
+## Files to Create/Update
+
+| File | Scope | Purpose |
+|------|-------|---------|
+| `AGENTS.md` | Project root | Index with reading order + links to other docs |
+| `.agents/documentation/codebase-summary.md` | Every repo/module | Standardized technical summary (7 sections) |
+| `.agents/documentation/project-nav.md` | Project root only | Comprehensive navigation map |
+
+> **Single-repo projects**: All 3 files live at the project root.
+> **Multi-repo projects**: `AGENTS.md` and `project-nav.md` at workspace root; each repo gets its own `AGENTS.md` and `codebase-summary.md`.
+
+## Workflow
+
+Execute steps in order:
+
+### 1. Explore the Project
+
+Walk the directory tree. Identify repos, packages/modules, entry points, dependencies, and key files. Read manifest or config files to understand structure and language.
+
+### 2. Create AGENTS.md (Project Root)
+
+```markdown
+# <Project Name>
+
+## Reading Order for AI Agents
+
+1. **Start here** → This file is your index
+2. **Comprehensive reference** → `.agents/documentation/project-nav.md` — Full project map
+3. **Technical summary** → `.agents/documentation/codebase-summary.md` — Dependencies, entry points, key files
+4. **Per-repo navigation** → Each repo has its own `AGENTS.md` and `.agents/documentation/codebase-summary.md`
+
+---
+
+This is a <language> project for <purpose>. It contains these repositories:
+
+- `<repo-name>` — <one-line description>
+- ...
+
+Documentation:
+
+- `.agents/documentation/codebase-summary.md`: top-level codebase summary
+- `.agents/documentation/project-nav.md`: navigation map
+- `<repo>/.agents/documentation/codebase-summary.md`: per-repo summaries
+```
+
+### 3. Create codebase-summary.md (Every Repo/Module)
+
+Use exactly these 7 sections in this order:
+
+```markdown
+# <module-name>
+
+## Summary
+One sentence: what the module does.
+
+## Dependencies
+- `<internal-dep>` — what it provides
+- External: `<library>` (<purpose>)
+
+## Used By
+- `<consumer>` — why it uses this module
+
+## Entry Point
+- `<path>` — one-line description
+
+## Key Files
+| File | Purpose |
+|------|---------|
+| `<path>` | One-line: what the file does |
+
+## Architecture
+- 2–4 bullets on design patterns and data flow.
+
+## Related
+- See `<module>` — how they work together
+
+## Documentation
+| Resource | Path |
+|----------|------|
+| Full docsite | `<path>` |
+```
+
+**Rules**: Information-dense, short. No project description (that's in project-nav.md). No architecture patterns duplicated from project-nav.md.
+
+### 4. Create project-nav.md (Project Root Only)
+
+Comprehensive map with these sections:
+
+1. **Project Overview** — What it does, repos/modules table, core capabilities
+2. **Architecture Principles** — Table: Principle | Detail | Key Files
+3. **Dependency Graph** — ASCII art showing module dependencies + prose explanation
+4. **Packages/Modules** — Per-module: Purpose, Key files, Key types/classes
+5. **Server** (if applicable) — Routes, key files, patterns
+6. **Plugins** (if applicable) — Categories, plugins table, key files
+7. **UI** (if applicable) — Components, services, themes, extensions
+8. **Apps** (if applicable) — Extensions, dependencies
+9. **Code Snippets** — Language-appropriate examples showing typical usage
+10. **Navigation Quick Reference** — Task → Path table for common development tasks
+11. **Documentation Links** — All doc resources with paths
+12. **Key Conventions & Patterns** — Naming, file structure, tool formats, etc.
+
+Add header note: `> **Primary reference** — Comprehensive navigation map for the project.`
+
+### 5. Cross-Reference and Verify
+
+- `AGENTS.md` → links to project-nav.md and codebase-summary.md
+- `codebase-summary.md` → Related section points to project-nav.md for overview
+- `project-nav.md` → no external cross-references needed (it's the primary source)
+- No duplicated content across files: project description, architecture patterns, quick-reference tables, and code snippets live ONLY in project-nav.md
+
+## Rules
+
+- **No redundancy**: Each piece of information lives in exactly one file
+- **codebase-summary.md preserves format**: The 7-section structure is standardized — do not change it
+- **Information-dense**: Keep files short. Use tables, bullets, one-line descriptions
+- **project-nav.md is canonical**: Single source of truth for project overview, architecture, snippets, and quick references
+- **Language-agnostic**: Adapt examples and conventions to the project's language and ecosystem

package/dist/skills/create-readme/SKILL.md

@@ -0,0 +1,103 @@
+---
+name: create-readme
+description: use this to create comprehensive README file for a given package or plugin or repository
+---
+
+## Workflow: README Creation
+
+### Phase 1: Discovery
+1. Read `/workspace/AGENTS.md` and `/workspace/.agents/documentation/project-nav.md` to understand project structure
+2. Read `/workspace/agent-smith/.agents/documentation/codebase-summary.md` and `/workspace/agent-smith/.agents/documentation/dpcumentation-map.md`
+3. Locate target in the codebase summary (note its position, dependencies, and purpose)
+
+### Phase 2: Deep Dive
+3. Navigate to directory: `/workspace/agent-smith/packages/{package-name}/` for a package, adapt the path
+4. Read `.agents/documentation/codebase-summary.md` for architecture, key files, and usage patterns
+5. Read documentation from `/workspace/agent-smith/docsite/public/doc/libraries/{package-name}/*.md` (get_started, usage, api)
+
+### Phase 3: Synthesis & Writing
+6. Create `README.md` with this structure:
+   - **Npm badge**: example:
+      ```
+      [![pub package](https://img.shields.io/npm/v/this-package-name)](https://www.npmjs.com/package/this-package-name)
+      ```
+   - **Title & Tagline**: Package name + one-sentence description. Mention that this package is part of the Agent Smith toolkit (repository: https://github.com/lynxai-team/agent-smith)
+   - **Features**: Bullet list of key capabilities (use emojis for visual hierarchy)
+   - **Documentation**: REQUIRED section with specific structure (see below)
+   - **Installation**: npm/yarn command in code block
+   - **Quick Start**: Minimal working example showing creation, init, and basic operations
+   - **Usage**: Detailed patterns with code examples (creation, initialization, verbose mode, reading/writing, error handling)
+   - **Complete Example**: Full working async function demonstrating all operations
+   - **API Reference**: Factory function signature + parameters table + interface + method summary table
+   - **Important Notes**: Browser-only warnings, limitations, related packages
+   - **Documentation Links**: References to full docsite
+   - **License**: MIT/appropriate license
+
+### Documentation Section Structure (REQUIRED)
+
+Every README must include a **Documentation** section with this exact structure: two subsections — "For AI Agents" and "For Humans" — providing appropriate links for each audience.
+
+#### Required Format
+
+```markdown
+## Documentation
+
+### For AI Agents
+- [Codebase Summary](.agents/documentation/codebase-summary.md) — Architecture, key files, and patterns for the {package-name} package
+- [Doc1](https://raw.githubusercontent.com/lynxai-team/agent-smith/refs/heads/main/docsite/public/doc/libraries/{package-name}/1.get_started.md) — Description
+- [Doc2](https://raw.githubusercontent.com/lynxai-team/agent-smith/refs/heads/main/docsite/public/doc/libraries/{package-name}/2.topic.md) — Description
+
+### For Humans
+- [Doc1](https://lynxai-team.github.io/agent-smith/libraries/{package-name}/) — Description
+- [Doc2](https://lynxai-team.github.io/agent-smith/libraries/{package-name}/topic) — Description
+```
+
+#### Rules
+
+1. **For AI Agents subsection**:
+   - Always start with a link to `.agents/documentation/codebase-summary.md` (relative path)
+   - List all relevant documentation files from `/workspace/agent-smith/docsite/public/doc/libraries/{package-name}/`
+   - Use raw GitHub URLs: `https://raw.githubusercontent.com/lynxai-team/agent-smith/refs/heads/main/docsite/public/doc/libraries/{package-name}/{filename}.md`
+   - Include a brief description after each link (after `—`)
+
+2. **For Humans subsection**:
+   - Use the docsite URL: `https://lynxai-team.github.io/agent-smith/libraries/{package-name}/`
+   - Convert paths: replace `docsite/public/doc/libraries/{package-name}/` with `libraries/{package-name}/`
+   - Remove file numbering prefixes (e.g., `1.get_started.md` → `/`)
+   - Include a brief description after each link (after `—`)
+
+#### Example from the types package README
+
+```markdown
+## Documentation
+
+### For AI Agents
+- [Codebase Summary](.agents/documentation/codebase-summary.md) — Architecture, key files, and patterns for the Agent Smith libraries
+- [Get Started](https://raw.githubusercontent.com/lynxai-team/agent-smith/refs/heads/main/docsite/public/doc/libraries/types/1.get_started.md) — Overview and installation
+- [Interfaces](https://raw.githubusercontent.com/lynxai-team/agent-smith/refs/heads/main/docsite/public/doc/libraries/types/2.interfaces.md) — Complete API reference of all exported types
+
+### For Humans
+- [Get Started](https://lynxai-team.github.io/agent-smith/libraries/types/) — Overview and usage guide
+- [Interfaces](https://lynxai-team.github.io/agent-smith/libraries/types/interfaces) — Full interface reference with tables
+```
+
+### Key Principles
+- **Information Density**: Every section must convey unique value; avoid repetition between Quick Start and Usage
+- **Code Examples**: All examples must be complete, runnable TypeScript with proper async/await
+- **Type Safety**: Show generic type parameters explicitly (e.g., `get<string>()`)
+- **Error Handling**: Document error cases explicitly with try/catch examples
+- **Browser vs Node**: Clearly state environment constraints early in the README
+- **Visual Hierarchy**: Use emojis for feature bullets, clear section headers, and tables for API reference
+
+### Output Validation
+Before finalizing, verify:
+- [ ] All code blocks are syntactically valid TypeScript
+- [ ] API signatures match actual implementation (check source files if needed)
+- [ ] Installation command uses correct package name (`@agent-smith/{package-name}`)
+- [ ] Documentation section follows the required structure with "For AI Agents" and "For Humans" subsections
+- [ ] Documentation links follow consistent pattern (raw GitHub URLs for AI agents, docsite URLs for humans)
+- [ ] No internal paths or implementation details leak into public documentation
+
+Documentation links base url: https://lynxai-team.github.io/agent-smith/
+
+Notify the user when the task is completed

package/dist/skills/create-task/SKILL.md

@@ -0,0 +1,55 @@
+---
+name: create-task
+description: entry point for creating tasks - analyzes request, chooses execution mode, creates task
+---
+
+# Create Task (Entry Point)
+
+This is the entry point skill for creating new tasks. It orchestrates the full workflow: analyze → decide mode → create task.
+
+## Workflow
+
+### Step 1: Analyze Request
+1. Read and understand the user's task request
+2. Identify scope, complexity, and requirements
+3. Note any special constraints or dependencies
+
+### Step 2: Choose Execution Mode
+1. Load the `task-mode` skill
+2. Evaluate the task against the decision checklist
+3. Determine whether to use `create-task-solo` (single agent) or `create-task-team` (multi-agent)
+4. Report your decision with reasoning
+
+### Step 3: Create Task
+1. Based on the mode decision, load either `create-task-solo` or `create-task-team` skill
+2. Execute that skill to create the task files
+3. Show the created structure to the user
+
+## Output Format
+
+After analysis and mode selection:
+
+```
+Task Analysis: [Brief summary of what needs to be done]
+Mode Decision: [create-task-solo | create-task-team]
+Reasoning: [Why this mode was chosen]
+Next Step: Creating task using [create-task-solo | create-task-team] skill...
+```
+
+After task creation:
+
+```
+Task Created: [.agents/tasks/task-name/]
+Files Created:
+- goals.md
+- plan.md
+- state.md
+- [additional files based on mode]
+```
+
+## Important Notes
+
+- Always use `task-mode` skill before creating a task
+- Never skip the mode selection step
+- Report decision to user before proceeding
+- The `create-task-solo` and `create-task-team` skills handle their own confirmation workflows

package/dist/skills/create-task-solo/SKILL.md

@@ -0,0 +1,146 @@
+---
+name: create-task-solo
+description: use this to create a new task with proper structure for single-agent (solo) execution
+---
+
+# Create Task Instructions (Single Agent)
+
+## Task Directory Structure
+
+Tasks are located in `.agents/tasks/[task-name]/`. Structure:
+
+```
+[task-name]/
+├── goals.md              # Goals and success criteria
+├── plan.md               # Full execution plan with all phases
+├── state.md              # Progress tracking
+└── notes.md              # Optional: additional context/notes
+```
+
+## Creating Task Files
+
+### 1. `goals.md` — Goals & Success Criteria
+
+```md
+# Goals and Success Criteria: [task-name]
+
+## Primary Goal
+[Clear statement of what the task achieves]
+
+## Success Criteria
+- [ ] Criterion 1
+- [ ] Criterion 2
+
+## Files Modified
+- `path/to/file` — description of changes
+
+## Exclusions
+[What is NOT included in this task]
+```
+
+### 2. `plan.md` — Full Execution Plan
+
+Contains all phases inline with step-by-step instructions. The agent retains context between phases, so steps can be concise.
+
+```md
+# Task Plan: [task-name]
+
+## Phase 1: [Phase Name]
+**Goal:** [What this phase achieves]
+
+### Step 1.1: [Step Name]
+- **File:** `path/to/file`
+- **Action:** [Detailed instructions on what to do]
+- **Verify:** [Verifiable success criteria]
+
+### Step 1.2: [Step Name]
+- **File:** `path/to/file`
+- **Action:** [Detailed instructions]
+- **Verify:** [Verifiable success criteria]
+
+## Phase 2: [Phase Name]
+**Goal:** [What this phase achieves]
+
+### Step 2.1: [Step Name]
+- **File:** `path/to/file`
+- **Action:** [Detailed instructions]
+- **Verify:** [Verifiable success criteria]
+```
+
+**Plan rules:**
+- Each phase has a clear goal
+- Each step specifies the file to modify, the action to take, and how to verify success
+- Success criteria must be verifiable (e.g., "test passes", "file contains X", "code compiles")
+- Steps should be atomic — one clear action per step
+
+### 3. `state.md` — Progress Tracking
+
+```md
+# Task State: [task-name]
+
+## Status: [planned | in-progress | completed]
+
+## Phases
+- [ ] Phase 1: [name]
+  - [ ] Step 1.1
+  - [ ] Step 1.2
+- [ ] Phase 2: [name]
+  - [ ] Step 2.1
+
+## Notes
+- Next: Phase X, Step Y
+```
+
+### 4. `notes.md` — Optional Context
+
+Use this file for any additional context, references, or information the agent should keep in mind while executing the task.
+
+---
+
+## Task Execution Workflow (Single Agent)
+
+When executing a task created with this skill, follow this workflow:
+
+### 1. Start/Resume Task
+1. Read `state.md` to determine current progress
+2. Identify the next step to execute
+3. If status is `completed`, notify user and exit
+
+### 2. Execute Steps
+For each step:
+1. Read the step instructions from `plan.md`
+2. Perform the action (edit files, run commands, etc.)
+3. Verify success criteria
+4. If verification fails, fix issues and re-verify
+5. Once verified, mark step as complete in `state.md`
+
+### 3. Update State
+After completing each phase:
+1. Mark all steps in that phase as `[x]` with `✓`
+2. Mark the phase as `[x]`
+3. Update "Next" note to point to the next phase/step
+4. If all phases complete, set status to `completed`
+
+### 4. Completion
+1. When all phases are done, notify user
+2. Ask user to verify the results
+3. Only delete the task directory if user confirms everything is correct
+
+---
+
+## Task Creation Workflow
+
+When creating a new task:
+
+1. **Analyze Request:** Understand what the user wants to accomplish
+2. **Define Goals:** Write `goals.md` with clear success criteria
+3. **Create Plan:** Write `plan.md` breaking work into phases and steps
+4. **Initialize State:** Write `state.md` with all items unchecked, status `planned`
+5. **Report to User:** Show the created task structure before execution
+6. **Wait for Confirmation:** Only execute if user approves the plan
+
+---
+
+## Choosing Execution Mode
+
+Use the `task-mode` skill to decide between single-agent (`create-task-solo`) and multi-agent (`create-task-team`) execution before creating a task. The `task-mode` skill provides a checklist and decision framework based on context isolation needs vs unified context requirements.