Execute Phase

Fallback rule: If a spawned agent completes its work (commits visible, SUMMARY.md exists) but the orchestrator never receives the completion signal, treat it as successful based on spot-checks and continue to the next wave/plan. Never block indefinitely waiting for a signal — always verify via filesystem and git state.

@/.claude/get-shit-done/references/agent-contracts.md @/.claude/get-shit-done/references/context-budget.md @~/.claude/get-shit-done/references/gates.md

• gsd-executor — Executes plan tasks, commits, creates SUMMARY.md
• gsd-verifier — Verifies phase completion, checks quality gates
• gsd-planner — Creates detailed plans from phase scope
• gsd-phase-researcher — Researches technical approaches for a phase
• gsd-plan-checker — Reviews plan quality before execution
• gsd-debugger — Diagnoses and fixes issues
• gsd-codebase-mapper — Maps project structure and dependencies
• gsd-integration-checker — Checks cross-phase integration
• gsd-nyquist-auditor — Validates verification coverage
• gsd-ui-researcher — Researches UI/UX approaches
• gsd-ui-checker — Reviews UI implementation quality
• gsd-ui-auditor — Audits UI against design requirements

• First positional token → PHASE_ARG
• Optional --wave N → WAVE_FILTER
• Optional --gaps-only keeps its current meaning
• Optional --cross-ai → CROSS_AI_FORCE=true (force all plans through cross-AI execution)
• Optional --no-cross-ai → CROSS_AI_DISABLED=true (disable cross-AI for this run, overrides config and frontmatter)

If --wave is absent, preserve the current behavior of executing all incomplete waves in the phase.

INIT=$(gsd-sdk query init.execute-phase "${PHASE_ARG}")
if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
AGENT_SKILLS=$(gsd-sdk query agent-skills gsd-executor 2>/dev/null)

Parse JSON for: executor_model, verifier_model, commit_docs, parallelization, branching_strategy, branch_name, phase_found, phase_dir, phase_number, phase_name, phase_slug, plans, incomplete_plans, plan_count, incomplete_count, state_exists, roadmap_exists, phase_req_ids, response_language.

If response_language is set: Include response_language: {value} in all spawned subagent prompts so any user-facing output stays in the configured language.

Read worktree config:

USE_WORKTREES=$(gsd-sdk query config-get workflow.use_worktrees 2>/dev/null || echo "true")

If the project uses git submodules, worktree isolation is skipped regardless of the workflow.use_worktrees config — the executor commit protocol cannot correctly handle submodule commits inside isolated worktrees. Sequential execution handles submodules transparently.

if [ -f .gitmodules ]; then
  echo "[worktree] Submodule project detected (.gitmodules exists) — falling back to sequential execution"
  USE_WORKTREES=false
fi

When USE_WORKTREES is false, all executor agents run without isolation="worktree" — they execute sequentially on the main working tree instead of in parallel worktrees.

Read context window size for adaptive prompt enrichment:

CONTEXT_WINDOW=$(gsd-sdk query config-get context_window 2>/dev/null || echo "200000")

When CONTEXT_WINDOW >= 500000 (1M-class models), subagent prompts include richer context:

• Executor agents receive prior wave SUMMARY.md files and the phase CONTEXT.md/RESEARCH.md
• Verifier agents receive all PLAN.md, SUMMARY.md, CONTEXT.md files plus REQUIREMENTS.md
• This enables cross-phase awareness and history-aware verification

When CONTEXT_WINDOW < 200000 (sub-200K models), subagent prompts are thinned to reduce static overhead:

• Executor agents omit extended deviation rule examples and checkpoint examples from inline prompt — load on-demand via @~/.claude/get-shit-done/references/executor-examples.md
• Planner agents omit extended anti-pattern lists and specificity examples from inline prompt — load on-demand via @~/.claude/get-shit-done/references/planner-antipatterns.md
• Core rules and decision logic remain inline; only verbose examples and edge-case lists are extracted
• This reduces executor static overhead by ~40% while preserving behavioral correctness

If phase_found is false: Error — phase directory not found. If plan_count is 0: Error — no plans found in phase. If state_exists is false but .planning/ exists: Offer reconstruct or continue.

When parallelization is false, plans within a wave execute sequentially.

Runtime detection for Copilot: Check if the current runtime is Copilot by testing for the @gsd-executor agent pattern or absence of the Task() subagent API. If running under Copilot, force sequential inline execution regardless of the parallelization setting — Copilot's subagent completion signals are unreliable (see <runtime_compatibility>). Set COPILOT_SEQUENTIAL=true internally and skip the execute_waves step in favor of check_interactive_mode's inline path for each plan.

REQUIRED — Sync chain flag with intent. If user invoked manually (no --auto), clear the ephemeral chain flag from any previous interrupted --auto chain. This prevents stale _auto_chain_active: true from causing unwanted auto-advance. This does NOT touch workflow.auto_advance (the user's persistent settings preference). You MUST execute this bash block before any config reads:

# REQUIRED: prevents stale auto-chain from previous --auto runs
if [[ ! "$ARGUMENTS" =~ --auto ]]; then
  gsd-sdk query config-set workflow._auto_chain_active false 2>/dev/null
fi

Look for a .continue-here.md in the current phase directory:

ls ${phase_dir}/.continue-here.md 2>/dev/null || true

If .continue-here.md exists, parse its "Critical Anti-Patterns" table for rows with severity = blocking.

If one or more blocking anti-patterns are found:

This step cannot be skipped. Before proceeding to check_interactive_mode or any other step, the agent must demonstrate understanding of each blocking anti-pattern by answering all three questions for each one:

1. What is this anti-pattern? — Describe it in your own words, not by quoting the handoff.
2. How did it manifest? — Explain the specific failure that caused it to be recorded.
3. What structural mechanism (not acknowledgment) prevents it? — Name the concrete step, checklist item, or enforcement mechanism that stops recurrence.

Write these answers inline before continuing. If a blocking anti-pattern cannot be answered from the context in .continue-here.md, stop and ask the user for clarification.

If no .continue-here.md exists, or no blocking rows are found: Proceed directly to check_interactive_mode.

If --interactive flag present: Switch to interactive execution mode.

Interactive mode executes plans sequentially inline (no subagent spawning) with user checkpoints between tasks. The user can review, modify, or redirect work at any point.

Interactive execution flow:

1. Load plan inventory as normal (discover_and_group_plans)

2. For each plan (sequentially, ignoring wave grouping):

   a. Present the plan to the user:

   ## Plan {plan_id}: {plan_name}
   
   Objective: {from plan file}
   Tasks: {task_count}
   
   Options:
   - Execute (proceed with all tasks)
   - Review first (show task breakdown before starting)
   - Skip (move to next plan)
   - Stop (end execution, save progress)
   

   b. If "Review first": Read and display the full plan file. Ask again: Execute, Modify, Skip.

   c. If "Execute": Read and follow ~/.claude/get-shit-done/workflows/execute-plan.md inline (do NOT spawn a subagent). Execute tasks one at a time.

   d. After each task: Pause briefly. If the user intervenes (types anything), stop and address their feedback before continuing. Otherwise proceed to next task.

   e. After plan complete: Show results, commit, create SUMMARY.md, then present next plan.

3. After all plans: proceed to verification (same as normal mode).

Benefits of interactive mode:

• No subagent overhead — dramatically lower token usage
• User catches mistakes early — saves costly verification cycles
• Maintains GSD's planning/tracking structure
• Best for: small phases, bug fixes, verification gaps, learning GSD

Skip to handle_branching step (interactive plans execute inline after grouping).

"none": Skip, continue on current branch.

"phase" or "milestone": Use pre-computed branch_name from init:

git checkout -b "$BRANCH_NAME" 2>/dev/null || git checkout "$BRANCH_NAME"

All subsequent commits go to this branch. User handles merging.

Report: "Found {plan_count} plans in {phase_dir} ({incomplete_count} incomplete)"

Update STATE.md for phase start:

gsd-sdk query state.begin-phase --phase "${PHASE_NUMBER}" --name "${PHASE_NAME}" --plans "${PLAN_COUNT}"

This updates Status, Last Activity, Current focus, Current Position, and plan counts in STATE.md so frontmatter and body text reflect the active phase immediately.

PLAN_INDEX=$(gsd-sdk query phase-plan-index "${PHASE_NUMBER}")

Parse JSON for: phase, plans[] (each with id, wave, autonomous, objective, files_modified, task_count, has_summary), waves (map of wave number → plan IDs), incomplete, has_checkpoints.

Filtering: Skip plans where has_summary: true. If --gaps-only: also skip non-gap_closure plans. If WAVE_FILTER is set: also skip plans whose wave does not equal WAVE_FILTER.

Wave safety check: If WAVE_FILTER is set and there are still incomplete plans in any lower wave that match the current execution mode, STOP and tell the user to finish earlier waves first. Do not let Wave 2+ execute while prerequisite earlier-wave plans remain incomplete.

If all filtered: "No matching incomplete plans" → exit.

Report:

## Execution Plan

**Phase {X}: {Name}** — {total_plans} matching plans across {wave_count} wave(s)

{If WAVE_FILTER is set: `Wave filter active: executing only Wave {WAVE_FILTER}`.}

| Wave | Plans | What it builds |
|------|-------|----------------|
| 1 | 01-01, 01-02 | {from plan objectives, 3-8 words} |
| 2 | 01-03 | ... |

This step runs after plan discovery and before normal wave execution. It identifies plans that should be delegated to an external AI command and executes them via stdin-based prompt delivery. Plans handled here are removed from the execute_waves plan list so the normal executor skips them.

Activation logic:

1. If CROSS_AI_DISABLED is true (--no-cross-ai flag): skip this step entirely.
2. If CROSS_AI_FORCE is true (--cross-ai flag): mark ALL incomplete plans for cross-AI execution.
3. Otherwise: check each plan's frontmatter for cross_ai: true AND verify config workflow.cross_ai_execution is true. Plans matching both conditions are marked for cross-AI.

CROSS_AI_ENABLED=$(gsd-sdk query config-get workflow.cross_ai_execution 2>/dev/null || echo "false")
CROSS_AI_CMD=$(gsd-sdk query config-get workflow.cross_ai_command 2>/dev/null || echo "")
CROSS_AI_TIMEOUT=$(gsd-sdk query config-get workflow.cross_ai_timeout 2>/dev/null || echo "300")

If no plans are marked for cross-AI: Skip to execute_waves.

If plans are marked but cross_ai_command is empty: Error — tell user to set workflow.cross_ai_command via gsd-sdk query config-set workflow.cross_ai_command "<command>".

For each cross-AI plan (sequentially):

1. Construct the task prompt from the plan file:

   • Extract <objective> and <tasks> sections from the PLAN.md
   • Append PROJECT.md context (project name, description, tech stack)
   • Format as a self-contained execution prompt

2. Check for dirty working tree before execution:

   if ! git diff --quiet HEAD 2>/dev/null; then
     echo "WARNING: dirty working tree detected — the external AI command may produce uncommitted changes that conflict with existing modifications"
   fi
   

3. Run the external command from the project root, writing the prompt to stdin. Never shell-interpolate the prompt — always pipe via stdin to prevent injection:

   echo "$TASK_PROMPT" | timeout "${CROSS_AI_TIMEOUT}s" ${CROSS_AI_CMD} > "$CANDIDATE_SUMMARY" 2>"$ERROR_LOG"
   EXIT_CODE=$?
   

4. Evaluate the result:

   Success (exit 0 + valid summary):

   • Read $CANDIDATE_SUMMARY and validate it contains meaningful content (not empty, has at least a heading and description — a valid SUMMARY.md structure)
   • Write it as the plan's SUMMARY.md file
   • Update STATE.md plan status to complete
   • Update ROADMAP.md progress
   • Mark plan as handled — skip it in execute_waves

   Failure (non-zero exit or invalid summary):

   • Display the error output and exit code
   • Warn: "The external command may have left uncommitted changes or partial edits in the working tree. Review git status and git diff before proceeding."
   • Offer three choices:
     • retry — run the same plan through cross-AI again
     • skip — fall back to normal executor for this plan (re-add to execute_waves list)
     • abort — stop execution entirely, preserve state for resume

5. After all cross-AI plans processed: Remove successfully handled plans from the incomplete plan list so execute_waves skips them. Any skipped-to-fallback plans remain in the list for normal executor processing.

For each wave:

1. Intra-wave files_modified overlap check (BEFORE spawning):

   Before spawning any agents for this wave, inspect the files_modified list of all plans in the wave. Check every pair of plans in the wave — if any two plans share even one file in their files_modified lists, those plans have an implicit dependency and MUST NOT run in parallel.

   Detection algorithm (pseudocode):

   seen_files = {}
   overlapping_plans = []
   for each plan in wave_plans:
     for each file in plan.files_modified:
       if file in seen_files:
         overlapping_plans.add(plan, seen_files[file])  # both plans overlap on this file
       else:
         seen_files[file] = plan
   

   If overlap is detected:

   • Warn the user:

     ⚠ Intra-wave files_modified overlap detected in Wave {N}:
       Plan {A} and Plan {B} both modify {file}
       Running these plans sequentially to avoid parallel worktree conflicts.
     

   • Override PARALLELIZATION to false for this wave only — run all plans in the wave sequentially regardless of the global parallelization setting.
   • This is a safety net for plans that were incorrectly assigned to the same wave. The planner should have caught this; flag it as a planning defect so the user can replan the phase if desired.

   If no overlap: proceed normally (parallel if PARALLELIZATION=true).

2. Describe what's being built (BEFORE spawning):

   Read each plan's <objective>. Extract what's being built and why.

   ---
   ## Wave {N}
   
   **{Plan ID}: {Plan Name}**
   {2-3 sentences: what this builds, technical approach, why it matters}
   
   Spawning {count} agent(s)...
   ---
   

   • Bad: "Executing terrain generation plan"
   • Good: "Procedural terrain generator using Perlin noise — creates height maps, biome zones, and collision meshes. Required before vehicle physics can interact with ground."

3. Spawn executor agents:

   Pass paths only — executors read files themselves with their fresh context window. For 200k models, this keeps orchestrator context lean (~10-15%). For 1M+ models (Opus 4.6, Sonnet 4.6), richer context can be passed directly.

   Worktree mode (USE_WORKTREES is not false):

   Before spawning, capture the current HEAD:

   EXPECTED_BASE=$(git rev-parse HEAD)
   

   Sequential dispatch for parallel execution (waves with 2+ agents): When spawning multiple agents in a wave, dispatch each Task() call one at a time with run_in_background: true — do NOT send all Task calls in a single message. git worktree add acquires an exclusive lock on .git/config.lock, so simultaneous calls race for this lock and fail. Sequential dispatch ensures each worktree finishes creation before the next begins (the round-trip latency of each tool call provides natural spacing), while all agents still run in parallel once created.

   # CORRECT: dispatch one Task() per message, each with run_in_background: true
   # → worktrees created sequentially, agents execute in parallel
   #
   # WRONG: multiple Task() calls in a single message
   # → simultaneous git worktree add → .git/config.lock contention → failures
   

   Task(
     subagent_type="gsd-executor",
     description="Execute plan {plan_number} of phase {phase_number}",
     model="{executor_model}",
     isolation="worktree",
     prompt="
       <objective>
       Execute plan {plan_number} of phase {phase_number}-{phase_name}.
       Commit each task atomically. Create SUMMARY.md.
       Do NOT update STATE.md or ROADMAP.md — the orchestrator owns those writes after all worktree agents in the wave complete.
       </objective>
   
       <worktree_branch_check>
       FIRST ACTION before any other work: verify this worktree's branch is based on the correct commit.
   
       Run:
       ```bash
       ACTUAL_BASE=$(git merge-base HEAD {EXPECTED_BASE})
       ```
   
       If `ACTUAL_BASE` != `{EXPECTED_BASE}` (i.e. the worktree branch was created from an older
       base such as `main` instead of the feature branch HEAD), hard-reset to the correct base:
       ```bash
       # Safe: this runs before any agent work, so no uncommitted changes to lose
       git reset --hard {EXPECTED_BASE}
       # Verify correction succeeded
       if [ "$(git rev-parse HEAD)" != "{EXPECTED_BASE}" ]; then
         echo "ERROR: Could not correct worktree base — aborting to prevent data loss"
         exit 1
       fi
       ```
   
       `reset --hard` is safe here because this is a fresh worktree with no user changes. It
       resets both the HEAD pointer AND the working tree to the correct base commit (#2015).
   
       If `ACTUAL_BASE` == `{EXPECTED_BASE}`: the branch base is correct, proceed immediately.
   
       This check fixes a known issue where `EnterWorktree` creates branches from
       `main` instead of the current feature branch HEAD (affects all platforms).
       </worktree_branch_check>
   
       <parallel_execution>
       You are running as a PARALLEL executor agent in a git worktree.
       Use --no-verify on all git commits to avoid pre-commit hook contention
       with other agents. The orchestrator validates hooks once after all agents complete.
       For `gsd-sdk query commit` (or legacy `gsd-tools.cjs` commit): add --no-verify flag when needed.
       For direct git commits: use git commit --no-verify -m "..."
   
       IMPORTANT: Do NOT modify STATE.md or ROADMAP.md. execute-plan.md
       auto-detects worktree mode (`.git` is a file, not a directory) and skips
       shared file updates automatically. The orchestrator updates them centrally
       after merge.
   
       REQUIRED: SUMMARY.md MUST be committed before you return. In worktree mode the
       git_commit_metadata step in execute-plan.md commits SUMMARY.md and REQUIREMENTS.md
       only (STATE.md and ROADMAP.md are excluded automatically). Do NOT skip or defer
       this commit — the orchestrator force-removes the worktree after you return, and
       any uncommitted SUMMARY.md will be permanently lost (#2070).
       </parallel_execution>
   
       <execution_context>
       @~/.claude/get-shit-done/workflows/execute-plan.md
       @~/.claude/get-shit-done/templates/summary.md
       @~/.claude/get-shit-done/references/checkpoints.md
       @~/.claude/get-shit-done/references/tdd.md
       ${CONTEXT_WINDOW < 200000 ? '' : '@~/.claude/get-shit-done/references/executor-examples.md'}
       </execution_context>
   
       <files_to_read>
       Read these files at execution start using the Read tool:
       - {phase_dir}/{plan_file} (Plan)
       - .planning/PROJECT.md (Project context — core value, requirements, evolution rules)
       - .planning/STATE.md (State)
       - .planning/config.json (Config, if exists)
       ${CONTEXT_WINDOW >= 500000 ? `
       - ${phase_dir}/*-CONTEXT.md (User decisions from discuss-phase — honors locked choices)
       - ${phase_dir}/*-RESEARCH.md (Technical research — pitfalls and patterns to follow)
       - ${prior_wave_summaries} (SUMMARY.md files from earlier waves in this phase — what was already built)
       ` : ''}
       - ./CLAUDE.md (Project instructions, if exists — follow project-specific guidelines and coding conventions)
       - .claude/skills/ or .agents/skills/ (Project skills, if either exists — list skills, read SKILL.md for each, follow relevant rules during implementation)
       </files_to_read>
   
       ${AGENT_SKILLS}
   
       <mcp_tools>
       If CLAUDE.md or project instructions reference MCP tools (e.g. jCodeMunch, context7,
       or other MCP servers), prefer those tools over Grep/Glob for code navigation when available.
       MCP tools often save significant tokens by providing structured code indexes.
       Check tool availability first — if MCP tools are not accessible, fall back to Grep/Glob.
       </mcp_tools>
   
       <success_criteria>
       - [ ] All tasks executed
       - [ ] Each task committed individually
       - [ ] SUMMARY.md created in plan directory
       - [ ] No modifications to shared orchestrator artifacts (the orchestrator handles all post-wave shared-file writes)
       </success_criteria>
     "
   )
   

   Sequential mode (USE_WORKTREES is false):

   Omit isolation="worktree" from the Task call. Replace the <parallel_execution> block with:

       <sequential_execution>
       You are running as a SEQUENTIAL executor agent on the main working tree.
       Use normal git commits (with hooks). Do NOT use --no-verify.
       </sequential_execution>
   

   The sequential mode Task prompt uses the same structure as worktree mode but with these differences in success_criteria — since there is only one agent writing at a time, there are no shared-file conflicts:

       <success_criteria>
       - [ ] All tasks executed
       - [ ] Each task committed individually
       - [ ] SUMMARY.md created in plan directory
       - [ ] STATE.md updated with position and decisions
       - [ ] ROADMAP.md updated with plan progress (via `roadmap update-plan-progress`)
       </success_criteria>
   

   When worktrees are disabled, execute plans one at a time within each wave (sequential) regardless of the PARALLELIZATION setting — multiple agents writing to the same working tree concurrently would cause conflicts.

4. Wait for all agents in wave to complete.

   Completion signal fallback (Copilot and runtimes where Task() may not return):

   If a spawned agent does not return a completion signal but appears to have finished its work, do NOT block indefinitely. Instead, verify completion via spot-checks:

   # For each plan in this wave, check if the executor finished:
   SUMMARY_EXISTS=$(test -f "{phase_dir}/{plan_number}-{plan_padded}-SUMMARY.md" && echo "true" || echo "false")
   COMMITS_FOUND=$(git log --oneline --all --grep="{phase_number}-{plan_padded}" --since="1 hour ago" | head -1)
   

   If SUMMARY.md exists AND commits are found: The agent completed successfully — treat as done and proceed to step 5. Log: "✓ {Plan ID} completed (verified via spot-check — completion signal not received)"

   If SUMMARY.md does NOT exist after a reasonable wait: The agent may still be running or may have failed silently. Check git log --oneline -5 for recent activity. If commits are still appearing, wait longer. If no activity, report the plan as failed and route to the failure handler in step 6.

   This fallback applies automatically to all runtimes. Claude Code's Task() normally returns synchronously, but the fallback ensures resilience if it doesn't.

5. Post-wave hook validation (parallel mode only):

   When agents committed with --no-verify, run pre-commit hooks once after the wave:

   # Run project's pre-commit hooks on the current state
   git diff --cached --quiet || git stash  # stash any unstaged changes
   git hook run pre-commit 2>&1 || echo "⚠ Pre-commit hooks failed — review before continuing"
   

   If hooks fail: report the failure and ask "Fix hook issues now?" or "Continue to next wave?"

5.5. Worktree cleanup (when isolation="worktree" was used):

When executor agents ran in worktree isolation, their commits land on temporary branches in separate working trees. After the wave completes, merge these changes back and clean up:

# List worktrees created by this wave's agents
WORKTREES=$(git worktree list --porcelain | grep "^worktree " | grep -v "$(pwd)$" | sed 's/^worktree //')

for WT in $WORKTREES; do
  # Get the branch name for this worktree
  WT_BRANCH=$(git -C "$WT" rev-parse --abbrev-ref HEAD 2>/dev/null)
  if [ -n "$WT_BRANCH" ] && [ "$WT_BRANCH" != "HEAD" ]; then
    CURRENT_BRANCH=$(git rev-parse --abbrev-ref HEAD)

    # --- Orchestrator file protection (#1756) ---
    # Snapshot orchestrator-owned files BEFORE merge. If the worktree
    # branch outlived a milestone transition, its versions of STATE.md
    # and ROADMAP.md are stale. Main always wins for these files.
    STATE_BACKUP=$(mktemp)
    ROADMAP_BACKUP=$(mktemp)
    [ -f .planning/STATE.md ] && cp .planning/STATE.md "$STATE_BACKUP" || true
    [ -f .planning/ROADMAP.md ] && cp .planning/ROADMAP.md "$ROADMAP_BACKUP" || true

    # Snapshot list of files on main BEFORE merge to detect resurrections
    PRE_MERGE_FILES=$(git ls-files .planning/)

    # Pre-merge deletion check: warn if the worktree branch deletes tracked files
    DELETIONS=$(git diff --diff-filter=D --name-only HEAD..."$WT_BRANCH" 2>/dev/null || true)
    if [ -n "$DELETIONS" ]; then
      echo "BLOCKED: Worktree branch $WT_BRANCH contains file deletions: $DELETIONS"
      echo "Review these deletions before merging. If intentional, remove this guard and re-run."
      rm -f "$STATE_BACKUP" "$ROADMAP_BACKUP"
      continue
    fi

    # Merge the worktree branch into the current branch (--no-ff ensures a merge commit so HEAD~1 is reliable)
    git merge "$WT_BRANCH" --no-ff --no-edit -m "chore: merge executor worktree ($WT_BRANCH)" 2>&1 || {
      echo "⚠ Merge conflict from worktree $WT_BRANCH — resolve manually"
      echo "  STATE.md backup:   $STATE_BACKUP"
      echo "  ROADMAP.md backup: $ROADMAP_BACKUP"
      echo "  Restore with: cp \$STATE_BACKUP .planning/STATE.md && cp \$ROADMAP_BACKUP .planning/ROADMAP.md"
      break
    }

    # Post-merge deletion audit: detect bulk file deletions in merge commit (#2384)
    # --diff-filter=D HEAD~1 HEAD shows files deleted by the merge commit itself.
    # Exclude .planning/ — orchestrator-owned deletions there are expected (resurrections
    # are handled below). Require ALLOW_BULK_DELETE=1 to bypass for intentional large refactors.
    MERGE_DEL_COUNT=$(git diff --diff-filter=D --name-only HEAD~1 HEAD 2>/dev/null | grep -vc '^\.planning/' || true)
    if [ "$MERGE_DEL_COUNT" -gt 5 ] && [ "${ALLOW_BULK_DELETE:-0}" != "1" ]; then
      MERGE_DELETIONS=$(git diff --diff-filter=D --name-only HEAD~1 HEAD 2>/dev/null | grep -v '^\.planning/' || true)
      echo "⚠ BLOCKED: Merge of $WT_BRANCH deleted $MERGE_DEL_COUNT files outside .planning/ — reverting to protect repository integrity (#2384)"
      echo "$MERGE_DELETIONS"
      echo "  If these deletions are intentional, re-run with ALLOW_BULK_DELETE=1"
      git reset --hard HEAD~1 2>/dev/null || true
      rm -f "$STATE_BACKUP" "$ROADMAP_BACKUP"
      continue
    fi

    # Restore orchestrator-owned files (main always wins)
    if [ -s "$STATE_BACKUP" ]; then
      cp "$STATE_BACKUP" .planning/STATE.md
    fi
    if [ -s "$ROADMAP_BACKUP" ]; then
      cp "$ROADMAP_BACKUP" .planning/ROADMAP.md
    fi
    rm -f "$STATE_BACKUP" "$ROADMAP_BACKUP"

    # Detect files deleted on main but re-added by worktree merge
    # (e.g., archived phase directories that were intentionally removed)
    # A "resurrected" file must have a deletion event in main's ancestry —
    # brand-new files (e.g. SUMMARY.md just created by the executor) have no
    # such history and must NOT be removed (#2501).
    DELETED_FILES=$(git diff --diff-filter=A --name-only HEAD~1 -- .planning/ 2>/dev/null || true)
    for RESURRECTED in $DELETED_FILES; do
      # Only delete if this file was previously tracked on main and then
      # deliberately removed (has a deletion event in git history).
      WAS_DELETED=$(git log --follow --diff-filter=D --name-only --format="" HEAD~1 -- "$RESURRECTED" 2>/dev/null | grep -c . || true)
      if [ "${WAS_DELETED:-0}" -gt 0 ]; then
        git rm -f "$RESURRECTED" 2>/dev/null || true
      fi
    done

    # Amend merge commit with restored files if any changed
    if ! git diff --quiet .planning/STATE.md .planning/ROADMAP.md 2>/dev/null || \
       [ -n "$DELETED_FILES" ]; then
      # Only amend the commit with .planning/ files if commit_docs is enabled (#1783)
      COMMIT_DOCS=$(gsd-sdk query config-get commit_docs 2>/dev/null || echo "true")
      if [ "$COMMIT_DOCS" != "false" ]; then
        git add .planning/STATE.md .planning/ROADMAP.md 2>/dev/null || true
        git commit --amend --no-edit 2>/dev/null || true
      fi
    fi

    # Safety net: commit any uncommitted SUMMARY.md before force-removing the worktree.
    # This guards against executors that skipped the git_commit_metadata step (#2070).
    UNCOMMITTED_SUMMARY=$(git -C "$WT" ls-files --modified --others --exclude-standard -- "*SUMMARY.md" 2>/dev/null || true)
    if [ -n "$UNCOMMITTED_SUMMARY" ]; then
      echo "⚠ SUMMARY.md was not committed by executor — committing now to prevent data loss"
      git -C "$WT" add -- "*SUMMARY.md" 2>/dev/null || true
      git -C "$WT" commit --no-verify -m "docs(recovery): rescue uncommitted SUMMARY.md before worktree removal (#2070)" 2>/dev/null || true
      # Re-merge the recovery commit
      git merge "$WT_BRANCH" --no-edit -m "chore: merge rescued SUMMARY.md from executor worktree ($WT_BRANCH)" 2>/dev/null || true
    fi

    # Remove the worktree
    if ! git worktree remove "$WT" --force; then
      WT_NAME=$(basename "$WT")
      if [ -f ".git/worktrees/${WT_NAME}/locked" ]; then
        echo "⚠ Worktree $WT is locked — attempting to unlock and retry"
        git worktree unlock "$WT" 2>/dev/null || true
        if ! git worktree remove "$WT" --force; then
          echo "⚠ Residual worktree at $WT — manual cleanup required after session exits:"
          echo "    git worktree unlock \"$WT\" && git worktree remove \"$WT\" --force && git branch -D \"$WT_BRANCH\""
        fi
      else
        echo "⚠ Residual worktree at $WT (remove failed) — investigate manually"
      fi
    fi

    # Delete the temporary branch
    git branch -D "$WT_BRANCH" 2>/dev/null || true
  fi
done

If workflow.use_worktrees is false: Agents ran on the main working tree — skip this step entirely.

If no worktrees found: Skip silently — agents may have been spawned without worktree isolation.

5.6. Post-merge test gate (parallel mode only):

After merging all worktrees in a wave, run the project's test suite to catch cross-plan integration issues that individual worktree self-checks cannot detect (e.g., conflicting type definitions, removed exports, import changes).

This addresses the Generator self-evaluation blind spot identified in Anthropic's harness engineering research: agents reliably report Self-Check: PASSED even when merging their work creates failures.

# Resolve test command: project config > Makefile > language sniff
TEST_CMD=$(gsd-sdk query config-get workflow.test_command --default "" 2>/dev/null || true)
if [ -z "$TEST_CMD" ]; then
  if [ -f "Makefile" ] && grep -q "^test:" Makefile; then
    TEST_CMD="make test"
  elif [ -f "Justfile" ] || [ -f "justfile" ]; then
    TEST_CMD="just test"
  elif [ -f "package.json" ]; then
    TEST_CMD="npm test"
  elif [ -f "Cargo.toml" ]; then
    TEST_CMD="cargo test"
  elif [ -f "go.mod" ]; then
    TEST_CMD="go test ./..."
  elif [ -f "pyproject.toml" ] || [ -f "requirements.txt" ]; then
    TEST_CMD="python -m pytest -x -q --tb=short 2>&1 || uv run python -m pytest -x -q --tb=short"
  else
    TEST_CMD="true"
    echo "⚠ No test runner detected — skipping post-merge test gate"
  fi
fi
# Detect test runner and run quick smoke test (timeout: 5 minutes)
TEST_EXIT=0
timeout 300 bash -c "$TEST_CMD" 2>&1
TEST_EXIT=$?
if [ "${TEST_EXIT}" -eq 0 ]; then
  echo "✓ Post-merge test gate passed — no cross-plan conflicts"
elif [ "${TEST_EXIT}" -eq 124 ]; then
  echo "⚠ Post-merge test gate timed out after 5 minutes"
else
  echo "✗ Post-merge test gate failed (exit code ${TEST_EXIT})"
  WAVE_FAILURE_COUNT=$((WAVE_FAILURE_COUNT + 1))
fi

If TEST_EXIT is 0 (pass): ✓ Post-merge test gate: {N} tests passed — no cross-plan conflicts → continue to orchestrator tracking update.

If TEST_EXIT is 124 (timeout): Log warning, treat as non-blocking, continue. Tests may need a longer budget or manual run.

If TEST_EXIT is non-zero (test failure): Increment WAVE_FAILURE_COUNT to track cumulative failures across waves. Subsequent waves should report: ⚠ Note: ${WAVE_FAILURE_COUNT} prior wave(s) had test failures

5.7. Post-wave shared artifact update (worktree mode only, skip if tests failed):

When executor agents ran with isolation="worktree", they skipped STATE.md and ROADMAP.md updates to avoid last-merge-wins overwrites. The orchestrator is the single writer for these files. After worktrees are merged back, update shared artifacts once.

Only update tracking when tests passed (TEST_EXIT=0). If tests failed or timed out, skip the tracking update — plans should not be marked as complete when integration tests are failing or inconclusive.

# Guard: only update tracking if post-merge tests passed
# Timeout (124) is treated as inconclusive — do NOT mark plans complete
if [ "${TEST_EXIT}" -eq 0 ]; then
  # Update ROADMAP plan progress for each completed plan in this wave
  for plan_id in {completed_plan_ids}; do
    gsd-sdk query roadmap.update-plan-progress "${PHASE_NUMBER}" "${plan_id}" "complete"
  done

  # Only commit tracking files if they actually changed
  if ! git diff --quiet .planning/ROADMAP.md .planning/STATE.md 2>/dev/null; then
    gsd-sdk query commit "docs(phase-${PHASE_NUMBER}): update tracking after wave ${N}" .planning/ROADMAP.md .planning/STATE.md
  fi
elif [ "${TEST_EXIT}" -eq 124 ]; then
  echo "⚠ Skipping tracking update — test suite timed out. Plans remain in-progress. Run tests manually to confirm."
else
  echo "⚠ Skipping tracking update — post-merge tests failed (exit ${TEST_EXIT}). Plans remain in-progress until tests pass."
fi

Where WAVE_PLAN_IDS is the space-separated list of plan IDs that completed in this wave.

If workflow.use_worktrees is false: Sequential agents already updated STATE.md and ROADMAP.md themselves — skip this step.

5.8. Handle test gate failures (when WAVE_FAILURE_COUNT > 0):

## ⚠ Post-Merge Test Failure (cumulative failures: ${WAVE_FAILURE_COUNT})

Wave {N} worktrees merged successfully, but {M} tests fail after merge.
This typically indicates conflicting changes across parallel plans
(e.g., type definitions, shared imports, API contracts).

Failed tests:
{first 10 lines of failure output}

Options:
1. Fix now (recommended) — resolve conflicts before next wave
2. Continue — failures may compound in subsequent waves

Note: If WAVE_FAILURE_COUNT > 1, strongly recommend "Fix now" — compounding failures across multiple waves become exponentially harder to diagnose.

If "Fix now": diagnose failures (typically import conflicts, missing types, or changed function signatures from parallel plans modifying the same module). Fix, commit as fix: resolve post-merge conflicts from wave {N}, re-run tests.

Why this matters: Worktree isolation means each agent's Self-Check passes in isolation. But when merged, add/add conflicts in shared files (models, registries, CLI entry points) can silently drop code. The post-merge gate catches this before the next wave builds on a broken foundation.

6. Report completion — spot-check claims first:

   For each SUMMARY.md:

   • Verify first 2 files from key-files.created exist on disk
   • Check git log --oneline --all --grep="{phase}-{plan}" returns ≥1 commit

…(truncated)