Preload Hermes Skills at Session Launch

When the task domain is known in advance, preload the relevant Hermes skills instead of spending the first turn discovering them. The -s flag loads each named skill into the session prompt before the first user message. It works in interactive and single-query CLI modes.

hermes -s hermes-agent-dev,github-auth
hermes chat -s github-pr-workflow -s github-auth

Preloading is most useful for repeatable operator entry points: repository review, release work, a known business integration, or a troubleshooting runbook.

Select a focused skill set

A skill should contribute concrete procedural value. Prefer the smallest set that covers:

the platform or product being operated;
the quality or security gate that applies;
the delivery surface, such as GitHub or Cloud Run;
a specialized tool whose syntax should not be guessed.

Do not preload an entire catalog. Every skill consumes prompt space, and overlapping skills can create noisy or conflicting guidance.

Build a repeatable launch command

List available skills and confirm their exact names.
Read the relevant skill before relying on it in an important workflow.
Put stable launch commands in team documentation or a reviewed wrapper.
Keep project-specific facts in project files, not copied into a generic skill.
Start the session in the correct working directory.

A launch command is not proof that a skill is current. Skills are procedural memory and should be patched when a real workflow reveals stale commands or missing verification steps.

Preload checklist

Each skill is relevant to the immediate task.
Exact skill names resolve without warnings.
The working directory points to the intended project.
No secrets are embedded in aliases or shell history.
Project instructions and skill instructions do not conflict.
The first response reflects the expected workflow.

Common pitfalls

Using preload as authorization. A skill can describe a write operation; it does not grant permission for risky or paid actions.
Loading too many skills. Prompt bloat reduces clarity and makes policy precedence harder to reason about.
Assuming session-global changes. The preload applies to the session that was launched, not every existing gateway conversation.
Freezing stale commands in aliases. Keep the skill as the source of procedural truth and review wrappers after updates.

Verification

Launch a disposable session with one known skill and ask for a concise summary of the workflow it contributes. Confirm the session starts in the intended directory and that the skill appears before the first substantive task. Then run a harmless, skill-specific read-only check.

Official reference: Hermes CLI interface.