Skip to content
Back to skills

B6p Preferences

ASecurity

See and change your own preferences for the BlueStep (B6P) tools in this project, in one place — how Claude writes its answers to you (short and plain, or complete and technical), when a long conversation gets summarized, and whether the client-data guard is on. Use when the user asks how to change how Claude answers, wants shorter, simpler or more technical answers, asks what they can configure or where a setting lives, or says "preferences", "settings" or "configuration" about these tools. ...

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 11, 2026
ai-agentsgobashgit

Works with

  • claude code
  • cursor
  • cli

Security analysis

A100/100

Scanned November 11, 2026

npx -y skills add Bluestep-Systems/bluestep-ai --skill b6p-preferences --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of B6p Preferences?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for B6p Preferences
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/bluestep-systems-b6p-preferences-bluestep-ai/badge)](https://www.skillsdirectory.com/skills/bluestep-systems-b6p-preferences-bluestep-ai)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: b6p-preferences
description: See and change your own preferences for the BlueStep (B6P) tools in this project, in one place — how Claude writes its answers to you (short and plain, or complete and technical), when a long conversation gets summarized, and whether the client-data guard is on. Use when the user asks how to change how Claude answers, wants shorter, simpler or more technical answers, asks what they can configure or where a setting lives, or says "preferences", "settings" or "configuration" about these tools. Changes only the user's own settings file, never the team's.
allowed-tools: Read Write Edit AskUserQuestion Bash(test:*) Bash(cat:*) Bash(claude:*) Bash(mkdir:*)
---

# /b6p-preferences — your settings for the BlueStep tools, in one place

Some settings are each person's own: how answers read, when a long conversation is summarized. They
live in `.claude/settings.local.json`, which is git-ignored, so a change here never reaches a teammate.
This skill shows them in plain words, says what each one does, and changes the one the user picks.
`/b6p-setup` asks the first one when it brings a project up to date, using the card below.

Everything the user reads is plain: what the setting does and what changes for them, never a key
name unless they ask. Ask with clickable options (`AskUserQuestion` in Claude Code), one card at a
time.

## On Cursor or Codex

Neither has output styles or these settings. Write nothing. Say that the project's `AGENTS.md`
already tells the agent to match the user's level, and where a personal preference goes instead:
on Cursor, a User Rule (Customize → Rules), on Codex, `~/.codex/AGENTS.md`. Offer the line to paste
for shorter answers: "Answer briefly in plain, everyday words; lead with the answer; no file names,
ids or technical terms unless I ask." Then stop.

## Step 1 — Read what is set now (read only)

On Claude Code, read `.claude/settings.local.json`, `.claude/settings.json` and `~/.claude/settings.json`. A missing
file is fine. The value that applies is the first one found in that order.

| Preference | Key | How to show it |
|---|---|---|
| How Claude answers | `outputStyle` | `bluestep-tools:Plain answers` → "Short and plain"; `default` or not set → "Complete and technical"; any other name → that name, as the user's own choice |
| When a long conversation is summarized | `autoCompactWindow` | a number of tokens, shown as "at about 400K" |
| Client-data guard | `phi-guard` plugin | installed or not, and on or off here (see step 3) |

Show one short table, preference · now · what it does, then ask "What do you want to change?" with
the three preferences and **Nothing**.

## Step 2 — How Claude answers

Before the card, say in one line: "This is only for you, in this project, and you can change it
again any time with `/b6p-preferences`." Then ask "How do you want Claude to answer you?" with these
two options, each with its example, in the user's language:

- **Short and plain** (recommended if you don't write code) — the answer first, in everyday words,
  with no file names or technical terms unless you ask. Example: "The list is empty because the unit
  was renamed. Check its name on the platform."
- **Complete and technical** (recommended if you write code) — more detail: files, lines of code and
  commands. Example: "`staffQuery.search` filters on `unitName` (`app.ts:6`); after the rename it
  returns `[]`."

When the current value is another style the user picked (for example `Concise`), add a third
option, **Keep `<name>`**.

On the answer, merge one key into `.claude/settings.local.json` — create the file with just that key
if it is absent, never touch its other keys:

- Short and plain → `"outputStyle": "bluestep-tools:Plain answers"` (exactly this; any other spelling
  matches no style and Claude falls back to none).
- Complete and technical → `"outputStyle": "default"`.
- Keep → write nothing.

Never write this key to Claude Code's `.claude/settings.json`: the team shares that file. Say it applies from the
next session. Both choices keep the project's answering rules (lead with the answer, ask only what
is yours to decide, a last list of what was not checked); the short style pushes further toward
brevity and everyday words.

## Step 3 — The other two

- **When a long conversation is summarized.** Claude re-reads the whole conversation on every
  message; past a point it summarizes the older part so answers stay accurate and cheaper. The team
  default is about 400K tokens, in Claude Code's `.claude/settings.json`. Offer **Keep the team default** or
  **Set my own** (a number from 100000 to 1000000); a personal value goes in
  `.claude/settings.local.json` as `autoCompactWindow` and wins over the team's, for this user only.
- **Client-data guard (`phi-guard`).** It blocks an edit that looks like real client data. Check
  whether it is installed and whether this project turns it off, the way
  `${CLAUDE_PLUGIN_ROOT}/skills/b6p-setup/references/enablement-claude-code.md` says. Not installed →
  offer it and, on a yes, hand over the install from that file; never run it. Turned off here →
  say so: that is a team decision in Claude Code's `.claude/settings.json`, so this skill does not change it.

## Report

One line per change: what changed, for whom (only you, this project) and when it applies (next
session). If nothing changed, say so in one line.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.