GitLab Boards & Labels

## Context
Brief description of what needs to be done and why.

## Scope
### Included
- Specific deliverable 1
- Specific deliverable 2

### Excluded
- What's explicitly NOT part of this work

## Acceptance Criteria
- [ ] Criterion 1 is met
- [ ] Criterion 2 is met
- [ ] Tests pass

### Blockers (check when resolved)
- [ ] Depends on issue #XY being complete

### Approval
- [ ] Read comment ← Check if you have feedback for AI
- [ ] Work complete and verified

## Context
Why this decision matters and what's at stake.

## Options
- [ ] A: First option with description
- [ ] B: Second option with description
- [ ] C: Third option with description
- [ ] Other: See comment below

## AI Recommendation
Option B is recommended because...

## ADR Link
docs/adr/20260124-decision-topic.md

### Approval
- [ ] Read comment
- [ ] Decision confirmed

1. AI completes work, moves card to "review"
   Card state: [ ] Read comment  [x] Criterion 1  [x] Criterion 2

2. Human reviews, has feedback
   Human adds comment: "Please also handle the edge case where X"
   Human checks: [x] Read comment

3. AI picks up card, sees checked "Read comment"
   AI reads comments, finds human's latest
   AI acts on instruction: handles edge case X
   AI UNCHECKS: [ ] Read comment    ← critical step (prevents infinite loop)
   AI moves card back to "review"

4. Human reviews again
   Satisfied? → Close the issue
   Not satisfied? → Add comment, check "Read comment" again (loop repeats)

AI Processing a Card (10-Step Protocol)

 1. CLAIM:    Move issue backlog/ready → in-progress
              glab issue update {IID} --unlabel "ready" --label "in-progress"

 2. READ:     Parse issue body - understand requirements

 3. CHECK:    Is "Read comment" checked? If yes, process feedback first

 4. BRANCH:   Create feature branch (issue-{IID}-{short-desc})

 5. EXECUTE:  Do the actual work (code, config, docs)

 6. COMMIT:   Stage, commit with Co-Authored-By footer

 7. PUSH:     Push branch to remote, create MR if needed

 8. UPDATE:   Check acceptance criteria boxes as each is completed
              Replace "- [ ]" with "- [x]" for each met criterion

 9. APPROVE:  If ALL criteria checked:
              Check "Work complete and verified" box
              Move to "done" (NOT closed)
              Add completion comment summarizing what was done

10. REPORT:   Inform user: "Work is in done column for your review"

PHASE Card Format:

Title: PHASE 3: Pipeline Integration

## Phase Goal
Integrate specification workflow into the CI/CD pipeline stages.

## Tasks to Create When Activated
| Task | Title              | Agent     | Description              |
|------|--------------------|-----------|--------------------------|
| T3.1 | Add spec templates | Developer | Create markdown templates |
| T3.2 | Map stage commands | Architect | Map pipeline stages       |
| T3.3 | Test integration   | QA        | Validate end-to-end       |

### Approval
- [ ] Read comment
- [ ] Phase tasks created

Step 1: AI or human creates DECISION card with options as checkboxes
         - [ ] A: Sequential pipeline (simpler, slower)
         - [ ] B: Parallel pipeline (faster, complex)
         - [ ] C: Hybrid approach
         - [ ] Other: See comment below

Step 2: Human reviews options, checks ONE checkbox
         - [ ] A: Sequential pipeline
         - [x] B: Parallel pipeline    ← human's choice
         - [ ] C: Hybrid approach

Step 3: Human moves card to "ready"

Step 4: AI scans board, finds DECISION in ready
         AI parses description, finds checked option = B
         AI adds confirmation comment
         AI removes decision-needed label
         AI moves to "done" and closes

If "Other" is chosen:
         Human checks "Other: See comment below"
         Human adds comment explaining their custom choice
         AI reads the comment to understand the decision

./scripts/replicate-labels.sh [source_group] [target_group]
# Default: administrators → developers

What it does:
1. Looks up group IDs dynamically (not hardcoded)
2. Fetches all labels from source group via GitLab API
3. Creates each label in target group
4. Handles "already exists" gracefully
5. Reports final count: "Created 27 labels, 3 already existed"