S
Skilldex
Get the CLI

skillpm suggest

Read the current project, search the registry for skills that fit it, and let you approve each suggestion before anything is installed.

Usage

skillpm suggest [--project-path <path>] [--yes] [--json]

Flags

| Flag | Default | Description | |---|---|---| | -p, --project-path <path> | current directory | Project to analyse | | -y, --yes | false | Approve every suggestion without prompting | | --json | — | Print the proposals as JSON without prompting or installing |

How it works

  1. Read the project. suggest builds a profile from files such as the README and package.json. If there is nothing to go on, it says so and stops instead of guessing.
  2. Search the registry. Search terms from the profile are run against the registry.
  3. Choose. A model picks the best fits from the skills the search returned. It can only choose from those results, so every suggestion is a real, installable skill.
  4. Approve. Each suggestion is shown with its qualified name, a reason, and a suggested scope. You install it at that scope, pick another, or skip it.
  5. Install. Approved skills are installed exactly as skillpm install <owner>/<name> would.

When the project needs something the registry does not have, suggest lists it separately as a gap and offers to draft it as a new skill in your project.

Requirements

Step 3 calls the Anthropic API, so suggest is the one command that needs a key:

export ANTHROPIC_API_KEY=sk-ant-...
# or
skillpm config set anthropicApiKey sk-ant-...

To use another endpoint that speaks the Anthropic Messages API, such as a LiteLLM proxy, set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. SKILLPM_SUGGEST_MODEL picks the model (default claude-sonnet-4-6).

Examples

skillpm suggest
skillpm suggest --project-path ../other-project
skillpm suggest --json