Blog
New perspectives, new possibilities.
129 articles
A Pre-Launch Test Checklist for Claude Code Skills
A plain question triggers a review, while a real request never fires the skill. Write steps, add approval points, and test with related and unrelated requests.
2026-11-13 ·DeepDojo.AI
Splitting a Long SKILL.md Into Smaller Files
Cram checklists and templates into SKILL.md and every trigger reads them all. Use Claude Code's layered loading: method in the main file, the rest on demand.
2026-11-13 ·DeepDojo.AI
Validating Claude Code's JSON Output
is_error is false and the JSON parses, but is the conclusion right? Check the exit code, validate fields with a schema, then confirm the input supports the answer.
2026-11-13 ·DeepDojo.AI
How to Set Permissions for Claude Code in CI
In cron jobs and CI, nobody reads claude -p permission prompts. Start read-only, set tools, approvals, and deny rules by real actions, then cap turns and budget.
2026-11-13 ·DeepDojo.AI
Claude Code Undid the Code but Not the Database
Rewind restored the code, but test tables and new dependencies stayed behind. Learn what checkpoints restore, save each attempt, and compare from one start.
2026-11-13 ·DeepDojo.AI
Cleaning Up a Messy CLAUDE.md
CLAUDE.md rules go stale, land on the wrong layer, or never load. Pick a layer by who needs the rule, keep only lines that change actions, and verify with /context.
2026-11-13 ·DeepDojo.AI
Why Claude Code's "Read-Only" Command Edits Files
allowed-tools listed only Read and Grep, yet Claude Code edited files. Treat who starts a command, what skips approval, and what is blocked as separate controls.
2026-11-13 ·DeepDojo.AI
Turn Repeated Prompts Into Claude Code Commands
Rewriting the same request each time makes review standards drift. Write it as a skill-based slash command with clear inputs, context, output, and stop points.
2026-11-13 ·DeepDojo.AI
Which Claude Code Features Should You Learn First?
Turn on commands, memory, hooks, MCP, and subagents at once, and failures get hard to trace. Start with what you repeat daily, then add features one gap at a time.
2026-11-13 ·DeepDojo.AI
Find the Config Breaking Claude Code With Safe Mode
Only one repo breaks, and every customization is a suspect. Narrow it down with --safe-mode, restore one category at a time, and verify the fix on the original path.
2026-11-12 ·DeepDojo.AI