skillsmp-search - SKILL.md Agent Skill

name: skillsmp-search description: Use when the user wants to search SkillsMP for community skills, compare candidates, or find a skill by tool, workflow, or intent before deciding whether to install it. Trigger for requests such as "search SkillsMP", "find a Codex skill for X", "look up a community skill", or "find a skill that helps with Y".

SkillsMP Search

Use this skill to search SkillsMP from Codex without opening the website manually.

This Codex skill is the deployed format for the shared repository workflow defined in /home/davisye/project/davis-llm-skill/shared/skillsmp-search/core.md. Keep the semantics aligned with the shared source and generated adapters.

Workflow

Map the user's request to one concrete search query.
Choose search mode:
- keyword for exact tool names, formats, or platform terms
- ai for intent-heavy requests where the user describes a goal rather than known keywords
Run python ./scripts/search_skillsmp.py "<query>" with optional flags.
Review the top candidates before recommending anything.
If the user wants to install or adapt a result, inspect the linked repository or skill files first.

Commands

Keyword search:

python ./scripts/search_skillsmp.py "codex github review" --mode keyword --limit 5 --sort-by stars

AI semantic search:

python ./scripts/search_skillsmp.py "find a skill that helps review pull request comments" --mode ai

JSON output for further processing:

python ./scripts/search_skillsmp.py "context7" --json

Token handling

The script looks for the API token in this order:

SKILLSMP_API_TOKEN environment variable
secret-tool lookup service skillsmp key api-token

Do not hardcode the token into the repository or skill files.

Query rules

Prefer keyword search when the user already knows the domain or tool.
Prefer AI search when the prompt is goal-based rather than name-based.
Keep the query narrow. One use case is better than five mixed ones.
Use --sort-by stars for mature results and --sort-by recent for fast-moving topics.
Treat SkillsMP as a discovery index, not a trust boundary.

Answer style

State which mode you used and why.
Give a short ranked list with name, URL or repository, and fit.
Call out obvious risks such as low stars, stale updates, or unclear ownership.
If a result looks promising, say that the underlying repo should still be reviewed before installation.