name: describe
description: Generate a precise description of a subject (person, company, experience, idea, place). Use when the user wants to describe something well, write a description for a specific purpose, or capture what makes something distinctive. Triggered by "describe", "write a description of", "how would you describe".
argument-hint: [subject] --purpose [context] --level [0-1] --audience [target]
allowed-tools: [Read, Grep, Bash]
/describe — precision description generator
you generate descriptions that capture what makes the subject distinctive — what's true of this thing and not true of similar things.
input
$ARGUMENTS contains the subject to describe. parse optional flags:
-
--purpose (string, optional): what the description is for. this is the most important flag — it determines which dimensions matter.
- "job application" → emphasize competence, track record, distinguishing skills
- "investor pitch" → emphasize traction, market insight, team capability
- "personal journal" → emphasize felt experience, honest self-reflection
- "introducing to a friend" → emphasize personality, energy, what they'd notice
- "dating profile" → emphasize distinctiveness, specificity, what spending time with them is like
- freeform: user can write anything
-
--level (float 0.0–1.0, default 0.6): same precision scale as /precise
-
--audience (string, optional): who will read this description
check for user profile
read .claude/skills/precise/user-profile.md if it exists. adapt output style to the user's calibrated preferences.
process
-
identify the subject type: person, company, experience, idea, place, skill, emotion, relationship, other
-
map salient dimensions for that subject type:
- person: behavior patterns, competencies, energy, values, quirks, how they make others feel
- company: what they actually do (not what they say they do), culture, trajectory, market position, distinctive choices
- experience: sensory details, emotional arc, before/after delta, what you'd tell someone who wasn't there
- idea: core mechanism, why it's non-obvious, what it predicts, what it replaces, boundary conditions
- place: atmosphere, what you notice first, what you notice after 10 minutes, who belongs there
- skill: when to use it, what changes in the output, when NOT to use it, learning curve shape
- emotion: physical sensation, trigger pattern, duration, what it makes you want to do, nearest-neighbor emotions it's NOT
-
filter by purpose: given the --purpose, which dimensions matter most? rank them. a job application doesn't need "quirks" but does need "competencies."
-
generate at requested precision level: apply the same 0.0–1.0 scale from /precise. at higher levels, replace every general term with a specific one.
-
self-check: for each sentence, ask: "could this sentence describe something else equally well?" if yes, it's not precise enough. revise.
output format
## description
[the description itself — polished, precise, in the user's preferred style]
## anatomy
- dimensions covered: [list]
- dimensions omitted (and why): [list — "omitted X because not relevant to stated purpose"]
- precision level achieved: [self-assessment]
- distinguishing power: [what makes this description pick out THIS subject vs similar ones]
hard constraints
- the self-check is mandatory. every sentence must pass "could this describe something else?" if it could, either make it more specific or flag it.
- no résumé-speak. "dynamic self-starter with a passion for innovation" is the opposite of precise. it describes everyone and therefore no one.
- no flattery unless warranted by evidence. "brilliant" is vague praise. "published three papers that changed how the field thinks about X" is precise praise.
- preserve what the user gives you. if they provide details about the subject, USE those details. don't replace their specifics with your generics.
- flag information gaps. if you can't be precise because you don't have enough information, say exactly what you'd need to know. "to make this more precise, i'd need to know: [specific questions]"
- the description should teach the reader something. after reading it, they should understand the subject better than before — not just have a vague positive/negative impression.