// SPF Smart Gateway - Configuration // Copyright 2026 Joseph Stone - All Rights Reserved // // Loads SPF rules, tiers, formulas, blocked paths. Defaults stored in LMDB. // // BLOCK I ADDITION: TransformerConfig struct (after MeshConfig pattern) // BLOCK NN-A: Added .claude.json, .mcp.json, .config/anthropic/, .anthropic/ to blocked_paths // BLOCK SEC-2: Added .qwen/, .xlaude/ agent config dirs + LIVE/CONFIG/ directory protection use serde::{Deserialize, Serialize, Deserializer}; use std::path::Path; /// Master SPF configuration loaded from CONFIG LMDB #[derive(Debug, Clone, Serialize, Deserialize)] pub struct SpfConfig { pub version: String, pub enforce_mode: EnforceMode, pub allowed_paths: Vec, pub blocked_paths: Vec, pub require_read_before_edit: bool, pub max_write_size: usize, pub tiers: TierConfig, pub formula: FormulaConfig, pub complexity_weights: ComplexityWeights, pub dangerous_commands: Vec, pub git_force_patterns: Vec, // ================================================================ // COMMAND WHITELIST FIELDS — Default-Deny Bash Security (BLOCK-01) // Empty defaults = everything blocked until configured. // Populated from LMDB commands DB by load_full_config() (BLOCK-02). // Enforced by Stage 0 in validate_bash() (BLOCK-03). // ================================================================ #[serde(default)] pub allowed_commands_user: std::collections::HashMap, #[serde(default)] pub allowed_commands_sandbox: std::collections::HashMap, #[serde(default)] pub user_fs_paths: Vec, } #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)] #[serde(rename_all = "lowercase")] pub enum EnforceMode { Soft, Max, } #[derive(Debug, Clone, Serialize, Deserialize)] pub struct TierConfig { pub simple: TierThreshold, pub light: TierThreshold, pub medium: TierThreshold, pub critical: TierThreshold, } #[derive(Debug, Clone, Serialize, Deserialize)] pub struct TierThreshold { pub max_c: u64, pub analyze_percent: u8, pub build_percent: u8, pub requires_approval: bool, } #[derive(Debug, Clone, Serialize, Deserialize)] pub struct FormulaConfig { /// W_eff: effective working memory in tokens pub w_eff: f64, /// Euler's number pub e: f64, /// C = (basic ^ basic_power) + (deps ^ deps_power) + (complex ^ complex_power) + (files * files_mult) pub basic_power: u32, pub deps_power: u32, pub complex_power: u32, pub files_multiplier: u64, } #[derive(Debug, Clone, Serialize, Deserialize)] pub struct ComplexityWeights { pub edit: ToolWeight, pub write: ToolWeight, pub bash_dangerous: ToolWeight, pub bash_git: ToolWeight, pub bash_piped: ToolWeight, pub bash_simple: ToolWeight, pub read: ToolWeight, pub search: ToolWeight, pub unknown: ToolWeight, } #[derive(Debug, Clone, Serialize, Deserialize)] pub struct ToolWeight { pub basic: u64, pub dependencies: u64, pub complex: u64, pub files: u64, } // ============================================================================ // COMMAND PERMISSION MODEL — Default-Deny Bash Security (BLOCK-01) // Per-command R/W/X flags for whitelist enforcement. // Stored in LMDB commands DB (BLOCK-02), checked by Stage 0 (BLOCK-03). // ============================================================================ /// Per-command permission flags for whitelist enforcement. /// Controls what operations a whitelisted command can perform. #[derive(Debug, Clone, Copy, Serialize, Deserialize)] pub struct CommandPerm { pub read: bool, // Can read files, list dirs, query info pub write: bool, // Can modify, create, delete files pub execute: bool, // Can spawn subprocesses (-exec, system()) } impl CommandPerm { pub fn read_only() -> Self { Self { read: true, write: false, execute: false } } pub fn read_write() -> Self { Self { read: true, write: true, execute: false } } pub fn full() -> Self { Self { read: true, write: true, execute: true } } } impl Default for SpfConfig { fn default() -> Self { Self { version: "1.0.0".to_string(), enforce_mode: EnforceMode::Max, allowed_paths: { let home = crate::paths::actual_home().to_string_lossy(); vec![ format!("{}/", home), ] }, blocked_paths: { let root = crate::paths::spf_root().to_string_lossy(); let home = crate::paths::actual_home().to_string_lossy(); let mut paths = vec![ crate::paths::system_pkg_path(), format!("{}/src/", root), format!("{}/LIVE/SPF_FS/blobs/", root), format!("{}/Cargo.toml", root), format!("{}/Cargo.lock", root), format!("{}/.claude/", home), // BLOCK NN-A: Claude/Anthropic config files — ZERO AI write access format!("{}/.claude.json", home), format!("{}/.mcp.json", home), format!("{}/.config/anthropic/", home), format!("{}/.anthropic/", home), // BLOCK SEC-2: External agent config directories — prevent config hijack // Blocks both home-level and SPFsmartGATE-level agent configs. // Prevents any AI agent from creating/modifying CLI routing files // that could redirect MCP tool calls to a different gate instance. format!("{}/.qwen/", home), format!("{}/.qwen/", root), format!("{}/.xlaude/", home), format!("{}/.xlaude/", root), // BLOCK SEC-2: Runtime config directory — protect gate behavior configs // Contains http.json, mesh.json, transformer.json, network.json, // whitelist-config.json — all control gate pipeline behavior. format!("{}/LIVE/CONFIG/", root), // System config and state — ZERO AI write access format!("{}/LIVE/CONFIG.DB", root), format!("{}/LIVE/LMDB5/", root), format!("{}/LIVE/state/", root), format!("{}/LIVE/storage/", root), format!("{}/hooks/", root), format!("{}/scripts/", root), ]; if cfg!(target_os = "windows") { paths.extend([ r"C:\Windows".to_string(), r"C:\Program Files".to_string(), r"C:\Program Files (x86)".to_string(), ]); } else { paths.extend([ "/tmp".to_string(), "/etc".to_string(), "/usr".to_string(), "/system".to_string(), ]); } paths }, require_read_before_edit: true, max_write_size: 100_000, tiers: TierConfig { simple: TierThreshold { max_c: 500, analyze_percent: 40, build_percent: 60, requires_approval: true }, light: TierThreshold { max_c: 2000, analyze_percent: 60, build_percent: 40, requires_approval: true }, medium: TierThreshold { max_c: 10000, analyze_percent: 75, build_percent: 25, requires_approval: true }, critical: TierThreshold { max_c: u64::MAX, analyze_percent: 95, build_percent: 5, requires_approval: true }, }, formula: FormulaConfig { w_eff: 40000.0, e: std::f64::consts::E, basic_power: 1, // ^1 per SPF protocol deps_power: 7, // ^7 per SPF protocol complex_power: 10, // ^10 per SPF protocol files_multiplier: 10, // ×10 per SPF protocol }, // Weights scaled for formula: C = basic^1 + deps^7 + complex^10 + files×10 // deps^7: 2→128, 3→2187, 4→16384, 5→78125 // complex^10: 1→1, 2→1024 complexity_weights: ComplexityWeights { edit: ToolWeight { basic: 10, dependencies: 2, complex: 1, files: 1 }, write: ToolWeight { basic: 20, dependencies: 2, complex: 1, files: 1 }, bash_dangerous: ToolWeight { basic: 50, dependencies: 5, complex: 2, files: 1 }, bash_git: ToolWeight { basic: 30, dependencies: 3, complex: 1, files: 1 }, bash_piped: ToolWeight { basic: 20, dependencies: 3, complex: 1, files: 1 }, bash_simple: ToolWeight { basic: 10, dependencies: 1, complex: 0, files: 1 }, read: ToolWeight { basic: 5, dependencies: 1, complex: 0, files: 1 }, search: ToolWeight { basic: 8, dependencies: 2, complex: 0, files: 1 }, unknown: ToolWeight { basic: 20, dependencies: 3, complex: 1, files: 1 }, }, dangerous_commands: vec![ "rm -rf /".to_string(), "rm -rf ~".to_string(), "dd if=".to_string(), "> /dev/".to_string(), "chmod 777".to_string(), "curl | sh".to_string(), "wget | sh".to_string(), "curl|sh".to_string(), "wget|sh".to_string(), ], git_force_patterns: vec![ "--force".to_string(), "--hard".to_string(), "-f".to_string(), ], // COMMAND WHITELIST DEFAULTS — EMPTY = DEFAULT-DENY (BLOCK-01) allowed_commands_user: std::collections::HashMap::new(), allowed_commands_sandbox: std::collections::HashMap::new(), user_fs_paths: vec![], } } } impl SpfConfig { /// Load config from JSON file, falling back to defaults pub fn load(path: &Path) -> anyhow::Result { if path.exists() { let content = std::fs::read_to_string(path)?; let config: Self = serde_json::from_str(&content)?; Ok(config) } else { log::warn!("Config not found at {:?}, using defaults", path); Ok(Self::default()) } } /// Save config to JSON file pub fn save(&self, path: &Path) -> anyhow::Result<()> { let content = serde_json::to_string_pretty(self)?; std::fs::write(path, content)?; Ok(()) } /// Get tier for a given complexity value /// CRITICAL tier requires explicit user approval. Lower tiers protected by other layers. pub fn get_tier(&self, c: u64) -> (&str, u8, u8, bool) { if c < self.tiers.simple.max_c { ("SIMPLE", self.tiers.simple.analyze_percent, self.tiers.simple.build_percent, self.tiers.simple.requires_approval) } else if c < self.tiers.light.max_c { ("LIGHT", self.tiers.light.analyze_percent, self.tiers.light.build_percent, self.tiers.light.requires_approval) } else if c < self.tiers.medium.max_c { ("MEDIUM", self.tiers.medium.analyze_percent, self.tiers.medium.build_percent, self.tiers.medium.requires_approval) } else { ("CRITICAL", self.tiers.critical.analyze_percent, self.tiers.critical.build_percent, self.tiers.critical.requires_approval) } } /// Check if a path is blocked (with canonicalization to prevent traversal bypass) /// FAIL-CLOSED: if canonicalize fails, try parent dir (for new files). /// If parent also fails, assume blocked. pub fn is_path_blocked(&self, path: &str) -> bool { let canonical = match std::fs::canonicalize(path) { Ok(p) => p.to_string_lossy().to_string(), Err(_) => { if path.contains("..") { return true; // Traversal in unresolvable path = always blocked } // Path doesn't exist yet — try resolving parent directory. // Handles new file creation in valid directories (e.g. write to sandbox). match std::path::Path::new(path).parent() { Some(parent) => match std::fs::canonicalize(parent) { Ok(p) => { let filename = std::path::Path::new(path) .file_name() .map(|f| f.to_string_lossy().to_string()) .unwrap_or_default(); format!("{}/{}", p.to_string_lossy(), filename) } Err(_) => return true, // Parent unresolvable = blocked (fail-closed) }, None => return true, // No parent = blocked (fail-closed) } } }; self.blocked_paths.iter().any(|blocked| canonical.starts_with(blocked)) } /// Check if a path is allowed (with canonicalization to prevent traversal bypass) /// FAIL-CLOSED: if canonicalize fails, path is NOT allowed. Period. /// Cannot confirm path is in an allowed location = deny. pub fn is_path_allowed(&self, path: &str) -> bool { let canonical = match std::fs::canonicalize(path) { Ok(p) => p.to_string_lossy().to_string(), Err(_) => { // FAIL-CLOSED: unresolvable path is never confirmed allowed. // This covers: traversal attacks, symlink tricks, non-existent // paths, encoded variants, null bytes — all denied. // For new file creation: validate_write uses is_write_allowed() // (hardcoded sandbox) as primary check, not this function. return false; } }; self.allowed_paths.iter().any(|allowed| canonical.starts_with(allowed)) } } // ============================================================================ // HTTP API CONFIGURATION // ============================================================================ /// HTTP transport configuration — loaded from LIVE/CONFIG/http.json #[derive(Debug, Clone, Serialize, Deserialize)] pub struct HttpConfig { pub transport: String, pub port: u16, pub bind: String, pub tls_enabled: bool, pub tls_cert: String, pub tls_key: String, pub auth_mode: String, pub api_key: String, } impl Default for HttpConfig { fn default() -> Self { Self { transport: "both".to_string(), port: 3900, // SEC-4: Default to localhost — external access requires explicit config. // Prevents accidental exposure on public interfaces. bind: "127.0.0.1".to_string(), tls_enabled: true, tls_cert: "tls/cert.pem".to_string(), tls_key: "tls/key.pem".to_string(), auth_mode: "both".to_string(), api_key: String::new(), } } } impl HttpConfig { /// Load HTTP config from JSON file, falling back to defaults pub fn load(path: &Path) -> anyhow::Result { if path.exists() { let content = std::fs::read_to_string(path)?; let config: Self = serde_json::from_str(&content)?; Ok(config) } else { log::warn!("HTTP config not found at {:?}, using defaults", path); Ok(Self::default()) } } } // ============================================================================ // MESH CONFIGURATION — Agent identity, role, team, discovery // ============================================================================ /// Dual-mode role — parsed from mesh.json "role" field. /// Shared between config.rs (MeshConfig), identity.rs (PeerInfo), and mesh.rs (PeerChannel). #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Hash)] pub enum AgentRole { /// Human-facing director — manages thinkers and workers #[serde(rename = "orchestrator")] Orchestrator, /// Co-problem-solver — shares live context with other thinkers #[serde(rename = "thinker")] Thinker, /// Pure executor — receives tasks, runs them, returns results #[serde(rename = "worker")] Worker, } impl<'de> Deserialize<'de> for AgentRole { fn deserialize(deserializer: D) -> Result where D: Deserializer<'de>, { let s = String::deserialize(deserializer)?; match s.to_lowercase().as_str() { "orchestrator" | "coordinator" | "agent" | "director" => Ok(AgentRole::Orchestrator), "thinker" | "problem-solver" => Ok(AgentRole::Thinker), "worker" | "executor" => Ok(AgentRole::Worker), _ => Ok(AgentRole::Orchestrator), // Unknown role defaults to Orchestrator } } } impl std::fmt::Display for AgentRole { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { match self { AgentRole::Orchestrator => write!(f, "orchestrator"), AgentRole::Thinker => write!(f, "thinker"), AgentRole::Worker => write!(f, "worker"), } } } impl Default for AgentRole { fn default() -> Self { Self::Orchestrator } } /// Mesh transport configuration — loaded from LIVE/CONFIG/mesh.json #[derive(Debug, Clone, Serialize, Deserialize)] pub struct MeshConfig { /// Enable mesh networking pub enabled: bool, /// Agent's role in the mesh team pub role: AgentRole, /// Team name this agent belongs to pub team: String, /// Agent display name (human-readable) pub name: String, /// Capabilities this agent exposes to mesh peers pub capabilities: Vec, /// Discovery mode: "auto" (mDNS + DHT), "local" (mDNS only), "manual" (groups only) pub discovery: String, /// ALPN protocol identifier pub alpn: String, /// QUIC bind port (0 = random, >0 = fixed — needed for peer JSON addr configs) #[serde(default)] pub port: u16, } impl Default for MeshConfig { fn default() -> Self { Self { enabled: true, role: AgentRole::Orchestrator, // legacy "agent" maps here team: "default".to_string(), name: String::new(), capabilities: vec!["tools".to_string()], discovery: "auto".to_string(), alpn: "/spf/mesh/1".to_string(), port: 0, } } } impl MeshConfig { /// Load mesh config from JSON file, falling back to defaults pub fn load(path: &Path) -> anyhow::Result { if path.exists() { let content = std::fs::read_to_string(path)?; let config: Self = serde_json::from_str(&content)?; Ok(config) } else { Ok(Self::default()) } } } // ============================================================================ // BLOCK I — TRANSFORMER CONFIGURATION // Follows MeshConfig pattern: struct + Default + load() // Loaded from LIVE/CONFIG/transformer.json // ============================================================================ /// Transformer runtime configuration — loaded from LIVE/CONFIG/transformer.json #[derive(Debug, Clone, Serialize, Deserialize)] pub struct TransformerConfig { /// Enable transformer runtime pub enabled: bool, /// Path to writer model checkpoint (relative to LIVE/MODELS/) pub writer_checkpoint: String, /// Path to researcher model checkpoint (relative to LIVE/MODELS/) pub researcher_checkpoint: String, /// Model dimension (embedding size) pub d_model: usize, /// Number of attention heads pub n_heads: usize, /// Number of transformer layers pub n_layers: usize, /// Vocabulary size (BPE merge count + special tokens) pub vocab_size: usize, /// Maximum sequence length pub max_seq_len: usize, /// Feed-forward hidden dimension (typically 4 × d_model) pub d_ff: usize, /// Learning rate for online training pub learning_rate: f64, /// Batch size for training (number of gate signals per batch) pub batch_size: usize, /// Enable online learning (train while serving) pub online_learning: bool, /// EWC lambda — elastic weight consolidation strength (0.0 = disabled) pub ewc_lambda: f64, /// Experience replay buffer size pub replay_buffer_size: usize, /// Temperature for generation (0.0 = greedy, 1.0 = full sampling) pub temperature: f64, } impl Default for TransformerConfig { fn default() -> Self { Self { enabled: false, writer_checkpoint: "writer_v1.spfc".to_string(), researcher_checkpoint: "researcher_v1.spfc".to_string(), // SPF Writer/Researcher spec: ~5M params d_model: 256, n_heads: 8, n_layers: 6, vocab_size: 8192, max_seq_len: 512, d_ff: 1024, // 4 × d_model learning_rate: 1e-4, batch_size: 32, online_learning: true, ewc_lambda: 0.4, replay_buffer_size: 10000, temperature: 0.7, } } } impl TransformerConfig { /// Load transformer config from JSON file, falling back to defaults pub fn load(path: &Path) -> anyhow::Result { if path.exists() { let content = std::fs::read_to_string(path)?; let config: Self = serde_json::from_str(&content)?; Ok(config) } else { Ok(Self::default()) } } /// Estimated parameter count for this configuration /// Rough formula: embedding + (n_layers × (4×d_model² + 2×d_model×d_ff)) + output pub fn estimated_params(&self) -> usize { let embedding = self.vocab_size * self.d_model; let pos_embedding = self.max_seq_len * self.d_model; // Per layer: self-attn (4 × d_model²) + cross-attn (4 × d_model²) + FFN (2 × d_model × d_ff) + norms let per_layer = 4 * self.d_model * self.d_model // self-attention Q/K/V/O + 4 * self.d_model * self.d_model // cross-attention Q/K/V/O + 2 * self.d_model * self.d_ff // FFN w1 + w2 + 4 * self.d_model; // layer norms (approx) let output = self.d_model * self.vocab_size; embedding + pos_embedding + self.n_layers * per_layer + output } /// Estimated memory usage in bytes (weights + Adam states) /// weights × 4 bytes × 3 (params + momentum + variance) pub fn estimated_memory_bytes(&self) -> usize { self.estimated_params() * 4 * 3 } } // ============================================================================ // BLOCK NP — NETWORK POOL CONFIGURATION // Follows MeshConfig/TransformerConfig pattern: struct + Default + load() // Loaded from LIVE/CONFIG/network.json // NetAdmin: orchestrates pool, assigns tasks, monitors workers (also does work) // Worker: accepts assignments from NetAdmin, executes, reports proof of work // ============================================================================ /// Role of this node in the network pool. /// NetAdmin orchestrates AND works. Worker accepts assignments only. #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)] #[serde(rename_all = "lowercase")] pub enum NodeRole { NetAdmin, Worker, } impl Default for NodeRole { fn default() -> Self { NodeRole::Worker } } /// A peer node in the network pool. #[derive(Debug, Clone, Serialize, Deserialize)] pub struct PoolPeer { /// Human-readable name (e.g. "ALPHA", "CHARLIE") pub name: String, /// Ed25519 public key hex — must match team.keys for mesh trust pub key_hex: String, /// QUIC port the peer listens on pub port: u16, /// Capabilities this worker exposes (e.g. ["tools", "web", "brain"]) #[serde(default)] pub capabilities: Vec, } /// Network pool configuration — loaded from LIVE/CONFIG/network.json /// Human-editable. One file per node. No recompile needed after changes. #[derive(Debug, Clone, Serialize, Deserialize)] pub struct NetworkConfig { /// This node's role in the pool #[serde(default)] pub role: NodeRole, /// Maximum concurrent workers in pool (hard cap: 8) #[serde(default = "default_pool_size")] pub pool_size: u8, /// Known worker peers — NetAdmin populates this, Workers leave empty #[serde(default)] pub peers: Vec, } fn default_pool_size() -> u8 { 8 } impl Default for NetworkConfig { fn default() -> Self { Self { role: NodeRole::Worker, pool_size: 8, peers: vec![], } } } impl NetworkConfig { /// Load network config from JSON file, falling back to defaults. /// Follows MeshConfig.load() pattern exactly. pub fn load(path: &Path) -> anyhow::Result { if path.exists() { let content = std::fs::read_to_string(path)?; let config: Self = serde_json::from_str(&content)?; Ok(config) } else { Ok(Self::default()) } } /// True if this node is the NetAdmin orchestrator. pub fn is_netadmin(&self) -> bool { self.role == NodeRole::NetAdmin } /// Effective pool size — enforces hard cap of 8 regardless of config value. pub fn effective_pool_size(&self) -> u8 { self.pool_size.min(8) } } // ============================================================================ // TESTS // ============================================================================ #[cfg(test)] mod tests { use super::*; #[test] fn tier_boundaries() { let config = SpfConfig::default(); assert_eq!(config.get_tier(0).0, "SIMPLE"); assert_eq!(config.get_tier(499).0, "SIMPLE"); assert_eq!(config.get_tier(500).0, "LIGHT"); assert_eq!(config.get_tier(1999).0, "LIGHT"); assert_eq!(config.get_tier(2000).0, "MEDIUM"); assert_eq!(config.get_tier(9999).0, "MEDIUM"); assert_eq!(config.get_tier(10000).0, "CRITICAL"); assert_eq!(config.get_tier(u64::MAX - 1).0, "CRITICAL"); } #[test] fn default_formula_exponents() { let config = SpfConfig::default(); assert_eq!(config.formula.basic_power, 1); assert_eq!(config.formula.deps_power, 7); assert_eq!(config.formula.complex_power, 10); assert_eq!(config.formula.files_multiplier, 10); assert_eq!(config.formula.w_eff, 40000.0); } #[test] fn default_enforce_mode_is_max() { let config = SpfConfig::default(); assert_eq!(config.enforce_mode, EnforceMode::Max); } #[test] fn blocked_paths_include_system_dirs() { let config = SpfConfig::default(); assert!(config.is_path_blocked("/tmp")); assert!(config.is_path_blocked("/tmp/evil.sh")); assert!(config.is_path_blocked("/etc/passwd")); assert!(config.is_path_blocked("/usr/bin/something")); } #[test] fn default_whitelists_are_empty() { let config = SpfConfig::default(); assert!(config.allowed_commands_user.is_empty(), "User whitelist must default empty (default-deny)"); assert!(config.allowed_commands_sandbox.is_empty(), "Sandbox whitelist must default empty (default-deny)"); assert!(config.user_fs_paths.is_empty(), "User FS paths must default empty"); } #[test] fn command_perm_constructors() { let r = CommandPerm::read_only(); assert!(r.read && !r.write && !r.execute); let rw = CommandPerm::read_write(); assert!(rw.read && rw.write && !rw.execute); let full = CommandPerm::full(); assert!(full.read && full.write && full.execute); } // ================================================================ // BLOCK NN-A — blocked_paths coverage tests // ================================================================ #[test] fn blocked_paths_include_claude_configs() { let config = SpfConfig::default(); let home = crate::paths::actual_home().to_string_lossy().to_string(); // .claude/ directory (pre-existing) assert!(config.blocked_paths.contains(&format!("{}/.claude/", home))); // BLOCK NN-A additions assert!(config.blocked_paths.contains(&format!("{}/.claude.json", home))); assert!(config.blocked_paths.contains(&format!("{}/.mcp.json", home))); assert!(config.blocked_paths.contains(&format!("{}/.config/anthropic/", home))); assert!(config.blocked_paths.contains(&format!("{}/.anthropic/", home))); } // ================================================================ // BLOCK SEC-2 — External agent config + CONFIG directory tests // ================================================================ #[test] fn blocked_paths_include_agent_config_dirs() { let config = SpfConfig::default(); let home = crate::paths::actual_home().to_string_lossy().to_string(); let root = crate::paths::spf_root().to_string_lossy().to_string(); // .qwen/ — home and root level assert!(config.blocked_paths.contains(&format!("{}/.qwen/", home)), "Home-level .qwen/ must be blocked to prevent config hijack"); assert!(config.blocked_paths.contains(&format!("{}/.qwen/", root)), "Root-level .qwen/ must be blocked to prevent config hijack"); // .xlaude/ — home and root level assert!(config.blocked_paths.contains(&format!("{}/.xlaude/", home)), "Home-level .xlaude/ must be blocked"); assert!(config.blocked_paths.contains(&format!("{}/.xlaude/", root)), "Root-level .xlaude/ must be blocked"); } #[test] fn blocked_paths_include_config_directory() { let config = SpfConfig::default(); let root = crate::paths::spf_root().to_string_lossy().to_string(); // LIVE/CONFIG/ directory — protects runtime JSON configs assert!(config.blocked_paths.contains(&format!("{}/LIVE/CONFIG/", root)), "LIVE/CONFIG/ must be blocked to protect runtime configs (http.json, mesh.json, etc.)"); // CONFIG.DB — already existed, verify still present assert!(config.blocked_paths.contains(&format!("{}/LIVE/CONFIG.DB", root)), "CONFIG.DB must remain blocked"); } // ================================================================ // BLOCK I — TransformerConfig tests // ================================================================ #[test] fn transformer_config_defaults() { let tc = TransformerConfig::default(); assert!(!tc.enabled); assert_eq!(tc.d_model, 256); assert_eq!(tc.n_heads, 8); assert_eq!(tc.n_layers, 6); assert_eq!(tc.vocab_size, 8192); assert_eq!(tc.max_seq_len, 512); assert_eq!(tc.d_ff, 1024); assert_eq!(tc.batch_size, 32); assert!(tc.online_learning); assert!((tc.learning_rate - 1e-4).abs() < 1e-10); assert!((tc.ewc_lambda - 0.4).abs() < 1e-10); assert!((tc.temperature - 0.7).abs() < 1e-10); } #[test] fn transformer_config_param_estimate() { let tc = TransformerConfig::default(); let params = tc.estimated_params(); assert!(params > 1_000_000, "Should be >1M params, got {}", params); assert!(params < 20_000_000, "Should be <20M params, got {}", params); } #[test] fn transformer_config_memory_estimate() { let tc = TransformerConfig::default(); let mem = tc.estimated_memory_bytes(); let expected = tc.estimated_params() * 12; assert_eq!(mem, expected); assert!(mem < 500_000_000, "Memory should be <500MB, got {}MB", mem / 1_000_000); } #[test] fn transformer_config_serialize_roundtrip() { let tc = TransformerConfig::default(); let json = serde_json::to_string_pretty(&tc).unwrap(); let back: TransformerConfig = serde_json::from_str(&json).unwrap(); assert_eq!(back.d_model, tc.d_model); assert_eq!(back.n_heads, tc.n_heads); assert_eq!(back.n_layers, tc.n_layers); assert_eq!(back.vocab_size, tc.vocab_size); assert_eq!(back.enabled, tc.enabled); assert!((back.temperature - tc.temperature).abs() < 1e-10); } #[test] fn transformer_config_custom_values() { let json = r#"{ "enabled": true, "writer_checkpoint": "custom_writer.spfc", "researcher_checkpoint": "custom_research.spfc", "d_model": 512, "n_heads": 16, "n_layers": 12, "vocab_size": 16384, "max_seq_len": 1024, "d_ff": 2048, "learning_rate": 3e-4, "batch_size": 64, "online_learning": false, "ewc_lambda": 0.8, "replay_buffer_size": 50000, "temperature": 1.0 }"#; let tc: TransformerConfig = serde_json::from_str(json).unwrap(); assert!(tc.enabled); assert_eq!(tc.d_model, 512); assert_eq!(tc.n_heads, 16); assert_eq!(tc.n_layers, 12); assert_eq!(tc.d_ff, 2048); assert!(!tc.online_learning); assert!((tc.learning_rate - 3e-4).abs() < 1e-10); } // ================================================================ // BLOCK NP — NetworkConfig tests // ================================================================ #[test] fn network_config_default_is_worker() { let nc = NetworkConfig::default(); assert_eq!(nc.role, NodeRole::Worker); assert_eq!(nc.pool_size, 8); assert!(nc.peers.is_empty()); assert!(!nc.is_netadmin()); } #[test] fn network_config_pool_size_capped_at_8() { let mut nc = NetworkConfig::default(); nc.pool_size = 20; assert_eq!(nc.effective_pool_size(), 8); nc.pool_size = 4; assert_eq!(nc.effective_pool_size(), 4); nc.pool_size = 0; assert_eq!(nc.effective_pool_size(), 0); } #[test] fn network_config_netadmin_serialize_roundtrip() { let json = r#"{ "role": "netadmin", "pool_size": 8, "peers": [ { "name": "ALPHA", "key_hex": "4c327f6d", "port": 4901, "capabilities": ["tools", "web"] } ] }"#; let nc: NetworkConfig = serde_json::from_str(json).unwrap(); assert_eq!(nc.role, NodeRole::NetAdmin); assert!(nc.is_netadmin()); assert_eq!(nc.peers.len(), 1); assert_eq!(nc.peers[0].name, "ALPHA"); assert_eq!(nc.peers[0].port, 4901); assert_eq!(nc.peers[0].capabilities, vec!["tools", "web"]); } #[test] fn network_config_worker_serialize_roundtrip() { let json = r#"{"role": "worker", "pool_size": 8, "peers": []}"#; let nc: NetworkConfig = serde_json::from_str(json).unwrap(); assert_eq!(nc.role, NodeRole::Worker); assert!(!nc.is_netadmin()); assert!(nc.peers.is_empty()); } #[test] fn network_config_defaults_on_missing_fields() { // role and pool_size have serde defaults — partial JSON must still work let json = r#"{}"#; let nc: NetworkConfig = serde_json::from_str(json).unwrap(); assert_eq!(nc.role, NodeRole::Worker); assert_eq!(nc.pool_size, 8); assert!(nc.peers.is_empty()); } }