propose-skill

Detects repeating session workflows and drafts new skill proposals for approval.

Updated May 3, 2026
One-click install
npx skills add https://github.com/spikelab/multiplai-cc-mktplace --skill propose-skill-spikelab
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: propose-skill
Source: https://github.com/spikelab/multiplai-cc-mktplace/tree/main/plugins/multiplai-dev/skills/propose-skill
Command: npx skills add https://github.com/spikelab/multiplai-cc-mktplace --skill propose-skill-spikelab

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Repeating multi-step workflows in Claude Code sessions go unnoticed and get re-explained every time; this Skill spots those recurring patterns and turns them into a reviewed, tested skill proposal instead of letting them accrete informally. ## Core Features & Use Cases - Pattern Detection: Reads session diary entries, learnings files, and conversation context to identify recurring tool chains, command sequences, and multi-step workflows. - Trajectory Extraction: With --from-session, reads the raw session transcript to separate stable steps from judgment points and capture dead ends that summaries drop. - Gated Installation: Drafts to a temporary directory, runs a promotion gate that validates frontmatter and executes bundled scripts with --help, and only installs after explicit user approval. - Use Case: After a session where you repeatedly ran the same cost-report commands, say "make this a skill" and get a drafted SKILL.md with the stable commands as literal steps and the decision criteria preserved as judgment points. ## Quick Start Ask the assistant to formalize the workflow we just repeated in this session into a new skill proposal.

Frequently Asked Questions about propose-skill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I turn a repeated Claude Code workflow into a skill?

Say "formalize this" or "make this a skill" and the Skill analyzes session context for recurring tool chains and command patterns. It drafts a SKILL.md, shows it for approval, and only writes files after you explicitly say yes.

How do I create a skill from the current session transcript?

Pass --from-session to read the raw JSONL transcript instead of diary summaries. This captures the actual tool calls, real arguments, and dead ends, then splits them into stable steps and judgment points for the draft.

Does propose-skill work without the multiplai-context plugin?

Yes, but with reduced context. Without the plugin's .multiplai/diary/ and .multiplai/learnings/ artifacts, it relies on the current conversation alone, so cross-session pattern detection is unavailable.

What does the promotion gate check before installing a skill?

The promote_skill.py script validates frontmatter keys and model/effort values, then executes every bundled script with --help expecting exit 0. A failing gate blocks installation until the reported issues are fixed.

When should I not create a skill from a pattern?

Skip patterns already covered by an existing skill, single-command actions too simple to warrant one, and one-off procedures unlikely to recur. If three sentences in CLAUDE.md would do the job, prefer that instead.