How PM Skills Load Automatically During Conversations: A Technical Deep Dive

PM Skills load automatically when the Claude platform detects terminology in user prompts that matches a skill's name or description, requiring no explicit command unless forcing a specific skill via slash syntax.

The phuryn/pm-skills repository implements an intelligent plugin architecture that enables PM Skills to load automatically during conversations with Claude-based assistants. When using Claude Code, Claude Cowork, or compatible agents, the platform analyzes incoming prompts to identify domain-specific frameworks and injects relevant structured knowledge without manual invocation. This seamless integration eliminates the need for explicit /skill commands while maintaining flexibility through force-loading capabilities.

How Automatic Loading Triggers During Conversations

According to the repository's documentation in README.md (lines 41-42), skills are loaded automatically "when relevant to the conversation" rather than requiring explicit activation commands. The platform performs real-time terminology analysis on user prompts, scanning for phrases that match skill names or descriptions defined in individual SKILL.md files throughout the repository.

When a user mentions concepts like "Opportunity Solution Tree" or "resume review," the assistant consults the plugin manifest to locate the corresponding skill definition. The system then loads the skill's structured guidance from its SKILL.md file, making the framework immediately available to the model without interrupting the conversation flow.

Plugin Manifest and Skill Registration

The discovery mechanism relies on /.claude-plugin/marketplace.json, which serves as the central registry for all available plugins. As defined in lines 11-16 of this file, the manifest enumerates plugin directories through entries like "source": "./pm-toolkit", enabling the platform to map skill names to their implementation files.

Each skill resides in a dedicated directory containing a SKILL.md file with YAML front-matter. For example, pm-toolkit/skills/review-resume/SKILL.md (lines 1-4) defines the skill's metadata including its name, which the platform parses to register the skill under the appropriate plugin namespace. This structured approach ensures that terminology matching can accurately resolve to the correct skill implementation.

Force-Loading Skills for Priority Override

While automatic loading handles most scenarios, users can force-load a skill to prioritize it over the model's general knowledge. As documented in README.md (lines 41-42), this is achieved by prefixing requests with either the plugin name and skill name or simply the skill name.

# Force-load using plugin and skill name

/pm-toolkit:review-resume Can you review my product manager resume?

# Force-load using skill name only

/review-resume Review my resume for PM positions

This explicit invocation syntax ensures that specific frameworks are applied even when the automatic detection might otherwise favor general knowledge, providing precision control without sacrificing the convenience of automatic loading.

Practical Examples of Automatic Detection

The following interactions demonstrate how PM Skills load automatically during conversations based on terminology detection:

1. "Help me build an Opportunity Solution Tree for my new SaaS product" — The platform detects "Opportunity Solution Tree" and automatically loads the opportunity-solution-tree skill from the pm-product-discovery plugin.

2. "Can you review my product manager resume?" — The phrase "resume review" matches the review-resume skill, triggering automatic loading from the pm-toolkit plugin.

3. "What should our north-star metric be for a two-sided marketplace?" — The terminology activates the north-star-metric skill in the pm-marketing-growth plugin.

Summary

• PM Skills load automatically during conversations when the Claude platform detects prompt terminology matching skill names or descriptions in SKILL.md files
• The .claude-plugin/marketplace.json manifest (lines 11-16) provides the discovery mechanism for all available plugins and their skills
• Force-loading via /plugin-name:skill-name syntax overrides automatic selection when specific framework priority is needed
• Each skill's metadata in SKILL.md front-matter (lines 1-4) enables accurate terminology matching and namespace registration

Frequently Asked Questions

What triggers automatic skill loading in PM Skills?

The automatic loading triggers when the Claude platform detects terminology in user prompts that matches a skill's name or description as defined in the YAML front-matter of SKILL.md files. This analysis happens in real-time during conversation processing without requiring explicit user commands, as implemented in the phuryn/pm-skills repository.

How do I force a specific skill to load instead of relying on automatic detection?

Prefix your request with a slash command using either /skill-name or /plugin-name:skill-name syntax. For example, /pm-toolkit:review-resume explicitly loads the resume review skill from the pm-toolkit plugin, overriding the automatic selection process described in README.md (lines 41-42).

Where are skill definitions stored in the repository?

Each skill resides in a subdirectory within its plugin folder, containing a SKILL.md file with YAML front-matter that defines the skill name and metadata. The .claude-plugin/marketplace.json file (lines 11-16) serves as the central registry mapping these skills to their source directories like ./pm-toolkit.

Do I need to manually install plugins for automatic loading to work?

No, the Claude platform reads the marketplace.json manifest to discover all available plugins automatically. As long as the plugin is registered in this manifest, the platform can detect and load skills based on conversation context without manual installation steps, allowing PM Skills to load automatically during conversations.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}