Generate Codebook Agent Skill

Generate Codebook Skill

You help a medical researcher turn a raw tabular dataset into a structured,

citable data dictionary (codebook). This is the *generator* side of the

dictionary-first workflow: it produces the artifact that /define-variables and

dictionary-first QC later consume. You generate code and review output — you do

not invent the meaning of coded values.

Communication Rules

• Communicate with the user in their preferred language.
• Variable names, codebook fields, and report output are in English.
• Medical terminology is always in English.

Philosophy

A codebook describes *what is in the data*, not *what the codes mean*. Column

distributions, types, and missingness are observable and safe to profile. The

meaning of a coded value (fatty_liver_grade = 0) is NOT observable from the

data — it lives in the authoritative data dictionary. This skill profiles the

former deterministically and explicitly flags the latter as [NEEDS DICTIONARY]

so a human fills it from the source. This is the generator counterpart to the

dictionary-first rule that /define-variables enforces on consumption.

Reference Files

• Schema + role rules: ${CLAUDE_SKILL_DIR}/references/codebook_schema.md — the

codebook.json schema, the role-inference heuristics, and how the output threads

into /define-variables and dictionary-first QC. Read this before interpreting output.

Deterministic Script

Run the bundled profiler rather than describing columns from memory:

python "${CLAUDE_SKILL_DIR}/scripts/generate_codebook.py" data.csv --out-dir .

Supports .csv/.tsv/.xlsx/.parquet/.dta/.sas7bdat. Flags: --max-levels N

(categorical cutoff, default 20), --json-only, --md-only. The script is

pandas-only, runs locally, and never sends data anywhere.

Workflow

Step 1: Profile (deterministic)

Run generate_codebook.py on the dataset. It writes codebook.json (machine-

readable) and codebook.md (review table), reporting per variable: role

(id / continuous / categorical / binary / date / text), dtype, missingness,

unique count, level frequencies or quantile summary, and a needs_dictionary flag.

Step 2: Review with the researcher (gate)

Present codebook.md and walk the user through it. Gate: the user confirms

the inferred roles (e.g., an integer-coded scale mis-read as continuous, or an id

column). Do not proceed to definition work until the user approves the role

assignments.

Step 3: Resolve [NEEDS DICTIONARY] items (gate)

For every variable flagged needs_dictionary: true, the level codes are

uninterpretable without the authoritative source. Gate: ask the user to

supply the meaning of each code from the real data dictionary (file/sheet/row),

or to confirm none exists. Fill label, units, and per-level meanings into the

codebook only from that source — never from inference. If the user cannot

supply it, leave the [NEEDS DICTIONARY] marker in place; do not erase it.

Step 4: Hand off

The completed codebook.json becomes the input dictionary for /define-variables

(operationalization) and the citation source for dictionary-first QC. Gate:

confirm with the user that no needs_dictionary flags remain unresolved before

the codebook is treated as authoritative for downstream analysis.

Scope Limitations

Supported

• Tabular files: CSV, TSV, Excel, Parquet, Stata (.dta), SAS (.sas7bdat).
• Per-variable profiling, role inference, missingness, level/range summaries.

NOT Supported

• Inventing or guessing the meaning of coded values (that is [NEEDS DICTIONARY]).
• Cleaning or transforming data — use /clean-data.
• De-identification — use /deidentify before sharing.
• Operationalizing exposure/outcome definitions — use /define-variables (this skill feeds it).

Cross-Skill Integration

• /define-variables consumes codebook.json as its data dictionary input.
• /clean-data profiles + cleans; this skill produces a durable dictionary artifact instead.
• /deidentify should run on the raw data before a codebook is shared externally.

Output Format

codebook.json (schema in references) and codebook.md (review table with a

"Columns requiring dictionary lookup" section). Summarize the counts

(rows, columns, needs_dictionary_count) in chat; do not paste the full JSON.

Worked Example

Input cohort.csv:

patient_id,age,sex,fatty_liver_grade,smoking_status,visit_date
1001,54,1,0,never,2023-01-15
1002,61,2,2,former,2023-02-03

Run:

python "${CLAUDE_SKILL_DIR}/scripts/generate_codebook.py" cohort.csv --out-dir .
# -> {"n_rows": ..., "n_columns": 6, "needs_dictionary_count": 2, "outputs": [...]}

codebook.md (excerpt):

| Variable            | Role        | Missing % | Unique | Needs dictionary |
| `patient_id`        | id          | 0.0       | N      |                  |
| `age`               | continuous  | 0.0       | ...    |                  |
| `sex`               | binary      | 0.0       | 2      | ⚠️ YES           |
| `fatty_liver_grade` | categorical | 0.0       | 5      | ⚠️ YES           |
| `smoking_status`    | categorical | 0.0       | 3      |                  |
| `visit_date`        | date        | 0.0       | ...    |                  |

sex and fatty_liver_grade are flagged because their levels are bare codes

(1/2, 0..4). smoking_status is not flagged — its levels are already

human-readable. The reviewer then:

• Opens the project's authoritative data dictionary.
• Fills sex: 1 = male, 2 = female and fatty_liver_grade: 0 = none … 4 = suspected

into the codebook from that source (citing file > sheet > row).

• Confirms no [NEEDS DICTIONARY] flags remain, then hands codebook.json to

/define-variables.

What the skill must never do: write sex: 1 = male because "that is the

usual coding." If the dictionary is unavailable, the flag stays.

Anti-Hallucination

• Never invent a variable's label, units, or the meaning of any coded level.
• Coded categorical/binary columns with bare codes are flagged [NEEDS DICTIONARY];

the meaning is filled only from the authoritative data dictionary, then cited.

• Role inference is a heuristic — surface it for user confirmation, do not assert it as ground truth.
• The profiler reads values locally; no data is sent to any model or network.