Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/org-quicko/skillset/llms.txt

Use this file to discover all available pages before exploring further.

The Skillset CLI is designed for unattended use in CI pipelines and automation scripts. This page covers the two mechanisms that make non-interactive operation reliable: environment variables that replace the login step, and the --json flag that makes every command’s output machine-parseable and prompt-free.

​
Skip login with environment variables

In CI, skip the skillset login step entirely. Set two environment variables and the CLI picks them up on every command, always overriding the stored config file: 
SKILLSET_REGISTRY=https://skills.example.com
SKILLSET_TOKEN=sk_live_…
SKILLSET_REGISTRY and SKILLSET_TOKEN take precedence over whatever skillset login stored on disk. You can mix them: setting only SKILLSET_REGISTRY points the CLI at that Registry while leaving Token resolution to the stored config (or omitting auth entirely for read-only commands).
Catalog reads — search, info, list, update — do not require a Token. You only need SKILLSET_TOKEN for write operations (publish) and for submitting URL-installed Skills.

​
Machine-readable output with --json

Every command accepts --json. In this mode:
  • The command’s result is written to stdout as a JSON object or array.
  • Any failure is written to stderr as { "error": { "code": "…", "message": "…" } } — never mixed into stdout, so a caller parsing stdout always gets a well-formed payload.
  • Exit code is 0 on success, non-zero on any failure.
  • No prompts are shown, regardless of whether a terminal is attached.
Because --json never prompts, some flags that are optional interactively become required:
CommandRequired under --json
login--token
install--scope
publish (batch)--yes
Omitting a required flag under --json causes the command to exit with a non-zero code and write a JSON error to stderr — it never hangs waiting for input.

​
Parsing JSON output with jq

# Extract Skill names from a search
skillset search code-review --json | jq -r '.items[].name'

# List all outdated Skills
skillset list --json | jq '[.skills[] | select(.status == "outdated")]'

# Get just the published Skill's id after a publish
skillset publish ./skills/code-review --json | jq -r '.id'

# Count how many Skills in a batch publish failed
skillset publish ./skills --yes --json | jq '[.[] | select(.status == "failed")] | length'

​
Exit codes

Exit codeMeaning
0Command completed successfully
1At least one error occurred
For batch commands (publish, update) the exit code is 1 if any single Skill in the batch failed, even when others succeeded. Inspect the per-Skill status fields in the JSON output to identify which ones need attention.

​
GitHub Actions example

The following workflow publishes all Skills under ./skills on every push to main: 
name: Publish Skills

on:
  push:
    branches: [main]
    paths:
      - "skills/**"

jobs:
  publish:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - name: Set up Node.js
        uses: actions/setup-node@v4
        with:
          node-version: "20"

      - name: Install Skillset CLI
        run: npm install -g @in-org-quicko/skillset-cli

      - name: Publish Skills
        env:
          SKILLSET_REGISTRY: ${{ secrets.SKILLSET_REGISTRY }}
          SKILLSET_TOKEN: ${{ secrets.SKILLSET_TOKEN }}
        run: skillset publish ./skills --yes --json
Store SKILLSET_REGISTRY and SKILLSET_TOKEN as encrypted secrets in your repository settings.

​
More CI examples

skillset install code-review \
  --agent claude-code \
  --scope project \
  --json

​
Error format

All failures under --json produce a JSON object on stderr: 
{
  "error": {
    "code": "not_found",
    "message": "No Skill named \"code-review\" on this Registry. Run `skillset search` to see what there is."
  }
}
code is the Registry’s own error code when the failure originated there (e.g. not_found, already_published, unauthorized), and cli_error for failures that never reached the Registry. Branching on code rather than message makes scripts resilient to wording changes.
Redirect stderr to a file while capturing stdout for parsing: skillset publish ./skills --yes --json 2>errors.json. This lets you inspect failures independently without interrupting your pipeline’s stdout processing.

​
Batch publish outcomes

When publishing a directory that contains multiple Skills, publish --json returns an array. Each element has a status of "published" or "failed": 
[
  {
    "name": "code-review",
    "status": "published",
    "id": "res_01j…",
    "published_at": "2025-01-15T10:00:00Z"
  },
  {
    "name": "pdf",
    "status": "failed",
    "error": "SKILL.md missing required field: description"
  }
]
The process exits with code 1 if any element has "status": "failed". A partial success is a failure from the pipeline’s perspective — check and fix the failing Skills before relying on the batch as a whole.

Build docs developers (and LLMs) love

Get started for freeTalk to us