Skip to main content
Skills are reusable sets of instructions that customize how Reliant’s agents behave for specific tasks. Think of them as saved expertise — a skill for “frontend design” or “incident response” gives agents the right context, examples, and rules without you repeating yourself every time. Skills use the shared Agent Skills format (SKILL.md), so they’re portable across tools that support the standard.

​
Quick start

Create a skill by adding a SKILL.md file in a skill directory:
A minimal SKILL.md looks like this:
That’s it — Reliant discovers it automatically and offers it to agents. Write the description for the agent: it is what an agent reads when deciding whether the skill applies to the task in front of it.

​
How agents use skills

Every agent turn carries a short list of the skills available in the project, each with its path and the start of its description. When a task matches one, the agent loads it with the skill tool and the skill’s instructions join the conversation. The tool can also list skills and search them by keyword, and it can read a large skill one section at a time, so a long skill doesn’t have to fill the context at once. Because loading is the agent’s decision, a good description matters more than anything else in the file. You can also simply ask: “load the release skill and follow it” works in any chat. A workflow can skip the decision and preload skills into a model call with the skills field of a Call LLM node, so the model starts with their instructions already in context.

​
How discovery works

Reliant discovers skills on the chat’s machine, in the project and in your home directory. It reads its own skill folders and also the folders Claude Code and Codex use, so skills you already have there work without copying them: When two skills share a name, the one with the lower priority number wins, so a skill in .reliant/skills/ overrides a Claude Code skill of the same name. Use .reliant.local/skills/ (gitignored) for personal variants and .reliant/skills/ for skills the team shares. In a project with several repositories, each repository’s own skill folders are read too, and its skills are listed with the repository as a prefix (api/deploy, web/deploy). A SKILL.md in a subdirectory of a skill is a sub-skill: .reliant/skills/go/error-handling/SKILL.md becomes go/error-handling, listed under go. Reliant ships a few built-in skills of its own: code-search, reliant-config and workflow-builder. General-purpose methodology skills, such as code review, debugging and refactoring, come from Forge and are offered in every project; a Forge project also gets Forge’s own framework skills.

​
SKILL.md format

​
Required fields

  • name — 1–64 characters, lowercase letters, digits, and hyphens. No leading/trailing hyphens, no consecutive hyphens. Must match the parent directory name.
  • description — 1–1024 characters.

​
Optional fields

  • license — License identifier.
  • compatibility — 1–500 characters describing compatibility notes.
  • metadata — A string-to-string map for custom key-value pairs.
  • allowed-tools — Space-delimited tool list (e.g. Bash(git:*) Read). When an agent loads the skill, these are shown to it as the tools the skill suggests. They are a hint, not a restriction: the agent’s tools stay the same.
Claude Code’s argument-hint, disable-model-invocation, user-invocable and paths fields are accepted so Claude Code skills parse unchanged, but Reliant doesn’t act on them. In Reliant’s own folders, any other frontmatter field is an error and the skill is skipped. Skills in the Claude Code and Codex folders are read leniently, and unknown fields there are ignored.

​
Supporting files

A skill directory can hold more than SKILL.md: examples, templates, reference notes, scripts. Mention them in the skill’s instructions by relative path, and the agent reads them from the skill’s directory when it needs them, like any other file. Only the SKILL.md body is loaded by the skill tool.

​
Compatibility

Reliant supports the shared Agent Skills format used by Claude and Codex, including SKILL.md with YAML frontmatter, supporting files, and discovery across project and global scopes. What Reliant does not support:
  • Plugin ecosystems or plugin execution contracts from other tools
  • Runtime side effects that bypass Reliant’s tool governance
  • External installer behaviors that conflict with Reliant’s installation model
To move skills into Reliant’s own folders, for example to override them or share them with your team, copy them into .reliant/skills/ or .reliant.local/skills/ yourself, ask the agent in a chat to do it, or run Import from Claude Code / Cursor.