Dry Run

Problem: Commands make real changes - creating files, spawning agents, modifying state, executing git operations. Testing commands on production workflows risks unintended modifications.

Solution: Add --dry-run flag that:

1. Executes all read operations (for accurate analysis)
2. Simulates all write operations (shows what would change)
3. Previews agent spawns (shows what Task() calls would occur)
4. Reports state changes (shows YAML mutations)
5. Outputs standardized summary (consistent format across commands)

The result: Safe command testing with full visibility into intended effects.

# In command process section
DRY_RUN=false
if echo "$ARGUMENTS" | grep -q -- "--dry-run"; then
  DRY_RUN=true
  # Remove flag from arguments for processing
  ARGUMENTS=$(echo "$ARGUMENTS" | sed 's/--dry-run//g' | xargs)
fi

Checking in command logic:

If DRY_RUN is true:
  - Collect intended operations in simulation log
  - Skip actual Write/Edit tool calls
  - Skip actual Task() spawns (log them instead)
  - Skip actual Bash commands with side effects
  - Execute Read/Grep/Glob normally
  - Output dry-run summary at end

════════════════════════════════════════════════════════════════════════════════
DRY-RUN MODE: No changes will be made
════════════════════════════════════════════════════════════════════════════════

📁 FILES THAT WOULD BE CREATED:
  ✚ specs/004-auth/spec.md (estimated ~150 lines)
  ✚ specs/004-auth/state.yaml (workflow state)
  ✚ specs/004-auth/NOTES.md (session notes)

📝 FILES THAT WOULD BE MODIFIED:
  ✎ specs/004-auth/state.yaml
    - phase: init → spec
    - status: pending → in_progress

🤖 AGENTS THAT WOULD BE SPAWNED:
  1. spec-phase-agent: "Execute spec phase for user authentication"
  2. clarify-phase-agent: "Clarify requirements"
  3. plan-phase-agent: "Generate implementation plan"

🔀 GIT OPERATIONS THAT WOULD OCCUR:
  • git checkout -b feature/004-auth
  • git add specs/004-auth/
  • git commit -m "feat: initialize auth feature workspace"

📊 STATE CHANGES:
  state.yaml:
    phase: spec → plan → tasks → implement
    status: pending → in_progress → completed

════════════════════════════════════════════════════════════════════════════════
DRY-RUN COMPLETE: 0 actual changes made
Run without --dry-run to execute these operations
════════════════════════════════════════════════════════════════════════════════

At command start, check for --dry-run in arguments:

### Step 0: Dry-Run Detection

Check for --dry-run flag:
```bash
DRY_RUN="false"
if [[ "$ARGUMENTS" == *"--dry-run"* ]]; then
  DRY_RUN="true"
  echo "DRY-RUN MODE ENABLED"
fi

If DRY_RUN is true:

• Initialize simulation log array
• Proceed with analysis but skip execution
• Collect all intended operations

**Important**: Remove `--dry-run` from arguments before parsing other flags to avoid conflicts.
</step>

<step number="2">
**Execute reads normally**

Read operations are safe and necessary for accurate simulation:

```markdown
**Safe operations (execute normally in dry-run):**
- Read tool: `Read file_path=...`
- Grep tool: `Grep pattern=...`
- Glob tool: `Glob pattern=...`
- Bash (read-only): `ls`, `cat`, `git status`, `git log`, `test -f`
- WebFetch: Documentation/API lookups
- WebSearch: Research queries

**Why**: Accurate simulation requires understanding current state.

Intercept and log write operations without executing:

**Write operations to simulate:**

**File writes** (Write tool):

Instead of: Write file_path="specs/001/spec.md" content="..." Log: "WOULD CREATE: specs/001/spec.md (~150 lines)"

**File edits** (Edit tool):

Instead of: Edit file_path="state.yaml" old_string="..." new_string="..." Log: "WOULD MODIFY: state.yaml (phase: init → spec)"

**Bash writes** (Bash tool with side effects):

Instead of: Bash command="mkdir -p specs/001" Log: "WOULD EXECUTE: mkdir -p specs/001"

Instead of: Bash command="git commit -m '...'" Log: "WOULD EXECUTE: git commit -m '...'"

Log Task() calls without spawning:

**Agent spawn simulation:**

Instead of:

Task tool call: subagent_type: "spec-phase-agent" description: "Execute spec phase" prompt: "..."

Log:

WOULD SPAWN AGENT: Type: spec-phase-agent Description: Execute spec phase Expected outputs: spec.md, updated state.yaml

**Why not spawn**: Agents perform real work. Dry-run must prevent cascading side effects.

Analyze YAML state mutations:

**State change tracking:**

Read current state.yaml, then log intended mutations:

```yaml
# Current state
phase: init
status: pending

# After command (simulated)
phase: spec
status: in_progress
phases:
  spec: completed

Log as:

STATE CHANGES (state.yaml):
  - phase: init → spec
  - status: pending → in_progress
  - phases.spec: (new) → completed

</step>

<step number="6">
**Output standardized summary**

Format all collected operations:

```markdown
**Summary template:**

At command end, output:

1. **Header** with dry-run banner
2. **Files created** (new files that would be written)
3. **Files modified** (existing files that would change)
4. **Agents spawned** (Task() calls that would execute)
5. **Git operations** (commits, branches, pushes)
6. **State changes** (YAML mutations)
7. **Footer** with execution instructions

Use consistent emoji prefixes:
- 📁 Files created
- 📝 Files modified
- 🤖 Agents spawned
- 🔀 Git operations
- 📊 State changes

All reads provide accurate context for simulation.

All writes are logged with expected effects.

Some operations need execution for accurate simulation while being side-effect-free.

Add to command frontmatter:

---
argument-hint: "[args] [--dry-run]"
---

Add to command process section:

### Step 0: Mode Detection

**Detect dry-run mode:**
```bash
DRY_RUN="false"
CLEAN_ARGS="$ARGUMENTS"
if [[ "$ARGUMENTS" == *"--dry-run"* ]]; then
  DRY_RUN="true"
  CLEAN_ARGS=$(echo "$ARGUMENTS" | sed 's/--dry-run//g' | xargs)
  echo "═══════════════════════════════════════════════════════════════════"
  echo "DRY-RUN MODE: No changes will be made"
  echo "═══════════════════════════════════════════════════════════════════"
fi

Initialize simulation log (if dry-run): If DRY_RUN is true, initialize arrays:

• FILES_CREATED=()
• FILES_MODIFIED=()
• AGENTS_SPAWNED=()
• GIT_OPERATIONS=()
• STATE_CHANGES=()

Add before each operation:
```markdown
If DRY_RUN is true:
  Log operation to appropriate array
  Skip execution
Else:
  Execute operation normally

Add at command end:

### Final Step: Dry-Run Summary (if applicable)

If DRY_RUN is true:
  Output dry-run summary using format from skill
  Exit without actual changes

### Step 0: Mode Detection

**Detect flags:**
```bash
DRY_RUN="false"
DEEP_MODE="false"
CLEAN_ARGS="$ARGUMENTS"

if [[ "$ARGUMENTS" == *"--dry-run"* ]]; then
  DRY_RUN="true"
  CLEAN_ARGS=$(echo "$CLEAN_ARGS" | sed 's/--dry-run//g')
fi

if [[ "$ARGUMENTS" == *"--deep"* ]]; then
  DEEP_MODE="true"
  CLEAN_ARGS=$(echo "$CLEAN_ARGS" | sed 's/--deep//g')
fi

CLEAN_ARGS=$(echo "$CLEAN_ARGS" | xargs)  # Trim whitespace

Step 1.1: Initialize Feature Workspace

If DRY_RUN is true:

Log: "WOULD CREATE: specs/NNN-feature-slug/state.yaml"
Log: "WOULD CREATE: specs/NNN-feature-slug/NOTES.md"
Log: "WOULD EXECUTE: python .spec-flow/scripts/spec-cli.py feature '...'"

Skip actual execution Else: Execute normally

</example_integration>
</command_integration>

<examples>
<example name="feature-dry-run">
**Input**: `/feature "add user dashboard" --dry-run`

**Output**:

════════════════════════════════════════════════════════════════════════════════ DRY-RUN MODE: No changes will be made ════════════════════════════════════════════════════════════════════════════════

📁 FILES THAT WOULD BE CREATED: ✚ specs/005-add-user-dashboard/state.yaml ✚ specs/005-add-user-dashboard/NOTES.md ✚ specs/005-add-user-dashboard/interaction-state.yaml ✚ specs/005-add-user-dashboard/domain-memory.yaml ✚ specs/005-add-user-dashboard/spec.md (after spec-phase-agent) ✚ specs/005-add-user-dashboard/plan.md (after plan-phase-agent) ✚ specs/005-add-user-dashboard/tasks.md (after tasks-phase-agent)

🤖 AGENTS THAT WOULD BE SPAWNED:

1. initializer → Initialize domain memory for feature
2. spec-phase-agent → Generate specification
3. clarify-phase-agent → Resolve ambiguities
4. plan-phase-agent → Create implementation plan
5. tasks-phase-agent → Break down into TDD tasks
6. worker (N times) → Implement each feature atomically
7. optimize-phase-agent → Run quality gates
8. ship-staging-phase-agent → Deploy to staging
9. finalize-phase-agent → Archive and document

🔀 GIT OPERATIONS THAT WOULD OCCUR: • git checkout -b feature/005-add-user-dashboard • git add specs/005-add-user-dashboard/ • git commit (multiple, after each phase) • git push origin feature/005-add-user-dashboard

📊 WORKFLOW PHASES: init → spec → clarify → plan → tasks → implement → optimize → ship → finalize

════════════════════════════════════════════════════════════════════════════════ DRY-RUN COMPLETE: 0 actual changes made Run /feature "add user dashboard" to execute these operations ════════════════════════════════════════════════════════════════════════════════

</example>

<example name="implement-dry-run">
**Input**: `/implement --dry-run`

**Output**:

════════════════════════════════════════════════════════════════════════════════ DRY-RUN MODE: No changes will be made ════════════════════════════════════════════════════════════════════════════════

📋 CURRENT STATE: Feature: specs/004-auth-system/ Phase: implement Tasks remaining: 8 of 12

📝 FILES THAT WOULD BE MODIFIED: ✎ specs/004-auth-system/domain-memory.yaml - features[F001].status: pending → completed - features[F002].status: pending → in_progress ✎ specs/004-auth-system/state.yaml - phase: implement (unchanged during execution) ✎ src/services/auth.ts (implementation) ✎ src/services/auth.test.ts (tests) ✎ src/routes/login.ts (implementation) ✎ src/routes/login.test.ts (tests)

🤖 AGENTS THAT WOULD BE SPAWNED:

1. worker → Implement F001: User login endpoint
2. worker → Implement F002: Session management
3. worker → Implement F003: Password validation
4. worker → Implement F004: Token refresh (... 4 more workers for remaining features)

🔄 TDD CYCLE PER TASK: For each worker: 1. Write failing tests 2. Implement feature 3. Verify tests pass 4. Update domain-memory.yaml 5. Commit changes

════════════════════════════════════════════════════════════════════════════════ DRY-RUN COMPLETE: 0 actual changes made Run /implement to execute 8 remaining tasks ════════════════════════════════════════════════════════════════════════════════

</example>

<example name="ship-dry-run">
**Input**: `/ship --staging --dry-run`

**Output**:

════════════════════════════════════════════════════════════════════════════════ DRY-RUN MODE: No changes will be made ════════════════════════════════════════════════════════════════════════════════

📋 PRE-FLIGHT CHECKS (would execute): ✓ Build verification ✓ Test suite (127 tests) ✓ Type checking ✓ Security scan ✓ Environment variables

🔀 GIT OPERATIONS THAT WOULD OCCUR: • git push origin feature/004-auth-system • gh pr create --base staging --head feature/004-auth-system • Wait for CI to pass • gh pr merge --auto

📦 DEPLOYMENT THAT WOULD OCCUR: • Merge to staging branch triggers deploy-staging.yml • Vercel/Railway builds from staging • Health check endpoints verified • Deployment URL: https://staging.example.com (estimated)

📊 STATE CHANGES: state.yaml: - phases.ship: pending → completed - deployment.staging.status: pending → deployed - deployment.staging.pr_number: (new) → #42

🤖 AGENTS THAT WOULD BE SPAWNED:

1. ship-staging-phase-agent → Execute staging deployment

════════════════════════════════════════════════════════════════════════════════ DRY-RUN COMPLETE: 0 actual changes made Run /ship --staging to deploy to staging environment ════════════════════════════════════════════════════════════════════════════════

</example>

<example name="quick-dry-run">
**Input**: `/quick "fix typo in README" --dry-run`

**Output**:

════════════════════════════════════════════════════════════════════════════════ DRY-RUN MODE: No changes will be made ════════════════════════════════════════════════════════════════════════════════

📋 TASK ANALYSIS: Type: Documentation fix Scope: Single file Estimated changes: <10 lines

📝 FILES THAT WOULD BE MODIFIED: ✎ README.md (typo fix)

🔀 GIT OPERATIONS THAT WOULD OCCUR: • git add README.md • git commit -m "docs: fix typo in README"

🤖 AGENTS THAT WOULD BE SPAWNED:

1. quick-worker → Execute atomic change with tests

════════════════════════════════════════════════════════════════════════════════ DRY-RUN COMPLETE: 0 actual changes made Run /quick "fix typo in README" to execute ════════════════════════════════════════════════════════════════════════════════

</example>
</examples>

<anti_patterns>
<anti_pattern name="executing-writes-in-dry-run">
**Problem**: Accidentally executing write operations in dry-run mode

**Bad approach**:
```markdown
If DRY_RUN:
  echo "Would write file"
Write file...  # WRONG: Executes regardless of dry-run!

Correct approach:

If DRY_RUN:
  Log: "WOULD CREATE: file.md"
Else:
  Write file_path="file.md" content="..."

Rule: Always wrap write operations in dry-run conditional.

Bad approach:

If DRY_RUN:
  echo "Would spawn agent"
Task(spec-phase-agent)  # WRONG: Agent runs and makes changes!

Correct approach:

If DRY_RUN:
  Log: "WOULD SPAWN: spec-phase-agent"
  Log: "Expected outputs: spec.md"
Else:
  Task(spec-phase-agent)

Rule: Never spawn agents in dry-run mode.

Bad approach:

DRY-RUN OUTPUT:
  Would create spec.md
  # Missing: state.yaml changes, git operations, other files

Correct approach:

DRY-RUN OUTPUT:
  FILES CREATED: spec.md, state.yaml, NOTES.md
  FILES MODIFIED: state.yaml (phase: init → spec)
  GIT OPERATIONS: checkout -b, add, commit
  AGENTS: spec-phase-agent, clarify-phase-agent

Rule: Log ALL operations that would occur.

Bad approach:

If DRY_RUN:
  Skip reading state.yaml  # WRONG: Can't accurately simulate!
  Guess at current state

Correct approach:

# Always read current state (safe operation)
STATE=$(cat specs/NNN/state.yaml)
CURRENT_PHASE=$(yq eval '.phase' ...)

If DRY_RUN:
  Log: "Current phase: $CURRENT_PHASE → would transition to: spec"

Rule: Execute reads for accurate simulation context.

1. Zero side effects: No files created/modified, no git operations, no agents spawned
2. Accurate preview: All intended operations listed in output
3. Consistent format: Standard output format across all commands
4. Readable output: Clear emoji prefixes, organized sections
5. Actionable: Output tells user how to execute for real
6. Complete: Shows files, agents, git ops, state changes

• --dry-run flag detected and removed from arguments
• All Write tool calls wrapped in dry-run conditional
• All Edit tool calls wrapped in dry-run conditional
• All Task() calls wrapped in dry-run conditional
• All Bash writes (mkdir, git commit, etc.) wrapped
• Read operations execute normally for context
• Output uses standard format with emoji prefixes
• Output includes all operation categories
• Output shows how to run without dry-run
• No actual changes occur in dry-run mode

Usage: Add --dry-run to any command:

/feature "add auth" --dry-run
/implement --dry-run
/ship --staging --dry-run

1. Flag detection: --dry-run recognized by command
2. Zero execution: No files created, modified, or deleted
3. No agents spawned: Task() calls logged but not executed
4. No git changes: No commits, branches, or pushes
5. Accurate preview: All operations that WOULD occur are listed
6. Standard format: Consistent output format with clear sections
7. Actionable output: User knows how to run command for real
8. Safe exploration: Users can test any command without risk
9. CI/CD ready: Can validate commands in automation pipelines
10. Repeatable: Same dry-run produces same output (deterministic)