Do
<CRITICAL>
This skill has MANDATORY requirements. You do NOT have a choice. You MUST follow every step exactly as written. This is not negotiable. This is not optional. You cannot rationalize your way out of this.
This skill handles EVERYTHING: planning, execution, review, and git. One command to rule them all.
</CRITICAL>
Overview
The main workflow skill. Automatically:
- Creates a plan if none exists (or resumes existing plan)
- Executes the plan step by step
- Runs mandatory code review
- Handles git operations if enabled
Mandatory Announcements
You MUST announce every action BEFORE taking it. No exceptions.
| Action | Required Announcement |
|---|
| Starting skill | "Using the do skill." |
| Checking plan | "Checking for existing plan..." |
| No plan found | "No plan found. Starting planning phase." |
| Plan found | "Plan found. Resuming execution." |
| Starting phase | "Phase [N]: [name]" |
| Loading module | "Loading [module]." |
| Asking question | "Question: [text]" |
| Noting answer | "Noted: [summary]." |
| Detecting | "Detecting [what]... [Result] detected from [source]." |
| Writing plan | "Writing plan to .b-claude/plan.md." |
| Executing step | "Executing step [N]: [description]" |
| Step complete | "Step [N] complete." |
| Batch complete | "Batch complete. Ready for feedback." |
| Starting review | "Starting mandatory code review." |
| Issue found | "Issue found: [severity] - [description]" |
| Committing | "Committing: [message]" |
| Complete | "Task complete." |
Red Flags - You Are Rationalizing If You Think:
| Thought | Reality |
|---|
| "I can skip planning" | NO. Planning is mandatory for new tasks. |
| "I already understand the request" | WRONG. Ask clarifying questions anyway. |
| "I know this codebase" | FALSE. Analyze it anyway. |
| "I'll combine steps" | NO. One step at a time. |
| "Code review is optional" | NEVER. Review is mandatory. |
| "This is simple" | DOESN'T MATTER. Full process every time. |
| "I'll announce at the end" | WRONG. Announce BEFORE each action. |
| "I'll skip the checkpoint" | NO. Every checkpoint is mandatory. |
Process
Step 0: Initialize
Step 0.1: Announce
"Using the do skill."
"Checking for .b-claude/preferences.md..."
Step 0.2: Preferences
If no preferences exist, ask ONE AT A TIME:
"Preference: Do you want emojis? (yes/no)"
"Preference: Auto-commit enabled? (yes/no)"
"Preference: Branch management? (yes/no)"
Save to .b-claude/preferences.md.
Step 1: Check for Plan
"Checking for existing plan in .b-claude/plan.md..."
- If plan exists with uncompleted steps → Go to Step 3 (Execute)
- If plan exists but all steps done → Ask: "Plan complete. New task?"
- If no plan exists → Go to Step 2 (Plan)
Step 2: Planning Phase
Only runs if no plan exists.
Phase 2.1: Think
"Phase 2.1: Think -- Understanding the request."
Ask clarifying questions ONE AT A TIME:
- What is the goal?
- What are the constraints?
- What are the acceptance criteria?
"Noted: [summary of answer]."
After all questions:
"Understanding established:
- Goal: [goal]
- Constraints: [constraints]
- Criteria: [criteria]"
CHECKPOINT: "Is this understanding correct?"
Phase 2.2: Analyze
"Phase 2.2: Analyze -- Scanning the project."
"Loading internal/language module."
Follow the language module to detect and announce:
- Language
- Frameworks
- Runtimes
- Environments
"Project analysis:
- Language: [lang]
- Frameworks: [list]
- Architecture: [pattern]
- Key files: [list]"
CHECKPOINT: "Is this analysis correct?"
Phase 2.3: Planify
"Phase 2.3: Planify -- Creating the plan."
Create a step-by-step plan with checkboxes:
markdown
1# Plan: [task name]
2
3## Context
4- Language: [lang]
5- Frameworks: [list]
6
7## Steps
8- [ ] Step 1: [description]
9- [ ] Step 2: [description]
10- [ ] Step 3: [description]
11...
"Writing plan to .b-claude/plan.md."
"Plan created with [N] steps."
CHECKPOINT: "Approve this plan to proceed?"
Step 3: Execution Phase
"Starting execution phase."
"Loading plan from .b-claude/plan.md."
Step 3.1: Branch (if enabled)
If branch management enabled:
"Creating branch: [branch-name]"
Step 3.2: Execute Steps
Batch size: 3 steps max. Stop after each batch.
For EACH step:
"Executing step [N]: [description]"
While working:
"Step [N] in progress: [action]"
When done:
"Step [N] complete. Marking as done."
Update checkbox in .b-claude/plan.md.
If auto-commit enabled:
"Committing: [message]"
CHECKPOINT after each batch:
"Batch complete. Steps [X-Y] done."
"Continue with next batch?"
STOP and wait.
Step 3.3: Handle Blockers
If blocked:
"BLOCKED at step [N]: [reason]"
"Need clarification: [question]"
STOP and wait.
Step 4: Code Review (MANDATORY)
"All steps complete. Starting mandatory code review."
Review all changes against:
- Language conventions
- Code quality (SOLID, complexity, etc.)
- Security (no secrets, injection, XSS)
For each issue:
"Issue found: [SEVERITY] - [file:line] - [description]"
"Code review complete. Verdict: [APPROVE/REQUEST_CHANGES]"
If issues found:
- CRITICAL/MAJOR: Fix immediately, re-commit
- MINOR: Ask user "Fix now or defer?"
Step 5: Completion
"Task complete."
"Summary:"
"- Steps completed: [N]"
"- Commits made: [N]"
"- Code review: [verdict]"
If git push enabled:
"Push to remote? (yes/no)"
STOP Conditions
STOP immediately and ask if:
- Request is ambiguous
- Multiple approaches possible
- Destructive action required
- CRITICAL issue found
- Step unclear
- Conflict with requirements
Internal Modules
Automatically loaded during analysis:
internal/language - Convention detection
<REMINDER>
You MUST announce EVERY step. You MUST wait at EVERY checkpoint. Code review is MANDATORY. No shortcuts. No exceptions.
</REMINDER>
