Nmem Skill Manage

Nowledge Mem Skill Manager

This skill allows the agent to list, suggest, and install skills from the user's Nowledge Mem knowledge base into the current project workspace.

Preferred Execution Hierarchy

1. REST API (Primary): The manage_skills.py script queries the nmem HTTP server (default http://127.0.0.1:14242).
2. CLI Fallback: If the server is unreachable or disabled, the script automatically falls back to invoking nmem skills list and nmem skills show to retrieve and install skills.

When to Use

• Use when the user asks to "install a skill", "update a skill", or "import skills from nmem".
• Use when setting up a new workspace to see what existing compiled procedures (ruling, check-lists, styling conventions) from Nowledge Mem would help this project.
• Use to keep local workspace skills in sync with updates on the Nowledge Mem server.

Workspace Target

All skills installed via this pipeline are placed in: <workspace-root>/.agents/skills/<skill-folder>/SKILL.md

Workflow

Step 1: Suggest and Analyze Relevance

To find out what skills might be relevant to the current workspace, run:

python3 hooks/nmem_entrypoint.py skill-manage suggest <workspace-root>

This script scans for Makefiles, workflow files, Flatpak/AetherPak configurations, and Git status to score and match skills on the server.

Step 2: Prompt the User (Rich Interface & Feedback Loop Optimization)

Do not install skills silently. Use Antigravity's rich interaction interfaces to solicit approval in a single turn:

• Option A (Interactive Prompt): Use the native ask_question tool to present a list of skills.
  • Set is_multi_select: true to let the user select multiple skills at once.
  • If a specific skill is highly recommended, list it first with (Recommended) prefix.
  • Follow up with a second ask_question or option on whether to commit them to the repository or keep them local.
• Option B (Proceed Artifact): For larger installations, write a skills_installation_plan.md artifact under <appDataDir>/brain/<conversation-id>.
  • Set RequestFeedback: true and UserFacing: true in the ArtifactMetadata to present a "Proceed" button.
  • Present the suggested skills in a markdown Table showing the Skill ID, Trust Badge (Proven, Checked, Draft), Description, Relevance reasons, and git commit strategy (e.g., local git exclude vs. committed).
  • The user can click "Proceed" to approve and install the plan in one click.

Step 3: Install/Update the Skill

Run the install command:

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

• If the skill is in candidate stage, the script will automatically trigger compile and wait for it.
• It will download the body directly using GET /skills/<id>?include_body=true (which avoids modifying the global activation state on the server).
• If --ignore is specified, it automatically appends .agents/skills/<skill-folder>/ to <workspace-root>/.git/info/exclude.

Step 4: Verify

Verify that the files have been written under <workspace-root>/.agents/skills/ and notify the user.

Step 5: Undo a Skill merge

If you need to revert a Skill merge (restore the absorbed Skill and remove the merge's pending evidence from the kept Skill), run:

python3 hooks/nmem_entrypoint.py skill-manage restore-merge <skill_id>

Note: This option is only available until that evidence is applied or changed.