Skip to content

Enhance agent orchestrator with real tool execution and persistence #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
seanebones-lang merged 1 commit into from
Mar 17, 2026
Merged

Enhance agent orchestrator with real tool execution and persistence #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 25 additions & 0 deletions .cursor/config/agents-activation.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,31 @@
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP Beta Tester"
},
"refactoring": {
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP Refactoring"
},
"architecture": {
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP Architecture"
},
"debugging": {
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP Debugging"
},
"observability": {
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP Observability"
},
"cicd": {
"enabled": true,
"auto_trigger": true,
"prefix": "CTO Eleven MCP CI/CD"
}
},
"orchestrator": {
Expand Down
162 changes: 115 additions & 47 deletions src/lib/agent-loop.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,18 +14,21 @@
// Types
// ============================================================================

export type ToolName =
export type ToolName =
| 'read_file'
| 'write_file'
| 'list_files'
| 'delete_file'
| 'move_file'
| 'patch_file'
| 'run_command'
| 'search_code'
| 'create_branch'
| 'create_pull_request'
| 'get_diff'
| 'get_commit_history'
| 'web_search'
| 'web_browse'
| 'think'
| 'complete'

Expand Down Expand Up @@ -154,17 +157,18 @@ Create a new GitHub repository. **Enables full project spawning from description
Returns: Repository object with owner, name, fullName, url, defaultBranch.

### 6. run_command
Execute a terminal command.
Execute a terminal command. Supports piped commands, redirects, and shell expressions.
\`\`\`json
{
"name": "run_command",
"arguments": {
"command": "string (required) - Command to execute",
"cwd": "string (optional) - Working directory"
"command": "string (required) - Command to execute (supports pipes, &&, ||)",
"cwd": "string (optional) - Working directory",
"timeout": "number (optional) - Timeout in ms (default: 30000, max: 300000)"
}
}
\`\`\`
Allowed commands: npm, npx, yarn, node, git, tsc, eslint, prettier, jest, python, pip, cargo, go
Allowed commands: npm, npx, yarn, node, git, tsc, eslint, prettier, jest, vitest, python, pip, cargo, go, make, docker, curl

**Python Libraries Available** (import directly in code_execution):
- Data: pandas, numpy, json, csv, PyMuPDF (fitz), Pillow, PyYAML
Expand All @@ -174,7 +178,7 @@ Allowed commands: npm, npx, yarn, node, git, tsc, eslint, prettier, jest, python
- Utility: asyncio, concurrent.futures, pathlib, logging, hashlib, secrets
- See full catalog: src/lib/agent-tools-catalog.ts

### 6. move_file
### 7. move_file
Move or rename a file. **Automatically commits to GitHub** when a repository is connected.
\`\`\`json
{
Expand All @@ -186,20 +190,35 @@ Move or rename a file. **Automatically commits to GitHub** when a repository is
}
\`\`\`

### 7. search_code
Advanced code search using GitHub Code Search API (more powerful than regex).
### 8. patch_file
Apply surgical edits to a file without rewriting the entire content. More token-efficient than write_file for small changes.
\`\`\`json
{
"name": "patch_file",
"arguments": {
"path": "string (required) - Path to the file to patch",
"patches": "array (required) - Array of patch objects: [{ \"find\": \"old text\", \"replace\": \"new text\" }]"
}
}
\`\`\`
Each patch replaces the first occurrence of \`find\` with \`replace\`. Patches are applied sequentially.

### 9. search_code
Advanced code search. Uses GitHub Code Search API for repos, or local ripgrep-style search.
\`\`\`json
{
"name": "search_code",
"arguments": {
"query": "string (required) - Search query (supports GitHub search syntax)",
"query": "string (required) - Search query (supports regex patterns)",
"language": "string (optional) - Filter by language (e.g., 'typescript', 'javascript')",
"path": "string (optional) - Filter by path"
"path": "string (optional) - Filter by path",
"context_lines": "number (optional) - Lines of context around matches (default: 0)",
"exclude": "string (optional) - Glob pattern to exclude (default: node_modules)"
}
}
\`\`\`

### 8. create_branch
### 10. create_branch
Create a new Git branch.
\`\`\`json
{
Expand All @@ -211,7 +230,7 @@ Create a new Git branch.
}
\`\`\`

### 9. create_pull_request
### 11. create_pull_request
Create a pull request.
\`\`\`json
{
Expand All @@ -225,7 +244,7 @@ Create a pull request.
}
\`\`\`

### 10. get_diff
### 12. get_diff
Get differences between commits or branches.
\`\`\`json
{
Expand All @@ -238,7 +257,7 @@ Get differences between commits or branches.
}
\`\`\`

### 11. get_commit_history
### 13. get_commit_history
Get commit history for a branch or specific file.
\`\`\`json
{
Expand All @@ -251,7 +270,7 @@ Get commit history for a branch or specific file.
}
\`\`\`

### 12. web_search
### 14. web_search
Search the web for information using DuckDuckGo (free, no API key required).
\`\`\`json
{
Expand All @@ -264,7 +283,7 @@ Search the web for information using DuckDuckGo (free, no API key required).
\`\`\`
Returns: Formatted search results with titles, URLs, and snippets.

### 13. web_browse
### 15. web_browse
Fetch and extract readable content from a web page.
\`\`\`json
{
Expand All @@ -276,7 +295,7 @@ Fetch and extract readable content from a web page.
\`\`\`
Returns: Extracted text content from the web page (up to 10KB).

### 14. think
### 16. think
Record your reasoning process (doesn't execute anything).
\`\`\`json
{
Expand All @@ -287,7 +306,7 @@ Record your reasoning process (doesn't execute anything).
}
\`\`\`

### 15. complete
### 17. complete
Mark the task as complete.
\`\`\`json
{
Expand All @@ -301,10 +320,22 @@ Mark the task as complete.

## Response Format

For each step, respond with:
For each step, you can respond with one or more tool calls.

**Single tool call:**
1. **Thought**: Your reasoning about what to do next
2. **Action**: A tool call in JSON format

**Parallel tool calls** (for independent operations like reading multiple files):
1. **Thought**: Your reasoning about what to do next
2. **Actions**: Multiple tool calls in a JSON array format
\`\`\`json
[
{ "name": "read_file", "arguments": { "path": "file1.ts" } },
{ "name": "read_file", "arguments": { "path": "file2.ts" } }
]
\`\`\`

Example:
\`\`\`
**Thought**: I need to first read the package.json to understand the project structure.
Expand All @@ -322,12 +353,13 @@ Example:

## Important Rules

1. **One action at a time**: Only call one tool per response
2. **Wait for results**: After each action, wait for the observation before proceeding
3. **Handle errors**: If a command fails, analyze the error and try to fix it
1. **Parallel when possible**: You may call multiple tools in a single response when they are independent (e.g., reading multiple files). Use a JSON array of tool calls.
2. **Sequential when dependent**: If one tool's output feeds into another, wait for results before proceeding.
3. **Handle errors**: If a command fails, analyze the error and try to fix it. Transient failures will be retried automatically.
4. **Be thorough**: Read relevant files before making changes
5. **Test your work**: Run tests or build commands to verify changes
6. **Complete the task**: Always call 'complete' when finished
5. **Use patch_file for small edits**: Prefer patch_file over write_file when making targeted changes to existing files.
6. **Test your work**: Run tests or build commands to verify changes
7. **Complete the task**: Always call 'complete' when finished
`

// ============================================================================
Expand Down Expand Up @@ -365,39 +397,75 @@ Remember: You are autonomous. Take initiative, make decisions, and complete the
}

// ============================================================================
// Tool Call Parser
// Tool Call Parser (supports single and parallel tool calls)
// ============================================================================

export function parseToolCall(response: string): ToolCall | null {
// Look for JSON in the response
const jsonMatch = response.match(/```json\s*([\s\S]*?)\s*```/)
if (!jsonMatch) {
// Try to find raw JSON
const rawJsonMatch = response.match(/\{[\s\S]*?"name"[\s\S]*?"arguments"[\s\S]*?\}/)
if (!rawJsonMatch) return null

/**
* Parse a single tool call from a parsed JSON object
*/
function toolCallFromParsed(parsed: Record<string, unknown>): ToolCall | null {
if (parsed && typeof parsed.name === 'string' && parsed.arguments && typeof parsed.arguments === 'object') {
return {
id: crypto.randomUUID(),
name: parsed.name as ToolName,
arguments: parsed.arguments as Record<string, unknown>,
}
}
return null
}

/**
* Parse one or more tool calls from agent response text.
* Supports:
* - Single tool call: { "name": "...", "arguments": {...} }
* - Parallel tool calls: [{ "name": "..." }, { "name": "..." }]
*/
export function parseToolCalls(response: string): ToolCall[] {
const results: ToolCall[] = []

// Try all JSON code blocks
const codeBlocks = [...response.matchAll(/```(?:json)?\s*([\s\S]*?)\s*```/g)]
for (const match of codeBlocks) {
try {
const parsed = JSON.parse(rawJsonMatch[0])
return {
id: crypto.randomUUID(),
name: parsed.name,
arguments: parsed.arguments || {},
const parsed = JSON.parse(match[1].trim())
// Array of tool calls (parallel)
if (Array.isArray(parsed)) {
for (const item of parsed) {
const tc = toolCallFromParsed(item)
if (tc) results.push(tc)
}
if (results.length > 0) return results
}
// Single tool call
const tc = toolCallFromParsed(parsed)
if (tc) results.push(tc)
if (results.length > 0) return results
} catch {
return null
// not valid JSON, try next block
}
}

try {
const parsed = JSON.parse(jsonMatch[1])
return {
id: crypto.randomUUID(),
name: parsed.name,
arguments: parsed.arguments || {},
// Fallback: try to find raw JSON (single tool call)
const rawJsonMatch = response.match(/\{[\s\S]*?"name"[\s\S]*?"arguments"[\s\S]*?\}/)
if (rawJsonMatch) {
try {
const parsed = JSON.parse(rawJsonMatch[0])
const tc = toolCallFromParsed(parsed)
if (tc) results.push(tc)
} catch {
// ignore
}
} catch {
return null
}

return results
}

/**
* @deprecated Use parseToolCalls() instead for parallel support. Kept for backward compatibility.
*/
export function parseToolCall(response: string): ToolCall | null {
const calls = parseToolCalls(response)
return calls.length > 0 ? calls[0] : null
}

// ============================================================================
Expand Down
Loading