Resource and skill precedence

Context resources

Nerve loads global agent context and project/ancestor AGENTS.md files root-to-project. More local instructions augment or override through ordering as defined by the loader.

Nerve-specific project resources live under .nerve/. Global Nerve agent resources live under <NERVE_HOME>/agent/ (normally ~/.nerve/agent/). A project SYSTEM.md overrides the global system resource; append prompts from applicable scopes are concatenated.

Skills

Effective skill discovery is first-name-wins in this order:

project .nerve/skills;
.agents/skills from project/ancestors according to project discovery order;
global <NERVE_HOME>/agent skills;
global ~/.agents/skills;
enabled Built-in Nerve skills;
enabled Agent Browser skill guidance.

Within each skills root, Nerve recursively discovers SKILL.md, accepts direct root .md files, and honors .gitignore, .ignore, and .fdignore. Skills require a description; directory-based names must match their parent directory and use lowercase letters, digits, and single hyphens. Invalid files are omitted with diagnostics. disable-model-invocation: true keeps a skill out of model-visible discovery while preserving explicit application use.

Built-in Nerve and Agent Browser skills are disabled by default. A disabled skill contributes neither metadata nor instructions to the agent system prompt. Settings toggles do not delete source files, and changes apply to subsequent agent runs.

Project definitions therefore take precedence over global, built-in, and Agent Browser definitions with the same name. Built-in Nerve skills take precedence over same-named Agent Browser guidance. Inspect the Settings Skills view to see discovered scope and effective state.

Unsupported legacy paths

Legacy .pi directories are not loaded. Move Nerve-specific files into .nerve/ and portable skills into .agents/skills/.