Use when multiple fix attempts fail and you need to systematically restore to a working baseline and reimplement instead of fixing broken code.
Reverts to last known working baseline after multiple failed fix attempts, then systematically reimplements changes one step at a time with local verification between each commit. Use when stuck in "fix the fix" loops or when accumulated changes make root cause unclear.
/plugin marketplace add TheBushidoCollective/han/plugin install jutsu-scratch@hanThis skill is limited to using the following tools:
Enforces methodical problem-solving by reverting to last known working state and reimplementing step-by-step instead of trying to fix accumulated broken changes.
When something breaks after multiple failed fix attempts:
# Find last known working state
git log --oneline -20
git show origin/beta:path/to/file.sh # Check beta/main branch
git diff origin/beta -- path/to/file.sh # What changed?
# Verify baseline works
git stash
git checkout origin/beta -- path/to/file.sh
# Test it - does it work?
# Get exact differences
git diff baseline..current -- path/to/file.sh
# Understand each change
# For EACH diff hunk, ask:
# - Why was this changed?
# - What problem was it trying to solve?
# - Did it actually solve that problem?
# Hard revert to working state
git checkout origin/beta -- path/to/file.sh
git add path/to/file.sh
git commit -m "Revert to working baseline from beta"
# Verify baseline works
./test-locally.sh
# Must pass before proceeding
Critical: Don't proceed until baseline is verified working.
Make ONE small change
# Example: Replace sed with awk in ONE function
# Don't change 5 things at once
Test locally immediately
./run-generation-script.sh
terraform validate
# Must pass before committing
Commit if working
git add changed-file.sh
git commit -m "Replace sed with awk in function X"
If it breaks, revert immediately
git reset --hard HEAD~1
# Try different approach or understand why it broke
Repeat for next change
# Run full test suite
make test
terraform validate
# Compare with original broken state
# Did we achieve the goal without breaking things?
# Push only after local verification
git push
Before committing ANY change:
# Made 10 changes trying to "optimize" variable filtering
# Each fix broke something new
# Spent day+ debugging
# Check beta branch - does it work?
git show origin/beta:terraform/build-module.sh > /tmp/beta-version.sh
bash /tmp/beta-version.sh # Verify it works
# Revert to beta version
git checkout origin/beta -- terraform/build-module.sh
# Now reimplement ONLY what's needed (e.g., sed→awk for portability)
# One function at a time, test each change
# Assume it's a CI environment issue
# Try 5 different "fixes" based on guesses
# Each creates new errors
# Find last passing pipeline
git log --oneline | head -20
# Check what changed since then
git diff <last-passing-commit>
# Revert suspicious changes
# Test locally before pushing
# Find working baseline
git log --oneline --all | grep "known working feature"
git show origin/beta:path/to/file
# Compare with baseline
git diff origin/beta -- path/to/file
git diff <working-commit> -- path/to/file
# Revert to baseline
git checkout origin/beta -- path/to/file
git checkout <working-commit> -- path/to/file
# Test locally
terraform validate
mix test
yarn test
# Verify no changes after running script
git diff # Should be empty if script is idempotent
This skill should be used when the user asks to "create an agent", "add an agent", "write a subagent", "agent frontmatter", "when to use description", "agent examples", "agent tools", "agent colors", "autonomous agent", or needs guidance on agent structure, system prompts, triggering conditions, or agent development best practices for Claude Code plugins.
This skill should be used when the user asks to "create a slash command", "add a command", "write a custom command", "define command arguments", "use command frontmatter", "organize commands", "create command with file references", "interactive command", "use AskUserQuestion in command", or needs guidance on slash command structure, YAML frontmatter fields, dynamic arguments, bash execution in commands, user interaction patterns, or command development best practices for Claude Code.
This skill should be used when the user asks to "create a hook", "add a PreToolUse/PostToolUse/Stop hook", "validate tool use", "implement prompt-based hooks", "use ${CLAUDE_PLUGIN_ROOT}", "set up event-driven automation", "block dangerous commands", or mentions hook events (PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, SessionEnd, UserPromptSubmit, PreCompact, Notification). Provides comprehensive guidance for creating and implementing Claude Code plugin hooks with focus on advanced prompt-based hooks API.