Telusko Docs

Build your first skill, /telusko-status, that injects live git and backend data into the prompt, and fix the permission error that injected commands can cause.

Our first skill is simple but useful: a quick status report for the project. Current branch, recent commits, what is in specs/, and whether the backend is running, summarised in a few lines. Perfect before starting any work.

Step 1: Create the folder and file

Create a `skills` folder inside `.claude`

.claude/skills/

Create a folder for the skill

The folder name becomes the command name:

.claude/skills/telusko-status/

It must be a folder, not a file.

Create `SKILL.md` inside it

The file name is exactly SKILL.md: uppercase SKILL, singular, with a .md extension.

.claude/skills/telusko-status/SKILL.md

Step 2: Write the skill

.claude/skills/telusko-status/SKILL.md

---
description: Shows the current state of the Telusko Workflow Engine, including git branch, recent commits, contents of specs/, and whether the backend is responding. Use when you need quick situational awareness before starting work.
allowed-tools: Bash(git branch *) Bash(git log *) Bash(ls *) Bash(curl -sf http://localhost:8000/*)
---

## Current branch
!`git branch --show-current`

## Recent commits
!`git log --oneline -5`

## Specs in this repo
!`ls -la specs/ 2>/dev/null || echo "no specs/ directory"`

## Backend health
!`curl -sf http://localhost:8000/api/tasks > /dev/null && echo "backend up" || echo "backend not responding"`

## Instructions
Summarize the above into a single-paragraph status report.
Flag anything unusual: uncommitted work, missing spec files, backend down.
Keep it under 5 lines.

The opening --- must be on line 1

Claude Code only reads the frontmatter when --- is the very first line of the file. In the project repository, most SKILL.md files start with an empty line before ---. In that case Claude Code treats the whole file, including the frontmatter, as plain instructions, and silently ignores the description, allowed-tools and other settings. If an editor keeps inserting a blank line, delete it before saving.

The frontmatter

Field	Required	Purpose
`description`	Recommended	Tells both humans and Claude what the skill does and when to use it. If omitted, the first line of the body is used
`name`	No	Defaults to the folder name (`telusko-status`)
`allowed-tools`	No	Tools the skill may use without asking for permission while it runs

For sub-agents, name and description were both required. For skills, name is optional because the folder name already provides it. The description is what really matters.

The !`command` lines

This is what makes skills powerful. You already know ! from shell mode. Inside a skill, a line like !`git branch --show-current` means:

Before Claude even sees this skill, run this command and put its output here.

So Claude never reads the text git branch --show-current. It reads the actual branch name, for example main. By the time Claude reads the skill, all the live data is already filled in. That is called dynamic context injection.

Step 3: Use it

Type / and look for telusko-status, or list all skills:

/skills

Then run it:

/telusko-status

Claude replies with a short report: branch, last commits, which spec files exist, and whether the backend is up.

Skill not showing up?

Claude Code watches skill folders and picks up new skills within a few seconds. The exception: if the .claude/skills/ folder itself did not exist when the session started, run /reload-skills (or restart Claude Code) once.

Troubleshooting: "Shell command permission check failed"

In the video, the first run failed with:

Shell command permission check failed for pattern "..."

Here is why. Injected !`...` commands run before Claude is involved, so they never show a permission prompt. Claude Code checks each one against your permission rules, and if a command is not explicitly allowed, the whole skill aborts.

There are two fixes:

Fix	Where	Scope
`allowed-tools` in the skill's frontmatter (shown above)	`SKILL.md`	Only while this skill runs. Recommended
An `allow` rule in your settings, for example `"Bash(curl -sf http://localhost:8000/*)"`	`.claude/settings.json`	Always

Two more gotchas we hit along the way:

When copying a command into a permission rule, do not include the leading !. The rule is Bash(git log *), not Bash(!git log *).
Make sure the settings file is actually saved and correctly named (.claude/settings.json).

Once the commands are allowed, the skill runs and returns the five-line summary.

Let Claude update settings for you

Claude Code has a bundled /update-config skill. Describe the permission you need, for example "allow curl to localhost:8000 for this project", and it edits the right settings file for you. Review the change before accepting it.

Official references

Written By: Shramik Masti

Creating a Custom Status Skill