ask-kit-skills
ask-kit kit: Claude Code skills and agents (subagents) on this machine and in the current repository: which skill or agent covers a job, what one says about a topic, which are installed and from where (user, project, plugin, claude.ai), which plugins the marketplaces offer, whether two overla
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/temet-ai--ask-kit-skills)HTML
<a href="/listings/temet-ai--ask-kit-skills"><img src="/api/listings/temet-ai--ask-kit-skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
You need [ask-kit](https://github.com/temet-ai/ask-kit#set-it-up) and Node.js.
Paste this into your agent:
Read ~/.config/ask-kit/AGENTS.md, then add this kit: github.com/temet-ai/ask-kit-skills
Or run it yourself:
ask-kit kit install temet-ai/ask-kit-skills # shows what it reads, changes and runs ask-kit kit install temet-ai/ask-kit-skills --yes # installs and enables it
Restart your agents so `ask` lists `skills`. Project skills and agents come from the repository your agent works in; a question that names another folder is answered for that folder.
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 77 lines
- # ask-kit-skills
- An [ask-kit](https://github.com/temet-ai/ask-kit) kit that lets your coding agent's `ask` tool
- answer questions about Claude Code skills and agents on your machine, and write, fix, rename and
- retire them when asked. A cheap model reads the skill and agent folders outside your agent's
- context and answers with names and `path:line`.
- Questions from its bench, of the kinds agents dug for by hand in skill and agent folders:
- - Which skill should I use to plan a night patrol route, and where does it live?
- - Does any skill or agent still mention morpork-report, the old name of the city-report skill? Which ones?
- - Is the frontmatter of the broken-skill skill valid? If not, what is wrong and on which line?
- ## Install
- You need [ask-kit](https://github.com/temet-ai/ask-kit#set-it-up) and Node.js.
- Paste this into your agent:
- ```
- Read ~/.config/ask-kit/AGENTS.md, then add this kit: github.com/temet-ai/ask-kit-skills
- ```
- Or run it yourself:
- ```sh
- ask-kit kit install temet-ai/ask-kit-skills # shows what it reads, changes and runs
- ask-kit kit install temet-ai/ask-kit-skills --yes # installs and enables it
- ```
- Restart your agents so `ask` lists `skills`. Project skills and agents come from the repository
- your agent works in; a question that names another folder is answered for that folder.
- ## What it carries
- | | |
- |---|---|
- | Scripts | `covers` (which skill or agent fits a job, plus marketplace plugins that match), `says` (the sections of a skill or agent about a topic, with `path:line`), `inventory` (what is installed, from where, plugin versions, marketplace search), `mentions` (every skill and agent that names a string), `lint` (frontmatter and description problems), `evaluate` (one skill folder against the checks a skill needs before it ships: description with what, when and trigger phrases, SKILL.md length, reference links and files, script permissions, placeholder text, each finding with file and line), `overlaps` (duplicates and near-duplicates) |
- | Skills | `write-skill`, `scaffold-skill` (the questions to settle before a new skill folder is written, and the scaffold call), `write-agent`, `fix-description`, `merge`, `retire`, `stale-sweep`: playbooks the investigator loads when the job calls for one |
- | Roles | `author` for requests to create or change a skill or agent; `auditor` for questions about validity, overlap and staleness |
- It reads `~/.claude/skills`, `~/.claude/agents`, `~/.claude/commands`, the repository's `.claude`
- folders, the plugins in `~/.claude/plugins/installed_plugins.json` (only the installed version of
- each) and the skills synced from claude.ai. It reads nothing else under `~/.claude`, and refuses
- files whose names look like credentials.
- Each script also runs with no model: `ask-kit run skills lint origin=user`, or
- `ask(do: "skills.lint")` from an agent.
- What `lint` checks was taken from the Claude Code documentation and checked against
- `claude plugin validate`: a frontmatter that does not parse (an unclosed quote, a tab in a block),
- a missing or glued closing `---`, no name or description, unknown or misspelled fields, tools an
- agent names that do not exist, a bad model, effort or colour, a description with no cue for when to
- use it or over the 1,536-character listing cap, a SKILL.md over 500 lines, broken links to files
- beside it, files beside it that nothing links to, scripts that are not executable, placeholder text left in the body. Settings, hooks, CLAUDE.md, rules and memory are not searched.
- ## What it can change
- When the request asks for a change, it creates, edits, renames and retires skill and agent files:
- `claude.create`, `claude.scaffold`, `claude.write`, `claude.edit`, `claude.setField`,
- `claude.rename`, `claude.retire`. A question never changes anything: each of those throws unless the caller asked
- for a change (`act`).
- - It writes only inside `~/.claude/skills`, `agents` and `commands` and the repository's
- `.claude/skills`, `agents` and `commands`. A plugin's copy and a synced claude.ai skill are
- refused, and so is any text that looks like a key or token.
- - `claude.scaffold(spec)` writes a whole skill folder from a spec (name, description, purpose, sections,
- reference files, scripts): SKILL.md with the spec's content, each reference file and script, scripts
- made executable. It refuses an existing folder and a spec with a gap or placeholder text.
- - Every change returns the lint of the file it wrote.
- - Retiring moves the file to `~/.claude/retired/` and returns the command that moves it back.
- Nothing is deleted.
- ## How well it works
- `bench/questions.json` holds six questions and three script runs, put to the folder `bench/fixtures/watch`, each with patterns
- its answer must hold. It has not been run on a model yet; `ask-kit kit test skills` runs it.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.