Run a git command and return its output.

* Denies the spawn request and exits * @param {string} reason - Reason for denial

* Reads the agent registry from disk * @param {string} registryPath - Path to registry file * @returns {Object} Registry object

* Writes the agent registry to disk * @param {string} registryPath - Path to registry file * @param {Object} registry - Registry object to write

* Removes quotes from a string value * @param {*} value - Value to unquote * @returns {*} Unquoted value

* Updates the completion status for an agent in the registry * @param {string} registryPath - Path to registry file * @param {string} agentId - Agen

* Build the combined context message

* Creates the delegation message to return to the user * @param {Object} hookData - Hook data object * @param {string} agentLogPath - Path to agent

* Get tool calls from assistant messages

* Loads an agent definition from file * @param {string} agentType - Type of agent to load * @returns {Object} Agent definition with systemPrompt, mo

* Load ignore patterns from a file (one pattern per line, # for comments)

* Normalizes a list by removing duplicates and empty values * @param {Array} list - List to normalize * @returns {Array} Normalized list

* Resolves the file path for an agent definition * @param {string} baseDir - Base directory to search in * @param {string} agentType - Agent type to

* Main registry function - runs all observers in parallel

* Converts a value to a normalized list * @param {*} value - Value to convert * @returns {Array|null} List or null

* Updates the PID for an agent in the registry * @param {string} registryPath - Path to registry file * @param {string} agentId - Agent ID * @param

* Writes the initial log file for an agent * @param {string} agentLogPath - Path to agent log file * @param {string} description - Task description

* Interactive annotation session

* Auto-annotate prompts using Claude Sonnet 4.5

* Background worker that processes all directories changed in the session

* Background worker that generates and saves the history entry

* Background validation worker

* Background validation worker (runs as detached process)

* Calculate confusion matrix

* Calculate effort correlation

* Calculate per-category metrics

* Calculate metrics

* Run activity tracker logic on a single prompt * This replicates the logic from activity-tracker.js

* Categorize prompt with specific model

* Checks if the current agent has permission to spawn the subagent * @param {string|null} parentAgentId - Parent agent ID * @param {string} subagent

* Specialized observers (existing ones)

Check for outdated Anthropic model usage.

Check for 'any' type usage in TypeScript/JavaScript.

Check for fallback and legacy patterns in code files only.

Check for outdated OpenAI model usage.

* Collects CLAUDE.md files from file directory up to cwd, then includes HOME

* Creates a registry entry for an agent * @param {string} agentId - Unique agent identifier * @param {number} currentDepth - Current recursion depth

Create a focused prompt for fixing errors in a specific file.

Detect which output style to use based on prompt keywords.

* Extracts user prompts from transcript JSONL files * @param {string} transcriptPath - Path to transcript file * @returns {Array<{prompt: string, se

Find the nearest package.json to determine project root.

* Generates a unique agent ID * @returns {string} Agent ID