Skill · CLAUDE.md rule

If It's Deterministic, Don't Make It Guess

A forcing function that sends exact-answer work to a script instead of the model's reasoning.

TypeCLAUDE.md rule

You ask the agent to count how many records match, or to reformat two hundred entries, or to total a column. It answers — confidently, in prose. It looks right. It's off by three, because a language model didn't count: it estimated, the way you'd estimate a jar of marbles from across the room.

There is a clean line between two kinds of work. One has a single correct answer a machine can compute: counting, parsing a fixed format, transforming data, arithmetic, lookups. The other needs judgement: what to build, how to word it, which trade-off to take. Models are extraordinary at the second and quietly unreliable at the first — so the rule routes each kind to the tool that is actually good at it:

✕ Without the rule

"Count / total / reformat this" — answered in prose

▼

Looks right, sounds certain

▼

Off by three — and you trusted the number

✓ With the rule

Writes a script for the exact-answer part

▼

Runs it, shows the script and the output

▼

The number is the number — not an estimate

Drop it into your CLAUDE.md. It won't slow down the work that genuinely needs reasoning — that's most of it. What it catches is the quiet failure: the confident wrong count, the missed row, the entry silently dropped in a reformat. The tell is structure. If the input has a consistent shape, the answer belongs to a parser, not a paragraph.

Paste into CLAUDE.md

## Deterministic work goes to code, not prose
When a task has exactly one correct answer a script can compute — counting, parsing a fixed format, transforming data, math, lookups — write and run a script instead of producing the answer by hand.
1. If the input has a consistent structure, reach for a parser, regex, or query — not free-form reasoning.
2. Show the script and its real output, not just the final number.
Reserve your reasoning for judgement, generation, and ambiguity — the work that has no single right answer.

Prefer your agent install it?

Add this rule to my CLAUDE.md, verbatim:

## Deterministic work goes to code, not prose
When a task has exactly one correct answer a script can compute — counting, parsing a fixed format, transforming data, math, lookups — write and run a script instead of producing the answer by hand.
1. If the input has a consistent structure, reach for a parser, regex, or query — not free-form reasoning.
2. Show the script and its real output, not just the final number.
Reserve your reasoning for judgement, generation, and ambiguity — the work that has no single right answer.

Did this help?

Tags#economics #claude.md #skill

Other skills

01Verify Before You Call It Done 02Plan First, Then Code 03Map the Blast Radius Before You Edit 05Diagnose From the Symptom, Not the Guess 06Make It Argue Against Its Own Plan 07Hand Off to Your Next Session 08A Review That Doesn't Rubber-Stamp 09Run to the gate, then stop — not step-by-step 10A CLAUDE.md skeleton — turn a description into operating law 11The weekly review prompt — five minutes that turn experience into skill 12Split the work into focused slices, each with clean context 13Is this even worth an agent — ask before you open the chat 14Probe with a small task before you trust it in a new domain

Back to skills

The author

Each story here wraps a lesson paid for in full.

craftagentsomeone building and learning at once

What are you building with agents? Want to trade notes, push back, or build something together — drop a line.

Write hello@craftagent.cloud

58pieces12clustersVI·ENbilingual

✕ Without the rule

✓ With the rule

Get new pieces by email