precise-calibrate

name	precise-calibrate
description	Calibrate your language precision preferences. Use when the user wants to set up their style profile, take the precision quiz, or configure how /precise adapts to them. Triggered by "calibrate", "set up my profile", "learn my style".
argument-hint	[{"optional":"path to writing sample"}]
allowed-tools	["Read","Write","Bash"]

/precise-calibrate — style profile builder

you run an adaptive quiz to learn how the user likes things described. the output is a style profile saved to .claude/skills/precise/user-profile.md that /precise reads on every invocation.

mode 1: quiz (default, no arguments)

run exactly 5 rounds. each round:

1. present a scenario (describing a person, a company, an experience, an idea, or a place)
2. show 3 descriptions of the same thing, each in a distinctly different style:
   • one abstract/conceptual
   • one concrete/sensory/specific
   • one structured/analytical vary the exact styles across rounds — don't repeat the same three archetypes.
3. ask: "which one do you like best? (1/2/3)"
4. use the answer to update your model of the user's preferences

what you're inferring (axes)

• abstract ↔ concrete: do they prefer "a liminal space" or "the hallway between the kitchen and the back door where nobody ever turns the light on"?
• emotional ↔ analytical: "it felt like the air had been punched out of the room" vs "morale dropped 40% in Q3 based on pulse survey data"
• terse ↔ elaborate: "sharp. relentless. right." vs a full paragraph
• formal ↔ colloquial: "the organization demonstrated resilience" vs "they just refused to die"
• metaphorical ↔ literal: "a cathedral of code" vs "a well-structured 50k-line codebase with clear module boundaries"

adaptive selection

each question should maximize information gain. if the first two answers both lean concrete, don't ask another concrete-vs-abstract question — probe a different axis (terse vs elaborate, metaphorical vs literal).

after round 5, present the inferred profile to the user for confirmation.

mode 2: writing sample (argument provided)

if the user provides a path to a writing sample via $ARGUMENTS:

1. read the file
2. analyze: sentence length distribution, adjective density, abstraction level, metaphor frequency, formality markers, terse-vs-elaborate ratio
3. infer the same 5-axis profile
4. present to user for confirmation/adjustment

output: user profile

after confirmation, write the profile to .claude/skills/precise/user-profile.md:

---
name: user-precision-profile
description: style preferences for /precise output calibration
type: user
---

## precision style profile

- abstract ↔ concrete: [score 0-1, where 0 = fully abstract, 1 = fully concrete]
- emotional ↔ analytical: [score 0-1]
- terse ↔ elaborate: [score 0-1]
- formal ↔ colloquial: [score 0-1]
- metaphorical ↔ literal: [score 0-1]

## interpretation

[2-3 sentences describing what this profile means in plain language. e.g., "you prefer descriptions that are concrete and terse — you'd rather hear one precise detail than three abstract adjectives. you're comfortable with metaphor when it earns its keep but don't want ornamental flourish."]

## calibration data

[list the quiz answers or writing sample analysis that produced this profile]

hard constraints

• 5 questions max in quiz mode. respect the user's time. if you can infer with high confidence in 3, stop early.
• no psychoanalysis. you're inferring description preferences, not personality. keep it about language.
• the profile is a starting point. tell the user they can re-run /precise-calibrate anytime or manually edit the profile file.
• make the quiz interesting. the descriptions should be genuinely well-written and distinct. the user should enjoy the process, not endure it.