// Safe, sequential merge protocol for integrating N parallel worktree branches back into main. Defines merge ordering, quality gates between merges, conflict classification, and the updated worktree naming convention for multi-instance dispatches.
| name | parallel-dispatch-merge |
| description | Safe, sequential merge protocol for integrating N parallel worktree branches back into main. Defines merge ordering, quality gates between merges, conflict classification, and the updated worktree naming convention for multi-instance dispatches. |
Merge N parallel worktree branches safely. One branch at a time. Quality gate between each. Dependency order first.
Priority (highest to lowest):
@<agent>[integration] branches always merge after all feature branches for that agent type.Multi-instance dispatches use scope-qualified names:
# Format: .wt/<agent-name>-<scope>
# Branch: wt/<agent-name>-<scope>-<timestamp>
# Setup
git worktree add .wt/<agent-name>-<scope> -b wt/<agent-name>-<scope>-$(date +%s) HEAD
# Examples
git worktree add .wt/backend-engineer-auth -b wt/backend-engineer-auth-$(date +%s) HEAD
git worktree add .wt/backend-engineer-tasks -b wt/backend-engineer-tasks-$(date +%s) HEAD
git worktree add .wt/frontend-engineer-auth-ui -b wt/frontend-engineer-auth-ui-$(date +%s) HEAD
git worktree add .wt/scout-infra -b wt/scout-infra-$(date +%s) HEAD
Single-instance dispatches retain the original convention:
git worktree add .wt/<agent-name> -b wt/<agent-name>-$(date +%s) HEAD
Execute for each branch, in merge order:
# Ensure main is clean
git status # must show clean working tree
git log --oneline -1 # confirm HEAD is expected
git merge --squash wt/<agent-name>-<scope>-<ts>
git commit -m "<type>(<scope>): <description>
Parallel dispatch: @<agent-name>[<scope>]
Scope card: <brief deliverable summary>"
Commit message includes the parallel dispatch context for traceability.
Run language-appropriate quality checks per GEMINI.md § Code Completion Mandate (the single source of truth for quality commands per language).
If quality gate fails: Fix issues before merging the next branch. The fix belongs to the scope of the branch that introduced the failure.
git worktree remove .wt/<agent-name>-<scope>
git branch -D wt/<agent-name>-<scope>-<ts>
Move to the next branch in merge order. Repeat Steps 1-5.
| Type | Description | Resolution Strategy |
|---|---|---|
| Textual | Same line modified differently by two branches | Analyze intent of both changes. Pick correct version or combine. Test after resolution. |
| Semantic | Different changes that interact logically (e.g., both add a dependency) | Requires understanding both features. May need new code that satisfies both intents. |
| Structural | File reorganization conflicts (renamed, moved) | Re-apply changes on the new structure. |
| Integration | Router/config aggregation (both add routes) | Combine both additions. Order by convention (alphabetical, feature grouping). |
# After resolving conflicts
git add <resolved-files>
git commit -m "<type>(<scope>): <description>
Conflict resolution: merged @<agent>[<scope-A>] changes with @<agent>[<scope-B>]
Resolution: <brief rationale>"
If the same conflict pattern repeats across multiple merges, the decomposition was flawed:
Read-only nodes (scout, qa-analyst, security-engineer, ux-reviewer, incident-responder) don't produce branches. They produce documents and findings that are passed as context to downstream nodes:
No merge step needed for read-only nodes. Their output is context, not code.
In a full pipeline, merge points occur at phase boundaries:
SCOUT (read-only, no merge) →
DESIGN (merge design branches) →
PRE-MORTEM (read-only, no merge — produces advisory findings) →
BUILD (merge feature branches per level) →
Level M+1: merge parallel feature branches
Level M+2: merge integration branches
REVIEW (read-only, no merge) →
REMEDIATE (merge fix branches) →
VERIFY (single agent, no parallel merge)
Each merge point follows this same per-branch protocol.
Before declaring a merge point complete:
git status)Session bootstrap + workflows for Pathfinder semantic navigation tools. Covers: discovery protocol, tool chaining patterns (explore, impact, audit, debug), search optimization, LSP degraded mode, and error recovery.
Playwright browser automation via MCP. Covers E2E testing, UI review, web scraping, screenshot capture, and general browser interaction. MCP-first — CLI is fallback only.
Safe command execution: input sanitization, timeout handling, output capture, error propagation. For spawning processes, shell commands, system calls.
Git conventions: conventional commits, branch naming, PR hygiene, release tagging.
Structured incident workflow: severity classification, triage, diagnosis, mitigation, postmortem, and prevention. Template-driven with blameless review.
Constructs, validates, and traverses a Directed Acyclic Graph (DAG) from scope cards for safe level-based parallel dispatch. Determines execution order via topological sort. Detects cycles and invalid dependencies.