Skip to content
Back to skills

Just

ASecurity

"Just answer" mode. Use when a message starts with /just, or the user says "just tell me", "don't do it, tell me", "answer only", "audit mode". Answer literally and only the question asked; no edits, no side-effecting commands, no pushes, no "while I was at it", no next-step offers.

  • 6 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 19, 2026
ai-agentspythongobashgitdatabase

Security analysis

A100/100

Scanned September 19, 2026

npx -y skills add Recluse/HyperMnesia --skill just --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Just?

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

Security grade badge for Just
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/recluse-just/badge)](https://www.skillsdirectory.com/skills/recluse-just)

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: just
description: "Just answer" mode. Use when a message starts with /just, or the user says "just tell me", "don't do it, tell me", "answer only", "audit mode". Answer literally and only the question asked; no edits, no side-effecting commands, no pushes, no "while I was at it", no next-step offers.
---

# /just — answer, don't act

Agents are trained to be autonomous: ask "could we…" and they go and do it. This skill flips the
mode for one request: **answer literally what was asked, then stop.**

The rule is small, local and **checkable**: the tool log for this turn must contain no writing call.

## Contract

1. **Read, never change.** Allowed: Read, Grep, Glob, safe `git log/diff/show/status`, `ls`,
   `cat`, SELECT queries, web fetch/search. Forbidden: Edit, Write, any command that changes
   files, the database, the cluster, configs or git history; `git push` above all. If answering
   seems to require a change, that is an action, not an answer: **say what would have to be
   done, and don't do it.**
2. **Only the question asked.** "Could we add seven red lines?" → "Yes/no, because…", not the
   lines. No widening: no "also noticed", no "you should also", no list of next steps.
3. **Literally.** Don't infer an unstated intent. If the question is ambiguous enough that the
   answers diverge, ask one clarifying question and stop.
4. **Honest about uncertainty.** If it can't be verified without acting, say so: "not checked,
   because checking requires X."
5. **Short ending.** The answer ends with the answer. No "want me to do it?".

## Audit mode

An audit is valuable precisely because it diagnoses without intervening: look, assess, name —
and don't "fix it on the way", because the fix changes what hasn't been measured yet. `/just`
gives this per request; for a whole session the user says "audit mode", and the contract holds
until lifted ("audit mode off", "you can act now").

## Examples

- `/just will recall break if I delete rerank/.venv?` → "No: recall goes through `mem_ops` to
  Postgres and the embedder; that venv only holds the reranker service's dependencies. The
  service itself — yes, it will fail." (no commands)
- `/just how many map orphans in myrepo?` → run the read-only check — `python ci/freshness.py
  <repo_dir> myrepo`, without `--mark`, which is the flag that writes — and report the number.
  Don't fix.
- `/just fix the glob` → "That's an action, not a question. Not doing it in /just mode; lift the
  mode and I will." (don't fix)

## Checking the contract

After a `/just` answer, the tool log for that turn must contain no Edit/Write, and among Bash
calls nothing but reads. If it does, the contract was broken — that is a bug of the mode, not
"helpful initiative".

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.

Comments

Loading comments…