Skip to main content
Skills are reusable capability bundles that teach an agent how to perform specialized tasks — querying a database, triaging an alert, following an escalation playbook, or drafting release notes. In Agent Harness, skills come from the TrueFoundry Skills Registry — a centralized, versioned, RBAC-governed catalog. This is fundamentally different from other harness products where skill files are scattered across repos with no central governance.
For the SKILL.md format and how the core harness loads git-backed skills on demand, see Setup Skills in the TrueForge docs. This page covers the governance TrueFoundry adds on top: the versioned Skills Registry, RBAC, version pinning, and audit trails.

​
Why centralized skills matter

In Claude Managed Agents or LangSmith Managed Deep Agents, skills are files you commit to agent repos or upload via API. There is no shared registry, no version pinning independent of code deploys, and no access control per skill. TrueFoundry Skills Registry provides:
Platform teams govern which skills are available to which agents and teams. Agent builders pick from an approved catalog — no manual file management.

​
What a skill contains

A skill is a directory rooted at a SKILL.md file with YAML frontmatter (name + description) and a markdown body with instructions.
The description field is what the agent sees upfront to decide whether a skill is relevant. The full body is loaded only when the agent picks the skill (progressive disclosure).

​
Versioning and production stability

Skills are versioned artifacts in the Skills Registry:
  • Pin a specific version on a production agent for guaranteed behavior
  • Test new versions in staging or playground before promoting
  • Roll back instantly if a skill change causes regressions
  • Audit history of what changed, when, and by whom
This is especially important for regulated or customer-facing workflows where reproducibility is mandatory.

​
How skills are used at runtime

At run time, the harness:
  1. Exposes skill metadata (name, description) to the model at startup
  2. When the model determines a skill is relevant, loads the full SKILL.md body
  3. The agent follows the skill’s instructions and accesses supporting files as needed
This progressive disclosure pattern keeps context lean while giving the agent access to specialized procedures on demand.

​
Preload Skill / MCP tools into agent context

For each skill, you can configure Preload Skill / MCP tools into agent context. When enabled:
  • Full SKILL.md is preloaded into the agent’s context at startup
  • The agent does not need an extra turn to read the skill
  • Uses more context space upfront

​
Skills + MCP tools

Skills and MCP tools are complementary:
  • Skills define how to handle a task (the procedure, decision logic, formatting rules)
  • MCP tools provide the actions and data needed to execute it
Example: a “CRM follow-up” skill instructs the agent on decision logic and message tone, while MCP tools handle record lookup, ticket update, and Slack notification.

​
Access control

Skills inherit access control from their parent Repository in the Skills Registry. See Repositories for the shared repository concept, storage backing, and access-control model.
  • Creators and admins publish and manage skills
  • Teams and users with read access can attach skills to agents
  • End-users invoking an agent get implicit access to attached skills
This means platform teams can govern which skills are available to which agents without per-skill manual permissions.

​
Getting started

  1. Publish or import skills in the Skills Registry.
  2. Select skills while creating/editing an agent in the builder.
  3. Pin versions for production agents.
  4. Validate behavior in playground before promoting to broad usage.
For end-to-end setup, see Create, Manage, and Use Skills and Add Skills in the Agent Playground.