Skill

Expanded Definition

A skill is a modular instruction package that an agent loads at runtime to shape how it behaves, what tools it may use, and how it completes a task. In agentic systems, skills are more than documentation because they can directly influence execution paths, argument construction, and operational decisions.

Definitions vary across vendors and implementation patterns. Some skill systems are primarily prompt bundles, while others include metadata, code, configuration, and supporting references that are invoked conditionally. NHI Management Group treats the security relevance of a skill as arising from its runtime effect: if the package can alter execution authority or tool selection, it becomes part of the control plane and should be governed accordingly. That makes skills adjacent to, but not the same as, prompts, policies, playbooks, or plugins. The NIST Cybersecurity Framework 2.0 helps frame this as a governance and protection problem, especially where change control, access review, and traceability are required. For background on how NHIs become operationally risky when they are embedded in workflows, see the Ultimate Guide to NHIs and NIST Cybersecurity Framework 2.0.

The most common misapplication is treating a skill as harmless reference material, which occurs when teams allow runtime-loaded instructions to bypass the same review used for code, secrets, and privileged automation.

Examples and Use Cases

Implementing skills rigorously often introduces governance overhead, requiring organisations to weigh faster agent enablement against the cost of approval, testing, and rollback controls.

• A customer support agent loads a billing-dispute skill that narrows permitted tools to read-only account lookup and case creation, reducing the chance of unintended changes.
• An engineering agent uses a deployment skill that includes release checks, environment-specific parameters, and rollback steps, which improves repeatability but requires strict version control.
• A security triage agent receives a phishing-analysis skill that points to approved evidence-gathering steps and logging requirements, helping preserve chain of custody.
• An operations agent invokes a database-maintenance skill that encodes safe thresholds and escalation rules, limiting free-form tool use while speeding routine work.
• A procurement agent loads a vendor-review skill that instructs it to consult approved policy documents and compliance criteria before generating recommendations.

These patterns become especially important in organisations already struggling with NHI governance and secret sprawl, as described in the Ultimate Guide to NHIs. In practice, teams often map skill governance to NIST Cybersecurity Framework 2.0 functions for access control, change management, and monitoring.

Why It Matters in NHI Security

Skills matter because they can quietly become an execution layer for privilege. If a skill can alter an agent's planning, tool access, or output constraints, then compromised or poorly reviewed skills can redirect an otherwise trusted agent into unsafe actions. That risk is amplified when skills are shared across teams, updated without provenance, or loaded from locations that are not covered by normal software supply-chain checks. NHI Management Group has found that only 5.7% of organisations have full visibility into their service accounts, which is a useful signal for the broader pattern: when identity governance is weak, runtime control packages tend to spread faster than oversight.

Security teams should treat skills as auditable assets with owners, versioning, approval states, and revocation paths. They should also verify whether a skill can read secrets, call external tools, or override default guardrails, because those behaviors create governance obligations similar to other privileged NHI components. Organisational failures often show up only after an agent has executed an unintended action or exposed data, at which point skill governance becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• Should organisations prioritise tool scoping or skill governance first for AI agents?
• What should organisations do when an agent skill can silently replace another skill?
• Should organisations use security skill prompts instead of access controls for AI agents?
• What breaks when agent skill registries can deliver malicious installers?