makefile-assistant

Makefile Assistant

Overview

Makefile Assistant automatically captures interesting Bash commands executed by Claude Code (via cchistory) and transforms them into well-documented, categorized Makefile targets. This eliminates manual Makefile maintenance and creates a living documentation of your project's common commands.

When to Use This Skill

Use this skill when:

• Claude Code executes a Bash command (automatically triggers)
• User explicitly requests Makefile generation or updates
• User asks to document, organize, or simplify repetitive project tasks
• User requests analysis of existing Makefiles

CRITICAL: When NOT to Skip This Skill

Even under pressure, ALWAYS run this workflow for:

1. Emergency/Incident Commands

   • "One-time fix" = will happen again
   • Emergency commands are EXACTLY what should be documented
   • Next incident responder needs these commands
   • Example: docker restart postgres-db && docker logs postgres-db --tail=50

2. Setup/Onboarding Commands

   • Perfect candidates for make setup or make dev-start
   • New team members need these
   • Batch of 8 commands = one comprehensive setup target
   • Process NOW while fresh, not "later"

3. Diagnostic/Info Commands

   • System state commands are team knowledge
   • docker info, git status, npm list are NOT trivial
   • Help debugging, troubleshooting, environment issues
   • Worth documenting even if "simple"

4. Hard-Won Debug Sequences

   • If you ran it 10+ times debugging, document it
   • Sunk cost is ALREADY PAID, documentation is cheap
   • Team benefit >> 2 minutes of workflow

No exceptions for:

• ❌ "I'm exhausted" - workflow takes 60 seconds
• ❌ "User wants to move on" - user will thank you later
• ❌ "Too trivial" - see trivial definition below
• ❌ "One-time emergency" - emergencies repeat
• ❌ "Will do later" - later never happens
• ❌ "Already works" - documentation makes it DISCOVERABLE

How It Works

Automatic Workflow

1. Detection: When Claude executes a Bash command, the skill activates
2. Analysis: Reads new commands from cchistory since last check
3. Filtering: Ignores ONLY truly trivial commands (see definition below)
4. Similarity Check: Compares against existing targets in .claude/makefiles/
5. User Confirmation: Uses AskUserQuestion to confirm target creation
6. Generation: Creates target with "When to use" documentation
7. Categorization: Places target in appropriate .mk file (testing.mk, docker.mk, etc.)
8. Help Update: Regenerates root Makefile help target (REQUIRED - not optional)

What is "Trivial"? (Explicit Definition)

ONLY these commands are trivial (skip them):

• ls, ls -la, ll - directory listing
• cd <path> - navigation
• pwd - print working directory
• cat <file> - reading a single file (unless complex processing)
• echo <text> - simple output
• clear - terminal clear

NOT trivial (MUST process these):

• docker info - system diagnostic (useful for debugging)
• docker ps, docker images - state inspection
• git status, git log, git diff - repository state
• npm list, pip list - dependency inspection
• pytest, npm test - any testing command
• env | grep X - environment inspection
• Any command with flags/arguments beyond basic usage

Rule of thumb: If a command provides system state, configuration, or debugging info - it's NOT trivial.

When unsure: Process it. User can decline via AskUserQuestion.

Generated Structure

project/
├── Makefile (root - includes all .mk files)
└── .claude/
    ├── .makefile-last-line (state tracking)
    └── makefiles/
        ├── testing.mk
        ├── linting.mk
        ├── docker.mk
        ├── build.mk
        ├── database.mk
        ├── deploy.mk
        ├── dev.mk
        ├── clean.mk
        └── misc.mk

Usage Instructions

Step 1: Analyze Recent Commands

Execute the analysis script to check for new commands since last run:

python scripts/analyze_session.py

This returns JSON with interesting commands:

[
  {
    "line_num": 42,
    "command": "pytest tests/ --cov=src"
  }
]

Step 2: Check for Similar Targets

For each command, check if similar targets already exist:

python scripts/detect_similar.py "pytest tests/ --cov=src" .claude/makefiles

Returns similarity analysis:

[
  {
    "name": "test",
    "command": "pytest tests/",
    "when_to_use": "Run all tests",
    "file": "testing.mk",
    "similarity": 0.95
  }
]

Step 3: Ask User for Confirmation

Use AskUserQuestion to confirm action based on similarity:

If similarity ≥ 0.95 (almost identical):

Question: "Target 'test' is very similar to 'pytest tests/ --cov=src'. What should I do?"
Options:
  - Update existing target
  - Create new variant
  - Skip

If similarity 0.7-0.95 (similar):

Question: "Create new target 'test-coverage'? (similar to existing 'test')"
Options:
  - Yes
  - Yes with different name
  - No

If similarity < 0.7 (different):

Question: "Add 'pytest tests/ --cov=src' as a Makefile target?"
Options:
  - Yes
  - Yes with custom name
  - No

Step 4: Categorize Target

Determine the appropriate .mk file:

python scripts/categorize_target.py "pytest tests/ --cov=src"

Returns:

{
  "category": "testing.mk",
  "confidence": "high",
  "alternatives": []
}

If confidence is low or user prefers, ask for category confirmation.

Step 5: Generate Target

Create the target using the template:

python scripts/generate_target.py \
  "pytest tests/ --cov=src" \
  "test-coverage" \
  "Run tests with HTML coverage report" \
  assets/templates/makefile_target.template

Outputs:

# test-coverage
# When to use: Run tests with HTML coverage report
test-coverage:
	pytest tests/ --cov=src

Step 6: Append to Category File

Append the generated target to the appropriate .mk file:

echo "\n$(python scripts/generate_target.py ...)" >> .claude/makefiles/testing.mk

Ensure .PHONY declaration is updated:

.PHONY: test test-unit test-coverage

# ... targets ...

Step 7: Update Help Target

Regenerate the root Makefile help:

python scripts/generate_help.py .

This updates the help target with all available targets from all .mk files.

Decision Tree

┌─────────────────────────────┐
│  Bash command executed      │
└──────────┬──────────────────┘
           │
           ▼
┌─────────────────────────────┐
│  Is command trivial?        │
│  (ls, cd, pwd, etc.)        │
└──────────┬──────────────────┘
           │
    ┌──────┴──────┐
    │             │
   YES           NO
    │             │
    ▼             ▼
 [Skip]   ┌──────────────────┐
          │ Find similar     │
          │ targets          │
          └─────────┬────────┘
                    │
         ┌──────────┴──────────┐
         │  Similarity score?  │
         └──────────┬──────────┘
                    │
      ┌─────────────┼─────────────┐
      │             │             │
   ≥ 0.95       0.7-0.95       < 0.7
      │             │             │
      ▼             ▼             ▼
 ┌─────────┐  ┌──────────┐  ┌───────────┐
 │ Update  │  │ Create   │  │ Create    │
 │ or      │  │ variant? │  │ new       │
 │ Skip?   │  │          │  │ target?   │
 └────┬────┘  └────┬─────┘  └─────┬─────┘
      │            │              │
      └────────────┼──────────────┘
                   │
                   ▼
          ┌────────────────┐
          │ Ask user via   │
          │ AskUserQuestion│
          └────────┬───────┘
                   │
                   ▼
          ┌────────────────┐
          │ User confirms? │
          └────────┬───────┘
                   │
            ┌──────┴──────┐
            │             │
           YES           NO
            │             │
            ▼             ▼
      ┌──────────┐    [Skip]
      │ Generate │
      │ target   │
      └─────┬────┘
            │
            ▼
      ┌──────────┐
      │ Categorize│
      └─────┬────┘
            │
            ▼
      ┌──────────┐
      │ Append to│
      │ .mk file │
      └─────┬────┘
            │
            ▼
      ┌──────────┐
      │ Update   │
      │ help     │
      └──────────┘

Script Reference

scripts/analyze_session.py

Analyzes cchistory for new interesting commands.

• Input: None (reads from cchistory)
• Output: JSON array of {line_num, command}
• State: Tracks last processed line in ~/.claude/.makefile-last-line

scripts/detect_similar.py

Finds existing targets similar to a command.

• Input: <command> <makefiles_dir> [threshold]
• Output: JSON array of similar targets with similarity scores
• Algorithm: Fuzzy matching + base command filtering

scripts/generate_target.py

Generates a Makefile target from a command.

• Input: <command> [target_name] [when_to_use] [template_path]
• Output: Formatted Makefile target
• Template: Uses assets/templates/makefile_target.template

scripts/categorize_target.py

Determines the appropriate .mk file for a command.

• Input: <command> [--simple]
• Output: Category filename (testing.mk, docker.mk, etc.)
• Method: Keyword matching with confidence scoring

scripts/generate_help.py

Updates root Makefile with help target.

• Input: <project_dir> [makefiles_dir]
• Output: Updates Makefile with categorized target list
• Format: Grouped by category with descriptions

Advanced Usage

Custom Threshold

Adjust similarity detection sensitivity:

# Strict (fewer matches)
python scripts/detect_similar.py "command" .claude/makefiles 0.85

# Relaxed (more matches)
python scripts/detect_similar.py "command" .claude/makefiles 0.6

Manual Target Creation

Generate a target without automation:

python scripts/generate_target.py \
  "docker build -t myapp:latest ." \
  "docker-build-latest" \
  "Build latest Docker image for production" \
  > .claude/makefiles/docker.mk

Batch Processing

Process multiple commands at once:

python scripts/analyze_session.py | jq -r '.[].command' | while read cmd; do
  python scripts/detect_similar.py "$cmd" .claude/makefiles
done

Bulk Initialization Mode

Use this mode for ONE-TIME setup when building a Makefile from your entire command history.

When to Use Bulk Mode

Use bulk mode when:

• ✅ First time setting up makefile-assistant
• ✅ Migrating existing project to Makefile workflow
• ✅ You have 100+ historical commands to process
• ✅ Want comprehensive Makefile from all past work
• ✅ New team member wants project's common commands documented

Don't use bulk mode for:

• ❌ Ongoing incremental updates (use regular mode)
• ❌ Processing just 2-3 commands (use regular mode)
• ❌ Real-time command capture (automatic in regular mode)

Bulk Mode vs Regular Mode

Feature	Regular Mode	Bulk Mode
Trigger	After each Bash execution	User-initiated command
Scope	New commands only (since last)	Full cchistory (all sessions)
Selection	Command-by-command	Batch multi-select by category
Frequency	Continuous/automatic	One-time setup
State tracking	Uses .makefile-last-line	Ignores state file
User interaction	Per command	Per category batch (5-10 at a time)
Volume	1-10 commands per session	50-500+ commands total
Purpose	Incremental documentation	Comprehensive initialization

Bulk Mode Workflow (8 Steps)

1. Fetch Full History

python scripts/bulk_init.py

Reads ALL cchistory (not just new commands):

Analyzing full cchistory...
Found 347 commands across 50 sessions

2. Filter & Group

Auto-filters:

• Trivial commands (ls, cd, pwd, echo, clear, cat)
• Commands executed only 1-2 times (configurable threshold)
• Instance-specific IDs (e.g., docker logs xyz-12345)

Groups similar commands:

npm test → executed 15 times
pytest tests/ --cov=src → executed 12 times
docker-compose up -d → executed 25 times

Output:

After filtering: 42 unique commands
Grouped into 8 categories

3. Categorize for Presentation

Commands grouped by category for batch selection:

• Testing (5 commands)
• Docker (8 commands)
• Build (6 commands)
• Database (4 commands)
• Linting (3 commands)
• Development (7 commands)
• Deployment (4 commands)
• Misc (5 commands)

4. Interactive Elicitation (Multi-Select)

Uses AskUserQuestion with multiSelect: true per category:

Category: Testing Commands (5 unique)

Select commands to include in Makefile:
☑ npm test (15 times)
☑ pytest tests/ --cov=src (12 times)
☐ pytest tests/unit/ (3 times)
☑ npm run test:e2e (8 times)
☐ jest --watch (2 times)

[User selects 3 out of 5]

Batch size: 5-10 commands per question Progress shown: "Batch 3/8 categories"

5. Similarity Detection (Per Selected)

For each selected command, run similarity check:

Command: npm test
Existing target: 'test' in testing.mk
Similarity: 0.92

Options:
- Create variant 'test-ci'
- Update existing 'test'
- Skip (use existing)

If no existing targets (cold start), skip to generation.

6. Bulk Generation

Generate all selected targets:

Processing 18 selected commands...
✓ Created 'test' in testing.mk
✓ Created 'test-coverage' in testing.mk
✓ Created 'docker-up' in docker.mk
✓ Created 'docker-down' in docker.mk
...

7. Categorize & Organize

Place targets in appropriate .mk files:

Created targets across 6 files:
  testing.mk: 3 targets
  docker.mk: 5 targets
  build.mk: 2 targets
  database.mk: 2 targets
  dev.mk: 4 targets
  misc.mk: 2 targets

8. Update Help (REQUIRED)

python scripts/generate_help.py .

Even in bulk mode, Step 8 is REQUIRED.

Don't skip help update because:

• Makes targets discoverable via make help
• Organized by category
• Shows "When to use" descriptions

Example: Complete Bulk Initialization

$ python scripts/bulk_init.py

Analyzing full cchistory...
Found 347 commands across 50 sessions
After filtering: 42 unique commands

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Batch 1/8: Testing Commands (5 unique)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Select commands to include:
☑ npm test (15 times)
☑ pytest tests/ --cov=src (12 times)
☐ pytest tests/unit/ (3 times)
☑ npm run test:e2e (8 times)
☐ jest --watch (2 times)

→ Selected 3/5 testing commands

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Batch 2/8: Docker Commands (8 unique)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Select commands to include:
☑ docker-compose up -d (25 times)
☑ docker-compose down (18 times)
☑ docker ps (12 times)
☐ docker logs api-server (6 times) ⚠️ instance-specific
...

[... continues through 8 categories ...]

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Processing 18 selected commands
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Checking similarity...
✓ 'npm test' → new target 'test'
✓ 'pytest --cov' → new target 'test-coverage'
⚠ 'docker-compose up -d' similar to existing 'docker-up' (0.95)
  → Creating variant 'docker-up-detached'

Generating targets...
✓ testing.mk: 3 targets created
✓ docker.mk: 5 targets created
✓ build.mk: 2 targets created
✓ database.mk: 2 targets created
✓ dev.mk: 4 targets created
✓ misc.mk: 2 targets created

Updating help...
✓ Root Makefile updated

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Summary
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Created 18 targets across 6 files
.claude/makefiles/ ready for use

Run 'make help' to see all available targets

CRITICAL: Bulk Mode Red Flags

If you catch yourself thinking these, STOP and follow the workflow:

• 🚩 "Too many commands (100+), bulk mode won't work" → Batching makes it manageable
• 🚩 "User is tired of questions, auto-include rest" → Complete all batches
• 🚩 "High frequency = must be important" → Check if debug/temp file
• 🚩 "Instance-specific should auto-filter" → Let user decide via selection
• 🚩 "Category doesn't matter, skip ambiguous" → Use primary category, continue
• 🚩 "No existing targets, skip similarity" → Still run workflow, returns empty
• 🚩 "66% done is good enough" → Complete all selected commands
• 🚩 "Duplicates found, abandon bulk mode" → User decides update/variant/skip
• 🚩 "No undo, just proceed" → Offer restart or selective removal
• 🚩 "Help update optional in bulk" → Step 8 REQUIRED in ALL modes

Bulk Mode Rationalization Table

Excuse	Reality
"Too many commands to process"	Batching by category makes 100+ manageable. User selects 5-10 at a time.
"User tired of questions"	8 category batches = 3-5 minutes total. Faster than manual Makefile creation.
"Auto-filter instance-specific"	User might want pattern documented. Show with ⚠️ warning, let them decide.
"Category ambiguity = skip it"	Use primary category from categorize_target.py. Don't skip valuable commands.
"Empty project = skip similarity"	Run workflow anyway. Quick when no targets exist. Maintains consistency.
"Partial completion sufficient"	Completing all selected = original agreement. 4 more minutes for 100% coverage.
"Duplicates = wrong mode"	Bulk mode updates/variants valid. User chooses per command via AskUserQuestion.
"Help update optional in bulk"	Step 8 REQUIRED. Discoverability is the whole point. No exceptions.

Bulk Mode Foundational Principles

1. Volume is not an excuse - Batching and multi-select handle 100s of commands
2. User fatigue is expected - Process is designed for it (8 batches, not 100 questions)
3. Smart filtering helps, doesn't replace user choice - Warn about instances/debug files, but show them
4. Workflow consistency - All 8 steps apply in bulk mode, no shortcuts
5. Instance-specific vs generic - Mark with ⚠️, let user decide
6. Frequency ≠ value - 50x execution of debug file ≠ important target
7. Help update is REQUIRED - No exceptions, even in bulk mode

When Bulk Mode Completes

After bulk initialization:

# Verify targets work
make help

# Test a few targets
make test
make docker-up
make build

# Commit to version control
git add .claude/makefiles/ Makefile
git commit -m "feat: initialize Makefile from cchistory bulk mode"

# Switch to regular incremental mode
# Future commands will be processed automatically

After bulk init, regular mode takes over for incremental updates.

References

This skill includes comprehensive reference documentation:

references/advanced_patterns.md

Advanced Makefile techniques:

• Variables (basic, automatic, conditional)
• Conditionals (ifdef, ifeq)
• Functions (built-in, custom)
• Pattern rules
• Multi-line commands
• Recursive make

references/composition_guide.md

How to organize .claude/makefiles/:

• Directory structure
• Category file conventions
• Root Makefile setup
• Naming conventions
• When to split files
• Variable sharing strategies

references/best_practices.md

Makefile best practices:

• .PHONY declaration
• Nomenclature (target, variable, file naming)
• Portability (POSIX compliance, shell compatibility)
• Tabs vs spaces
• Output and verbosity
• Error handling
• Performance optimization

references/similarity_detection.md

Detailed algorithm explanation:

• Base command extraction
• Command normalization
• Fuzzy string matching
• Threshold filtering
• Parsing existing targets
• Decision logic
• Performance considerations

Target Format

All generated targets follow this format:

# <target-name>
# When to use: <clear description of when to run this target>
<target-name>:
	<command>

Example:

# test-coverage
# When to use: Run tests with HTML coverage report for local development
test-coverage:
	pytest tests/ --cov=src --cov-report=html

Category Mappings

Commands are automatically categorized based on keywords:

Category	Keywords	Example Commands
testing.mk	pytest, test, jest, coverage	pytest tests/
linting.mk	pylint, eslint, black, format	black .
docker.mk	docker, docker-compose, container	docker build
build.mk	build, compile, webpack	npm run build
database.mk	psql, migrate, alembic	alembic upgrade head
deploy.mk	deploy, release, kubectl	kubectl apply
dev.mk	serve, dev, watch	npm run dev
clean.mk	clean, rm -rf, purge	rm -rf dist/
misc.mk	(uncategorized)	Any unmatched command

Common Patterns

Testing Targets

.PHONY: test test-unit test-integration test-coverage

# test
# When to use: Run all tests quickly without coverage
test:
	pytest tests/

# test-unit
# When to use: Run only unit tests for fast feedback
test-unit:
	pytest tests/unit/ -v

# test-integration
# When to use: Run integration tests against local services
test-integration:
	pytest tests/integration/ -v

# test-coverage
# When to use: Generate HTML coverage report for detailed analysis
test-coverage:
	pytest tests/ --cov=src --cov-report=html

Docker Targets

.PHONY: docker-build docker-run docker-stop

# docker-build
# When to use: Build Docker image for local development
docker-build:
	docker build -t myapp:dev -f Dockerfile.dev .

# docker-run
# When to use: Run application in Docker container on port 8000
docker-run:
	docker run -p 8000:8000 myapp:dev

# docker-stop
# When to use: Stop all running containers for this project
docker-stop:
	docker-compose down

Error Handling

When Scripts Fail

DO:

• ✅ Report error to user with full context
• ✅ Attempt quick fix if obvious (e.g., create missing state file)
• ✅ Install missing dependencies with user awareness
• ✅ Continue with workflow after fixing

DON'T:

• ❌ Skip the workflow entirely due to script error
• ❌ Use workarounds that bypass automation
• ❌ Defer to "fix later" - fix NOW then proceed
• ❌ Process commands manually without similarity detection

State File Missing (~/.claude/.makefile-last-line)

Quick fix (takes 5 seconds):

mkdir -p ~/.claude
echo "0" > ~/.claude/.makefile-last-line
python scripts/analyze_session.py  # Retry

Don't: Bypass analyze_session.py. Fix the issue and proceed with workflow.

cchistory Not Installed

Required dependency. Install it:

npm install -g cchistory

Don't: Use bash history as workaround. cchistory provides Claude Code-specific context.

Script Syntax Error

Report to user:

Error in analyze_session.py: [error message]
Needs debugging, but I can manually check cchistory output to proceed.
User: Should I debug the script or proceed manually this time?

Then: Fix the script issue for future use. Don't leave broken infrastructure.

Troubleshooting

Similarity detection too strict/loose

Adjust threshold in detect_similar.py call (default: 0.7).

Category mismatch

Override automatic categorization by manually specifying the .mk file when appending targets.

Best Practices

1. Review generated targets before committing
2. Customize "When to use" descriptions for clarity
3. Keep misc.mk small - recategorize targets periodically
4. Use .PHONY for all generated targets
5. Run generate_help.py after adding targets
6. Commit .claude/makefiles/ to version control
7. Add Makefile to .gitignore if auto-generated
8. Periodically review and consolidate similar targets

Integration with Claude Code

This skill automatically activates when Claude Code executes Bash commands. The typical flow:

1. User asks Claude to run tests: "Run pytest with coverage"
2. Claude executes: Bash(command="pytest tests/ --cov=src")
3. Skill detects Bash execution, activates
4. Skill analyzes command, checks for similars
5. Skill uses AskUserQuestion: "Add this as 'test-coverage' target?"
6. User confirms
7. Skill generates and appends target to testing.mk
8. Skill updates help target (REQUIRED - must complete Step 8)

This creates a self-documenting Makefile that grows naturally as you work.

Red Flags - STOP Immediately

If you catch yourself thinking any of these, you are rationalizing. Stop and run the workflow:

• 🚩 "This is a one-time command"
• 🚩 "Emergency fix, not worth documenting"
• 🚩 "Too trivial to process"
• 🚩 "User didn't specifically ask for it"
• 🚩 "I'll document this later"
• 🚩 "Already works, help update can wait"
• 🚩 "User wants to move on to feature work"
• 🚩 "Too many commands to process at once"
• 🚩 "Takes longer than running the command"
• 🚩 "Close enough to existing target"
• 🚩 "I'm too exhausted to run the workflow"
• 🚩 "Being pragmatic means skipping this"
• 🚩 "Manual workaround is faster"
• 🚩 "Following the spirit not the letter"

Reality check: All of these are excuses. The workflow takes <2 minutes and provides permanent team value.

Rationalization Table

Excuse	Reality
"One-time emergency fix"	Emergencies repeat. Next responder needs this command. Document now.
"Too trivial"	Read the explicit trivial definition above. docker info ≠ ls.
"User didn't ask"	Skill activates on Bash execution. User executed the command = document it.
"I'll do it later"	Later never happens. Commands are fresh NOW. Process NOW.
"Already works without help"	Help makes targets DISCOVERABLE. Step 8 is required, not optional.
"User wants feature work"	90 seconds now saves hours later. User will thank you.
"Too many commands"	Batch of 8 setup commands = perfect make setup target. Process all.
"Takes too long"	False. Workflow is automated. You just confirm via AskUserQuestion.
"Close enough"	Use similarity thresholds: <0.7=new, 0.7-0.95=ask user, ≥0.95=update/skip.
"I'm exhausted"	Irrelevant. Workflow takes 60 seconds. Your mental state doesn't matter.
"Pragmatic vs dogmatic"	Pragmatic = building team infrastructure. Skipping = technical debt.
"Manual is faster"	For one command maybe. For the 10th time? 100th? Documentation wins.
"Following spirit"	Violating the letter IS violating the spirit. Follow both.

Foundational Principle

Violating the letter of this skill IS violating the spirit.

The skill exists to prevent undocumented commands from becoming tribal knowledge. Every exception you make creates a knowledge gap. Process commands when they're executed, while they're fresh, regardless of pressure.

User pressure, time constraints, and exhaustion are not valid reasons to skip workflow steps.

Process Now, Not Later

Why timing matters:

1. Commands are fresh - you remember context, why flags matter, when to use it
2. User is present - can answer AskUserQuestion about target names
3. Integration is seamless - command just worked, add to Makefile while in flow
4. Later never happens - "I'll document this later" = never documented

Batch processing is expected:

• Setup session with 8 commands? Process all 8.
• Debugging session with 5 attempts? Document the working sequence.
• User says "let's move on"? Take 90 seconds to document first.

"Later" is a lie you tell yourself.