gb/AGENTS.md

General Bots AI Agent Guidelines

⚠️ CRITICAL SECURITY WARNING

NEVER CREATE FILES WITH SECRETS IN THE REPOSITORY ROOT

Secret files MUST be placed in /tmp/ only:

• ✅ /tmp/vault-token-gb - Vault root token
• ✅ /tmp/vault-unseal-key-gb - Vault unseal key
• ❌ vault-unseal-keys - FORBIDDEN (tracked by git)
• ❌ start-and-unseal.sh - FORBIDDEN (contains secrets)

Why /tmp/?

• Cleared on reboot (ephemeral)
• Not tracked by git
• Standard Unix security practice
• Prevents accidental commits

---

🧭 LLM Navigation Guide

Reading This Workspace

For LLMs analyzing this codebase:

1. Start with Component Dependency Graph in README to understand relationships
2. Review Module Responsibility Matrix for what each module does
3. Study Data Flow Patterns to understand execution flow
4. Reference Common Architectural Patterns before making changes
5. Check Security Rules below - violations are blocking issues
6. Follow Code Patterns below - consistency is mandatory

---

🔐 Security Directives - MANDATORY

1. Error Handling - NO PANICS IN PRODUCTION

// ❌ FORBIDDEN
value.unwrap()
value.expect("message")
panic!("error")
todo!()
unimplemented!()

// ✅ REQUIRED
value?
value.ok_or_else(|| Error::NotFound)?
value.unwrap_or_default()
value.unwrap_or_else(|e| { log::error!("{}", e); default })
if let Some(v) = value { ... }
match value { Ok(v) => v, Err(e) => return Err(e.into()) }

2. Command Execution - USE SafeCommand

// ❌ FORBIDDEN
Command::new("some_command").arg(user_input).output()

// ✅ REQUIRED
use crate::security::command_guard::SafeCommand;
SafeCommand::new("allowed_command")?
    .arg("safe_arg")?
    .execute()

3. Error Responses - USE ErrorSanitizer

// ❌ FORBIDDEN
Json(json!({ "error": e.to_string() }))
format!("Database error: {}", e)

// ✅ REQUIRED
use crate::security::error_sanitizer::log_and_sanitize;
let sanitized = log_and_sanitize(&e, "context", None);
(StatusCode::INTERNAL_SERVER_ERROR, sanitized)

4. SQL - USE sql_guard

// ❌ FORBIDDEN
format!("SELECT * FROM {}", user_table)

// ✅ REQUIRED
use crate::security::sql_guard::{sanitize_identifier, validate_table_name};
let safe_table = sanitize_identifier(&user_table);
validate_table_name(&safe_table)?;

---

✅ Mandatory Code Patterns

Use Self in Impl Blocks

impl MyStruct {
    fn new() -> Self { Self { } }  // ✅ Not MyStruct
}

Derive Eq with PartialEq

#[derive(PartialEq, Eq)]  // ✅ Always both
struct MyStruct { }

Inline Format Args

format!("Hello {name}")  // ✅ Not format!("{}", name)

Combine Match Arms

match x {
    A | B => do_thing(),  // ✅ Combine identical arms
    C => other(),
}

---

❌ Absolute Prohibitions

• ❌ NEVER use .unwrap() or .expect() in production code (tests OK)
• ❌ NEVER use panic!(), todo!(), unimplemented!()
• ❌ NEVER use Command::new() directly - use SafeCommand
• ❌ NEVER return raw error strings to HTTP clients
• ❌ NEVER use #[allow()] in source code - FIX the code instead
• ❌ NEVER add lint exceptions to Cargo.toml - FIX the code instead
• ❌ NEVER use _ prefix for unused variables - DELETE or USE them
• ❌ NEVER leave unused imports or dead code
• ❌ NEVER use CDN links - all assets must be local
• ❌ NEVER use cargo clean - causes 30min rebuilds, use ./reset.sh for database issues
• ❌ NEVER create .md documentation files without checking botbook/ first

---

📏 File Size Limits - MANDATORY

Maximum 450 Lines Per File

When a file grows beyond this limit:

1. Identify logical groups - Find related functions
2. Create subdirectory module - e.g., handlers/
3. Split by responsibility:
   • types.rs - Structs, enums, type definitions
   • handlers.rs - HTTP handlers and routes
   • operations.rs - Core business logic
   • utils.rs - Helper functions
   • mod.rs - Re-exports and configuration
4. Keep files focused - Single responsibility
5. Update mod.rs - Re-export all public items

NEVER let a single file exceed 450 lines - split proactively at 350 lines

---

🔥 Error Fixing Workflow

Mode 1: OFFLINE Batch Fix (PREFERRED)

When given error output:

1. Read ENTIRE error list first
2. Group errors by file
3. For EACH file with errors: a. View file → understand context b. Fix ALL errors in that file c. Write once with all fixes
4. Move to next file
5. REPEAT until ALL errors addressed
6. ONLY THEN → verify with build/diagnostics

NEVER run cargo build/check/clippy DURING fixing Fix ALL errors OFFLINE first, verify ONCE at the end

Mode 2: Interactive Loop

LOOP UNTIL (0 warnings AND 0 errors):
  1. Run diagnostics → pick file with issues
  2. Read entire file
  3. Fix ALL issues in that file
  4. Write file once with all fixes
  5. Verify with diagnostics
  6. CONTINUE LOOP
END LOOP

---

🎭 Playwright Browser Testing - YOLO Mode

When user requests to start YOLO mode with Playwright:

1. Start the browser - Use mcp__playwright__browser_navigate to open http://localhost:3000
2. Take snapshot - Use mcp__playwright__browser_snapshot to see current page state
3. Test user flows - Use click, type, fill_form, etc.
4. Verify results - Check for expected content, errors in console, network requests
5. Validate backend - Check database and services to confirm process completion
6. Report findings - Always include screenshot evidence with browser_take_screenshot

Bot-Specific Testing URL Pattern: http://localhost:3000/<botname>

Backend Validation Checks: After UI interactions, validate backend state via psql or tail logs.

---

🧪 Testing Strategy

Unit Tests

• Location: Each crate has tests/ directory or inline #[cfg(test)] modules
• Naming: Test functions use test_ prefix or describe what they test
• Running: cargo test -p <crate_name> or cargo test for all

Integration Tests

• Location: bottest/ crate contains integration tests
• Scope: Tests full workflows across multiple crates
• Running: cargo test -p bottest

Coverage Goals

• Critical paths: 80%+ coverage required
• Error handling: ALL error paths must have tests
• Security: All security guards must have tests

---

🐛 Debugging Rules

🚨 CRITICAL ERROR HANDLING RULE

STOP EVERYTHING WHEN ERRORS APPEAR

When ANY error appears in logs during startup or operation:

1. IMMEDIATELY STOP - Do not continue with other tasks
2. IDENTIFY THE ERROR - Read the full error message and context
3. FIX THE ERROR - Address the root cause, not symptoms
4. VERIFY THE FIX - Ensure error is completely resolved
5. ONLY THEN CONTINUE - Never ignore or work around errors

NEVER restart servers to "fix" errors - FIX THE ACTUAL PROBLEM

Log Locations

Component	Log File	What's Logged
botserver	botserver.log	API requests, errors, script execution, client navigation events
botui	botui.log	UI rendering, WebSocket connections
drive_monitor	In botserver logs with [drive_monitor] prefix	File sync, compilation
client errors	In botserver logs with CLIENT: prefix	JavaScript errors, navigation events

---

🎨 Frontend Standards

HTMX-First Approach

• Use HTMX to minimize JavaScript
• Server returns HTML fragments, not JSON
• Use hx-get, hx-post, hx-target, hx-swap
• WebSocket via htmx-ws extension

Local Assets Only - NO CDN

<!-- ✅ CORRECT -->
<script src="js/vendor/htmx.min.js"></script>

<!-- ❌ WRONG -->
<script src="https://unpkg.com/htmx.org@1.9.10"></script>

---

🚀 Performance & Size Standards

Binary Size Optimization

• Release Profile: Always maintain opt-level = "z", lto = true, codegen-units = 1, strip = true, panic = "abort".
• Dependencies:
  • Run cargo tree --duplicates weekly
  • Run cargo machete to remove unused dependencies
  • Use default-features = false and explicitly opt-in to needed features

Linting & Code Quality

• Clippy: Code MUST pass cargo clippy --all-targets --all-features with 0 warnings.
• No Allow: Do not use #[allow(clippy::...)] unless absolutely necessary and documented.

---

🔑 Memory & Main Directives

LOOP AND COMPACT UNTIL 0 WARNINGS - MAXIMUM PRECISION

• 0 warnings
• 0 errors
• Trust project diagnostics
• Respect all rules
• No #[allow()] in source code
• Real code fixes only

Remember:

• OFFLINE FIRST - Fix all errors from list before compiling
• BATCH BY FILE - Fix ALL errors in a file at once
• WRITE ONCE - Single edit per file with all fixes
• VERIFY LAST - Only compile/diagnostics after ALL fixes
• DELETE DEAD CODE - Don't keep unused code around
• GIT WORKFLOW - ALWAYS push to ALL repositories (github, pragmatismo)