All extensions are plain files. Put them in .neuralinverse/ inside a project to share them with your team through the repository, or in ~/.neuralinverse/ to use them everywhere. Project files override global ones with the same name. Organisation files delivered by a signed policy override both and show org as their source in /skills, /agents and /help.
| What | Project location | Global location |
|---|---|---|
| Project instructions | AGENTS.md or NEURALINVERSE.md in the repository | ~/.neuralinverse/AGENTS.md |
| Slash commands | .neuralinverse/commands/*.md | ~/.neuralinverse/commands/*.md |
| Sub-agent profiles | .neuralinverse/agents/*.md | ~/.neuralinverse/agents/*.md |
| Skills | .neuralinverse/skills/<name>/SKILL.md | ~/.neuralinverse/skills/<name>/SKILL.md |
| Answer styles | styles/*.md | styles/*.md |
Project instructions: AGENTS.md
AGENTS.md (or NEURALINVERSE.md; the names are set by instructionFiles) is loaded from your global configuration folder, then from every folder between the repository root and the working folder. Write build and test commands, conventions, and anything the agent should always know.
| Command | What it does |
|---|---|
/bootstrap | The agent studies the repository and drafts AGENTS.md |
/remember <note> | Appends a standing instruction |
/instructions | Shows what is loaded and opens AGENTS.md in your editor |
Agent instruction files managed by your organisation in Govern can be written to a local, untracked override file with neuralinverse context agent-files effective --write. See Organisation context.
Slash commands
A Markdown file with front matter becomes a command named after the file:
---
description: Write release notes for a version
argument-hint: <version>
---
Write release notes for version $1 from the commits since the previous tag.
Group them by feature, fix and internal change. Extra instructions: $ARGUMENTSSaved as .neuralinverse/commands/release-notes.md, this becomes /release-notes 2.4.0. $ARGUMENTS is everything after the command; $1 to $9 are individual arguments.
Sub-agents
The agent can hand self-contained work to a helper agent with the delegate tool. Helpers run in their own conversation and report back, which keeps the main context small. Named profiles give helpers a brief, a restricted tool set or a different model:
---
name: security-reviewer
description: Reviews changes for injection, auth and secrets problems
tools: read_file, search_text, find_files, list_dir, run_shell
model: anthropic/claude-opus-5-5
---
You review code for security problems. Report each finding with file:line, severity,
and a concrete fix. Do not modify files./agents lists the profiles; /agents new <name> creates one. Without a model, helpers use the subagent role model (see Providers and models).
Skills
A skill is a folder of instructions plus any supporting files: scripts, templates, references. Only its name and description are shown to the agent up front; the agent loads the full text with the load_skill tool when a task matches, so skills cost almost no context until used.
---
name: db-migration
description: How to write and verify database migrations in this repository
---
1. Create the migration with `make migration name=<name>` …/skills lists what is available.
Answer styles
/output-style (or the outputStyle setting) chooses how answers are written: default, concise, explanatory, review, or a custom style from a styles/*.md file.
Share a set of extensions
To share commands, agents, skills, hooks and MCP servers as one installable unit, package them as a plugin. To copy your own extensions to another machine, use neuralinverse settings export or settings sync (Settings reference).