orca-confirm
Handle team confirmation with automatic bypass mode detection for /orca
When & Why to Use This Skill
The Orca Team Confirmation skill is a specialized orchestration component designed to manage the approval workflow for multi-agent systems. It streamlines the transition from planning to execution by presenting proposed agent pipelines and intelligently handling user confirmations. Its standout feature is the automatic bypass mode detection, which prevents workflow bottlenecks in automated environments by recognizing when manual input is unavailable and proceeding based on predefined permissions.
Use Cases
- Optimizing autonomous agent workflows by automatically detecting bypass permissions to ensure continuous execution without manual intervention.
- Enhancing transparency in complex software development tasks by providing users with a structured markdown summary of proposed agent roles, project domains, and potential risks.
- Implementing Human-in-the-Loop (HITL) governance by allowing developers to review, confirm, or modify AI-suggested team compositions before resource-intensive tasks begin.
- Standardizing communication between different agent modules through structured status reporting (BYPASS, CONFIRMED, or MODIFIED) for reliable system parsing.
| name | orca-confirm |
|---|---|
| description | Handle team confirmation with automatic bypass mode detection for /orca |
Orca Team Confirmation Skill
Purpose: Handle user confirmation for agent team selection with automatic bypass mode detection.
Input: User's request is implicitly available. You receive the proposed team and project type as arguments.
Critical Rules (Read First)
- Ask for confirmation ONCE using AskUserQuestion
- Check response structure immediately after receiving it
- If response.answers is empty → BYPASS MODE → Return immediately
- NEVER retry AskUserQuestion if first attempt returned blank
- Return a structured result that main /orca can parse
Implementation
Step 1: Present Proposed Team (VISIBLE OUTPUT - NOT IN TOOL CALL)
CRITICAL: This must be VISIBLE MARKDOWN OUTPUT, not inside AskUserQuestion.
Output this as regular text so the user can see it:
## Proposed Pipeline
**Request:** [task description]
**Domain:** [PROJECT_TYPE]
**Complexity:** [simple/medium/complex]
### Agent Team
| Role | Agent |
|------|-------|
| Coordination | [grand-architect] |
| Architecture | [architect] |
| Implementation | [builder] |
| Specialists | [list them] |
| Gates | [gate agents] |
| Verification | [verification agent] |
### Files Likely Affected
- [from ContextBundle]
### Risks/Notes
- [any identified]
This output MUST appear BEFORE you call AskUserQuestion.
Step 2: Ask for Confirmation (ONCE ONLY - SIMPLE QUESTION)
Use AskUserQuestion with these exact options:
Question: "Proceed with this team?"
Options:
1. "Yes, proceed" → Proceed with proposed team
2. "Modify team" → Ask user what to change
Step 3: Check Response Structure
Immediately after AskUserQuestion returns, check the response:
# Pseudo-code for what you MUST check:
if response is None or response.get("answers") is None or len(response.get("answers", {})) == 0:
# BYPASS MODE DETECTED
# User has bypass permissions enabled
# They CANNOT respond to questions
# Proceed immediately with proposed team
Output:
" Proceeding with proposed team (bypass mode detected - confirmation unavailable)
Proposed Team: [list team]
You can interrupt if you need changes."
# Return to /orca with: BYPASS|original_team
else:
# INTERACTIVE MODE
# User provided an answer
user_answer = response["answers"][first_question_key]
if user_answer == "Yes, proceed":
Output: " Team confirmed"
# Return to /orca with: CONFIRMED|original_team
elif user_answer == "Modify team":
# Ask user what to change
# Adjust team based on their input
# Return to /orca with: MODIFIED|new_team_list
Step 4: Return Result to /orca
Format: STATUS|agent1,agent2,agent3
Valid statuses:
BYPASS- Bypass mode detected, proceeding with proposed teamCONFIRMED- User confirmed proposed teamMODIFIED- User requested changes, new team provided
Examples:
BYPASS|swiftui-developer,swiftdata-specialist,swift-testing-specialist
CONFIRMED|design-reviewer,accessibility-specialist,ui-engineer
MODIFIED|nextjs-14-specialist,tailwind-specialist,design-reviewer
Output Format
Your final output must be:
[Status message to user explaining what happened]
RESULT: STATUS|agent,list,here
The RESULT line is what /orca will parse.
Examples
Example 1: Bypass Mode (Blank Response)
Input: Proposed team for iOS project
AskUserQuestion returns: {"answers": {}}
Output:
Proceeding with proposed team (bypass mode detected - confirmation unavailable)
Proposed Team:
• swiftui-developer
• swiftdata-specialist
• swift-testing-specialist
You can interrupt if you need changes.
RESULT: BYPASS|swiftui-developer,swiftdata-specialist,swift-testing-specialist
Example 2: User Confirms
Input: Proposed team for Next.js project
AskUserQuestion returns: {"answers": {"q1": "Yes, proceed"}}
Output:
Team confirmed by user
Proceeding with:
• nextjs-14-specialist
• tailwind-specialist
• design-reviewer
RESULT: CONFIRMED|nextjs-14-specialist,tailwind-specialist,design-reviewer
Example 3: User Modifies
Input: Proposed team for backend project User says: "Remove quality-validator, add test-engineer" Output:
Team modified per user request
Original: backend-engineer, quality-validator
Modified: backend-engineer, test-engineer
RESULT: MODIFIED|backend-engineer,test-engineer
Testing Checklist
Before returning results, verify:
- Asked AskUserQuestion exactly ONCE
- Checked response.answers structure explicitly
- Returned RESULT line in correct format
- Did NOT retry on blank response
- Showed clear message to user about what's happening
Common Mistakes to Avoid
DON'T:
"I didn't receive a response, let me ask again..."
[Retries AskUserQuestion]
DO:
" Proceeding with proposed team (bypass mode detected)"
RESULT: BYPASS|team,list
DON'T:
# Unclear what happened
RESULT: team,list
DO:
# Clear status prefix
RESULT: BYPASS|team,list
This skill exists to solve one problem: Stop retrying when bypass mode is enabled.
It works by encapsulating the confirmation logic and enforcing single-attempt behavior.