nmem-skill-load

name	nmem-skill-load
description	Search Nowledge Mem for skills matching a query and load or inject them on-demand into the current conversation turn or workspace. Triggered by /nmem-skill-load <query> or proactive task discovery.

Nowledge Mem On-Demand Skill Loader

Dynamically discover, preview, and load compiled AI skills from Nowledge Mem into the active conversation turn or local workspace.

Primary Entry Point

Use this skill when:

• The user runs /nmem-load-skill <skill query> or asks to "load skill ".
• Antigravity discovers an unhandled domain task (e.g. Makefile, RPM packaging, Docker, Flatpak, specialized database migrations) where a relevant skill exists on the Nowledge Mem server but is not yet active locally.

---

Workflow

Step 1: Search for Matching Skills

Run load_skill.py search to query Nowledge Mem for candidate or compiled skills matching the query:

python3 hooks/nmem_entrypoint.py skill-load search "<skill query>"

The script returns JSON listing available matching skills with their id, name, description, stage, and relevance.

Step 2: Present Preview & Selection

If multiple matching skills are returned, present the choices to the user before loading:

• Use ask_question to let the user select the desired skill.
• If a single skill clearly matches, present a brief summary of the skill name and description.

Step 3: Fetch Skill Body

Retrieve the full skill body without altering the global activation state on the server:

python3 hooks/nmem_entrypoint.py skill-load fetch "<skill_id>"

Step 4: Load into Active Session

Select the appropriate loading mode based on user intent:

Mode A: Ephemeral Turn Injection (Zero-Restart)

When the user wants the skill loaded immediately for the current task without persisting files:

• Clean Formatting: Strip leading YAML frontmatter (---...---) from the fetched skill body before rendering to the user.
• No Raw XML Tags: Do NOT output raw <dynamically_loaded_skill> XML tags or unparsed YAML frontmatter headers in user-facing chat text.
• Present the ingested skill using a clean GitHub-style callout:

  > [!TIP]
  > **Loaded Skill Context**: `<skill_name>` (`<trust_badge>`)
  > *Ingested into active turn for this workspace.*
  
  <clean-markdown-body-content>
  

• Proceed directly to follow the injected skill's instructions for the current turn.

Mode B: Persistent Workspace Installation

When the user asks to install the skill for future use in the current repository:

python3 hooks/nmem_entrypoint.py skill-manage install "<skill_id>" "<workspace-root>" [--ignore]

• Installs the file to <workspace-root>/.agents/skills/<skill-name>/SKILL.md.
• Appends .agents/skills/<skill-name>/ to <workspace-root>/.git/info/exclude if --ignore is set.
• Notify the user once installed.