Using TraceMem Skills
TraceMem provides a comprehensive set of "skills" — specialized instruction files designed to teach AI agents how to correctly and safely use the TraceMem platform. These skills are hosted in the TraceMem Skills Repository and are fully compatible with the AI Skills standard.
The Skills Repository
The official skills are available at:
Each skill works as a "plugin" for your AI agent, providing it with the specific knowledge needed to perform TraceMem operations safely.
Importing Skills
You can easily add TraceMem skills to your project using the skills CLI.
Using npx skills add
To add a specific skill to your agent's environment, use the npx skills add command:
# Interactive mode (Recommended) - Select skills from a list
npx skills add tracemem/tracemem-skills
# Add a specific skill directly using the shorthand syntax
npx skills add tracemem/tracemem-skills --skill decision-envelope
This command will download the skill documentation and place it in your .agent/skills (or configured) directory, where your agent can reference it.
Bulk Installation
You can also clone the entire repository if you are building a highly capable agent that needs full platform access:
git clone https://github.com/tracemem/tracemem-skills.git .agent/skills
Available Skills
The following skills are available for installation. Use the skill name in the --skill flag.
| Skill Name | Description | Recommended For |
|---|---|---|
overview | High-level mental model of Envelopes and Traces. | All Agents |
decision-envelope | Start Here. How to open, operate, and close decisions. | All Agents |
intent-and-automation | How to choose the correct intent and automation_mode. | All Agents |
reading-data | How to use Data Products to read data (replacing direct DB access). | Reader Agents |
writing-data | How to perform governed inserts, updates, and deletes. | Action Agents |
approvals | How to handle requires_exception and wait for human approval. | High-Stakes Agents |
notes-and-context | How to "show your work" by adding audit context and reasoning. | Planning Agents |
traces-and-audit | Deep dive into the audit trail and proving "why". | Audit/Compliance Agents |
safety-and-failure-modes | Best practices for error handling, retries, and clean exits. | Reliability |
when-not-to-use | When to skip TraceMem (performance/triviality). | Optimization |
Why Use Skills?
AI agents (like those powered by Claude, GPT-4, or custom LLMs) function best when provided with clear, structured instructions on how to use tools. TraceMem skills are designed to:
- Reduce Hallucinations: Detailed invariants and "when NOT to use" sections prevent agents from guessing API behavior.
- Enforce Governance: Skills teach agents to respect approval workflows and policy denials.
- Improve Safety: Explicit instructions on handling PII and secrets help agents avoid data leaks.
- Standardize Behavior: Ensure all your agents follow the same decision lifecycle (Open -> Operate -> Close).
Example Workflow
Here is how an agent effectively uses an installed skill:
- User Request: "Please refund order #123."
- Agent Action: Agent recognizes this is a sensitive action.
- Skill Look-up: Agent reads
.agent/skills/decision-envelope/SKILL.mdand.agent/skills/writing-data/SKILL.md. - Execution:
- Agent follows the skill to open a decision with
intent="order.refund". - Agent checks policy.
- Agent performs the write.
- Agent closes the decision.
- Agent follows the skill to open a decision with
By using the standardized skills, the agent avoids common pitfalls like "forgetting to close the decision" or "writing data without a purpose string", because the skill file explicitly forbids those actions.