Imported from diegosouzapw/awesome-omni-skill (skills/tools/agent-ops-planning/SKILL.md). Install upstream with npx skills add diegosouzapw/awesome-omni-skill --skill agent-ops-planning. Copyright stays with the author.

Planning workflow

Works with or without aoc CLI installed. Issue tracking can be done via direct file editing.

Issue Tracking (File-Based — Default)

Build Commands (from constitution)

# Read actual commands from .agent/constitution.md to understand project structure
build: uv run python -m build
test: uv run pytest

Reference Documents

Implementation details are stored as markdown:

.agent/issues/references/{ISSUE-ID}-impl-plan.md

CLI Integration (when aoc is available)

When aoc CLI is detected in .agent/tools.json, these commands provide convenience shortcuts:

Preconditions

• Work should be tracked as an issue before planning begins
• Constitution exists and is baseline-ready (or stop and run constitution workflow)
• Baseline exists if any code change is expected (or stop and run baseline workflow first)

Issue-First Principle

Before starting detailed planning:

1. Check for existing issue: Is there already an issue for this work?

   • Yes → proceed with planning, reference the issue ID
   • No → suggest creating one first

2. Create issue if needed:

   This work isn't tracked yet. Create an issue first?
   
   Suggested: FEAT-{next}@{hash} — "{title from request}"
   Priority: {inferred priority}
   
   [Y]es, create and continue / [N]o, plan without issue
   

3. Reference issue throughout:

   • Plan title: "Plan for {ISSUE-ID}: {title}"
   • Update issue status to in_progress when planning starts
   • Link plan to issue in notes section

Iterations based on confidence

Read confidence from:

1. Task's confidence field (if set)
2. Otherwise, constitution's default confidence

Low Confidence Mandatory Interview (invoke agent-ops-interview)

When confidence is LOW, an interview is MANDATORY before planning begins.

This ensures assumptions are surfaced and clarified with the human before any design work.

Interview Trigger

🎯 LOW CONFIDENCE DETECTED — Mandatory Interview Required

Before planning {ISSUE-ID}, I need to clarify key aspects with you.
This is required for low confidence work to reduce implementation risk.

Starting structured interview (one question at a time)...

Interview Questions Template

Ask these questions ONE AT A TIME, waiting for response before proceeding:

1. Scope Boundaries

   Q1: What is explicitly OUT OF SCOPE for this issue?
   (List anything I should NOT touch or change)
   

2. Expected Behavior

   Q2: Can you describe the expected behavior in specific terms?
   (What should happen when X? What output for input Y?)
   

3. Edge Cases

   Q3: What edge cases should I consider?
   (Empty inputs, errors, concurrent access, etc.)
   

4. Testing Expectations

   Q4: What testing approach do you expect?
   (Unit tests? Integration? Manual verification? Specific scenarios?)
   

5. Success Criteria

   Q5: How will you know this is done correctly?
   (What will you check during code review?)
   

6. Known Constraints

   Q6: Are there any constraints I should know about?
   (Performance requirements, compatibility, dependencies, etc.)
   

Interview Notes Capture

After interview completes:

1. Summarize answers in issue notes section
2. Create .agent/issues/references/{ISSUE-ID}-interview.md if answers are extensive
3. Link interview notes from issue's spec_file or notes field

### Interview Summary (YYYY-MM-DD)
- **Out of scope**: {answer}
- **Expected behavior**: {answer}
- **Edge cases**: {answer}
- **Testing**: {answer}
- **Success criteria**: {answer}
- **Constraints**: {answer}

Interview Bypass (NOT RECOMMENDED)

User may skip interview, but must acknowledge the risk:

⚠️ Skipping interview for low confidence issue is NOT recommended.
This increases risk of incorrect implementation.

Skip anyway? [Y]es, I accept the risk / [N]o, let's do the interview

If skipped, log in issue: "Interview skipped by user — higher risk accepted"

Preconditions

• Constitution exists and is baseline-ready (or stop and run constitution workflow).
• Baseline exists if any code change is expected (or stop and run baseline workflow first).

Procedure

1. Intake:

   • restate the goal (1–3 lines)
   • list unknowns as explicit questions
   • stop and ask until clarified; no guessing

2. Plan iteration 1:

   • steps
   • files expected to change
   • files that must not change
   • test strategy
   • risks/unknowns
   • why it is minimal change

3. Validate iteration 1:

   • check every requirement
   • check constitution constraints
   • check baseline constraints
   • identify assumptions; convert assumptions into questions

4. Plan iteration 2+ (if confidence requires):

   • revise based on validation
   • tighten diffs and test plan

5. Generate Implementation Details (invoke agent-ops-impl-details):

   • Determine detail level from confidence (see table above)
   • Generate detailed implementation specification
   • Save to .agent/issues/references/{ISSUE-ID}-impl-plan.md
   • Link from issue's spec_file field

6. Final implementation plan:

   • numbered steps
   • acceptance criteria mapping
   • test plan mapping
   • reference to implementation details file

7. Approval gate (based on confidence):

   • low: HARD gate — wait for explicit approval, plan for single issue only
   • normal: SOFT gate — ask "Ready to implement?", continue if no objection
   • high: MINIMAL — proceed unless user objects

   LOW confidence additional requirements:

   • Plan must cover exactly 1 issue (no batching)
   • Reference document in .agent/issues/references/ is MANDATORY
   • Implementation will have HARD STOP after completion for human review
   • Test coverage threshold: ≥90% line, ≥85% branch on changed code

8. Update .agent/focus.md and issue status via agent-ops-state.

Implementation Details Integration

MANDATORY: After plan iterations are complete, you MUST generate implementation details.

This is not optional. Every plan must have an implementation details file.

Low Confidence → Extensive Details

For risky, complex, or uncertain changes:

Invoking agent-ops-impl-details with level: extensive

Output MUST include:
- ACTUAL EXECUTABLE CODE for each change (not pseudo-code)
- Complete function implementations with types
- Edge case handling with specific code
- Error scenarios with specific exception handling
- Full test cases with assertions
- Import statements
- Docstrings

Example extensive output:

# File: src/services/user.py
# Change: Add process_user function

from datetime import datetime
from typing import Optional
from .models import User, UserResult
from .exceptions import NotFoundError

def process_user(user_id: str, db: Database) -> UserResult:
    """Process user with validation.
    
    Args:
        user_id: The user's unique identifier
        db: Database connection
        
    Returns:
        UserResult with processed user data
        
    Raises:
        ValueError: If user_id is invalid
        NotFoundError: If user doesn't exist
    """
    if not user_id:
        raise ValueError("user_id is required")
    
    user = db.get_user(user_id)
    if user is None:
        raise NotFoundError(f"User {user_id} not found")
    
    return UserResult(
        id=user.id,
        name=user.name,
        processed_at=datetime.utcnow(),
    )

Normal Confidence → Normal Details

For standard features and typical changes:

Invoking agent-ops-impl-details with level: normal

Output includes:
- Function signatures with parameter types
- Pseudo-code for logic flow
- Data structure definitions
- API contracts (request/response)
- Key test scenarios (not full code)

High Confidence → Low Details

For simple, well-understood changes:

Invoking agent-ops-impl-details with level: low

Output includes:
- Files to change with brief description
- High-level approach (1-2 sentences per file)
- Dependencies and risks
- Basic test coverage outline

Detail Level Override

User can override the default level:

Plan with extensive details regardless of confidence? [y/n]

Or specify in the planning request: "Plan with extensive implementation details"

Issue Discovery During Planning

During planning, invoke agent-ops-tasks discovery procedure for:

1. Sub-tasks discovered:

   • Large feature breaks into multiple issues
   • Prerequisites that need addressing first
   • "Before we can do X, we need to Y"

2. Risks identified:

   • Technical risks → CHORE or TEST issues
   • Security concerns → SEC issues
   • Performance concerns → PERF issues

3. Dependencies found:

   • External blockers → blocked issues
   • Missing APIs/features → FEAT issues

Present after plan iteration:

📋 Planning revealed {N} additional work items:

- [FEAT] API endpoint for user preferences (prerequisite)
- [TEST] Integration tests needed for payment flow
- [DOCS] Update API documentation for new fields

Create issues for these? [A]ll / [S]elect / [N]one

These will be linked as dependencies/related to {ORIGINAL-ISSUE-ID}.

After creating sub-issues:

Created {N} related issues. What's next?

1. Continue planning {ORIGINAL-ISSUE-ID} (with dependencies noted)
2. Plan the prerequisite first ({NEW-ISSUE-ID})
3. Review all related issues