flavor

name	Flavor
description	Interactive interview that customizes DRL skill files for field-specific research conventions

Flavor Customization System

Overview

Customize the DRL skill files to match a researcher's field conventions, methodology preferences, journal requirements, and citation style. Flavor edits are version-controlled and reversible.

Step 1: Safety Checkpoint (H2 Mitigation)

Before editing any skill file, you MUST create a safety checkpoint.

1. Stage all current changes: git add .claude/skills/drl/
2. Commit with a descriptive message: git commit -m "checkpoint: pre-flavor skill state"
3. This ensures all flavor changes are reversible via git revert or git diff

If the commit fails because there are no changes, that is fine -- proceed.

Step 2: Interactive Interview

Use AskUserQuestion to gather field-specific configuration. Ask about each dimension:

2a. Research Field

• Primary field: economics, sociology, political science, psychology, public health, education, etc.
• Subfield or specialization (e.g., labor economics, organizational sociology)
• Interdisciplinary considerations

2b. Methodology Preferences

• Primary approach: quantitative, qualitative, mixed methods
• Preferred statistical methods: regression analysis, causal inference (IV, RDD, DiD), structural models, Bayesian methods
• Data analysis conventions: significance thresholds, effect size reporting, confidence intervals vs. p-values
• Robustness expectations: how many alternative specifications, which sensitivity analyses

2c. Journal Target

• Target journal name (if known)
• Journal formatting requirements (section structure, word limits, abstract format)
• Submission requirements (anonymous review, data availability statement, pre-registration)

2d. Citation Style

• Citation format: APA 7th, Chicago Author-Date, Chicago Notes, Harvard, journal-specific
• Bibliography conventions: full author lists vs. et al. threshold, DOI inclusion, URL formatting
• In-text citation style: (Author, Year) vs. Author (Year) vs. numbered

Step 3: Web Search for Field Conventions

Search for the target journal and field conventions:

1. Search for journal-specific author guidelines and submission requirements
2. Search for field-specific statistical reporting standards (e.g., APA reporting standards for psychology)
3. Search for common robustness check expectations in the field
4. Note any field-specific vocabulary or terminology conventions

Record findings to inform skill file edits.

Step 4: Skill File Editing

Read and edit the DRL skill files to incorporate field conventions.

Atomic Write Protocol (H2 Mitigation)

For each skill file edit, use a write-then-rename sequence to prevent partial writes from corrupting skill files:

# 1. Read the current file
cat .claude/skills/drl/<skill>/SKILL.md

# 2. Write modified content to a temporary file
# (use Write tool or redirect to .tmp)
# Result: .claude/skills/drl/<skill>/SKILL.md.tmp

# 3. Atomically replace the original with the temp file
mv .claude/skills/drl/<skill>/SKILL.md.tmp .claude/skills/drl/<skill>/SKILL.md

If step 3 fails, the original file is untouched. If step 2 fails, there is no corruption -- only a leftover .tmp file to clean up.

What Gets Customized

Edit the following skill files in .claude/skills/drl/:

1. spec-dev/SKILL.md: Adapt hypothesis format, literature gap framing, and methodology outline to field conventions
2. plan/SKILL.md: Adjust statistical method recommendations, variable naming conventions, and robustness battery to field norms
3. work/SKILL.md and sub-skills (work-writing/SKILL.md, work-analysis/SKILL.md): Set field-appropriate significance thresholds, effect size measures, and table/figure formatting
4. review/SKILL.md: Calibrate reviewer expectations for the field (e.g., IV validity checks for economics, effect size reporting for psychology)
5. compound/SKILL.md: Adjust paper structure to journal requirements, set abstract format, and configure citation style

Note: The procedural skills (lit-review, decision, compile, status) are intentionally excluded from flavor customization because their workflows are field-agnostic.

Customization Targets Within Each Skill

For each skill file, look for and adapt:

• Vocabulary: Replace generic terms with field-specific terminology
• Method defaults: Set appropriate default statistical methods for the field
• Quality thresholds: Adjust what counts as a sufficient robustness battery
• Output format: Match table and figure conventions to the journal style
• Review criteria: Calibrate what reviewers check based on field expectations

Step 5: Verify and Commit

1. Review all changes: git diff .claude/skills/drl/
2. Verify no skill file was corrupted (each still has valid YAML frontmatter and readable content)
3. Stage the changes: git add .claude/skills/drl/
4. Commit: git commit -m "flavor: configure for <field> targeting <journal>"

Reverting Flavor Changes

If the researcher wants to undo flavor customization:

1. Find the pre-flavor checkpoint: git log --oneline .claude/skills/drl/
2. Revert to the checkpoint commit
3. Or selectively revert individual files: git checkout <commit> -- .claude/skills/drl/<skill>/SKILL.md

Gate Criteria

Before concluding, verify:

• Safety checkpoint committed before any edits
• All four interview dimensions gathered (field, methodology, journal, citation)
• Web search performed for journal and field conventions
• Skill files edited via atomic write protocol
• Changes committed with descriptive message
• No skill file corrupted (YAML frontmatter intact)

Quality Criteria

• Safety checkpoint committed before edits
• All four interview dimensions gathered (field, methodology, journal, citation)
• Web search performed for field conventions
• Atomic write protocol followed for each edit
• YAML frontmatter intact in all edited files
• Changes committed with descriptive message

Common Pitfalls

• Editing skill files without creating a safety checkpoint first
• Applying changes that are too field-specific and break the general workflow
• Not searching for the actual journal requirements (guessing instead)
• Forgetting to update the citation style in the compound skill
• Making vocabulary changes that conflict with beads or DRL workflow terminology