TL Fixes Blockers and Regresses

Purpose

Team Lead triages the UAT blocker list, invokes BPE-09 (Fix Bug) once per blocker, then sends Saga back for a targeted regression run over the affected scenarios. Loop until the regression verdict is CLEAN. Non-blockers are filed as deferred issues and do not gate close.

Entry condition: MIN-05 Saga produced a UAT Report with Blockers > 0. If blockers = 0, skip this activity — proceed directly to MIN-07.

Step 1: Triage Blockers

Read the UAT Report. For each blocker issue:

gh issue view {bug_number} --json number,title,body,labels

Extract: failing scenario, step ID, expected vs actual, affected feature. Group by affected file/feature. Prioritize: fix bugs that unblock the most downstream journey steps first.

Step 2: Invoke BPE-09 Per Blocker

For each blocker, run BPE-09 Fix Bug (Activity 203) as a fresh dr-dobbs subagent (cheap-fast model).

The BPE-09 input is the Bug Report filed by Saga in MIN-05. Pass it directly:

dr-dobbs — run BPE-09 Fix Bug for bug #{bug_number}.

Bug Report: gh issue view {bug_number} --json number,title,body

Follow BPE-09 steps exactly:
1. Analyze why existing tests missed the bug.
2. Write a failing test that reproduces it (Red).
3. Implement the minimal correct fix (Green).
4. Full regression: pytest tests/
5. Commit: fix({scope}): {bug_title} — Fixes #{bug_number}
6. Post NODE_PASS on the bug issue.

Scope: do not touch files unrelated to this bug. Do not patch symptoms.

Launch as a fresh Task subagent. Never resume a prior worker.

On NODE_PASS:

gh issue comment {bug_number} --body "<!-- FIX_COMMITTED -->\nCommit: {sha}\nTest gap: {one-line analysis}\n<!-- /FIX_COMMITTED -->"

On NODE_FAIL after worker’s internal BPE-09 retry: dispatch one additional TL-guided fix subagent with a narrower prompt targeting only the failing assertion. If still failing: escalate with drift-escalated label and await human decision.

Step 3: Defer Non-Blockers

For each Major and Minor defect from the UAT Report:

gh issue edit {issue_number} --add-label "deferred"
gh issue comment {issue_number} --body "Deferred. Non-blocker from iteration {slug}. Fix in next iteration."

Step 4: Regression UAT (Saga in Regression Mode)

After all blocker fixes are committed, invoke Saga as a fresh Task subagent (medium model):

Saga — regression run for iteration {slug}.
Re-execute only the scenarios / journeys that had failures in the prior UAT report:
  {affected_scenario_list}
Follow the full MIN-05 protocol (survey → execute → classify → report).
Produce a verdict: CLEAN or NEW_BLOCKERS_FOUND.

CLEAN → proceed to MIN-07 TL Closes Sprint with Lessons and Release.

NEW_BLOCKERS_FOUND → loop back to Step 1. Each loop iteration must make net progress (at least one fewer blocker than the prior loop). If the same blocker persists across two fix cycles: escalate.

Exit Condition

Saga regression verdict is CLEAN and no open blocker-labelled issues remain on the milestone.

Success Criteria

• Every UAT blocker passed through BPE-09 Fix Bug (test-gap analysis + failing test + fix + regression)
• Every fix has a commit where the reproducing test was red before and green after
• Full regression suite green after all fixes
• Saga regression run returns CLEAN over affected scenarios
• Non-blockers labeled deferred and removed from milestone gate
• Zero open blockers before proceeding to MIN-07

Agent

Team Lead (Orchestrator) — advanced reasoning model. Triages blockers, assembles BPE-09 prompts, monitors worker results, coordinates regression Saga invocation.

dr-dobbs (Worker) — cheap-fast model. Runs BPE-09 Fix Bug per blocker: one bug per invocation, test-first, full regression before NODE_PASS.

Saga (Regression UAT) — medium model. Re-runs only affected scenarios from tests/e2e or tests/uat.

Rules

• do-test-first (via BPE-09)
• do-not-mock-in-integration-tests (via BPE-09)
• do-follow-commit-convention
• do-small-increments