Process Directory List

Process Directory List

Process the directory-instructions.json file and generate AGENTS.md files for each directory by invoking the /generate-instructions skill.

Purpose

This skill orchestrates the batch generation of AGENTS.md files for multiple directories in the monorepo. It reads the directory list JSON file and systematically processes each directory item.

Input

Reads from directory-instructions.json in the project root, which has this structure:

{
  "items": [
    {
      "id": "1",
      "directory": "app/controllers",
      "output_file": "AGENTS.md",
      "status": "pending",
      "analyzed_at": null
    }
  ]
}

Steps

1. Read the Directory List

• Load directory-instructions.json from the current working directory
• Parse the JSON and extract the items array
• Display summary: "Found X directories to process"

2. Spawn Agents for Each Directory

For each item in the items array where status is "pending":

Create Agents

Use Codex agent delegation to process each pending directory in parallel.

Each agent receives this prompt:

Process directory "{directory_path}" from directory-instructions.json
(ID: {id}):

Steps:
1. Read directory-instructions.json
2. Update item {id} to set status="in_progress"
3. Write the updated JSON back
4. Run: /generate-instructions {directory_path}
5. Read directory-instructions.json again
6. Update item {id} to set status="completed" and
   analyzed_at="{ISO timestamp}"
7. Write the final JSON back
8. If any errors occur, set status="error" instead

The /generate-instructions skill will analyze the directory and create the
AGENTS.md file.

Launch All Agents in Parallel

CRITICAL: Spawn all agents in a single batch so each directory is processed concurrently.

Example:

spawn_agent(worker, "Process app/controllers and generate AGENTS.md")
spawn_agent(worker, "Process app/models and generate AGENTS.md")
spawn_agent(worker, "Process app/services and generate AGENTS.md")
... (one agent per pending directory)

This allows all directories to be processed simultaneously.

3. Monitor and Report Results

After all agents have completed, display a summary:

Completed: 6 directories
Errors: 1 directory
Skipped: 1 directory (already completed)

Launched 7 agents in parallel

Usage Notes

• Idempotent: Running this skill multiple times will only process pending items
• Resumable: If interrupted, re-run to continue from where it left off
• Parallel Processing: Spawns multiple agents simultaneously, one per directory
• Status tracking: Each agent updates the JSON file for its directory independently
• Efficient: All directories are processed concurrently for maximum speed

Example Workflow

# 1. First create the directory list
/create-directory-list

# 2. Then process all directories
/process-directory-list

# 3. Check results
cat directory-instructions.json

Important

• Use Codex agent delegation for parallel processing
• Each agent should invoke /generate-instructions directly; do not reimplement its logic
• Each directory gets its own dedicated agent for independent processing
• Agents must update the JSON file after their directory completes
• Preserve the JSON structure and formatting when updating the file. Do not add entries.
• Launch all agents in a single batch for true parallelism