Filling a profile

A new profile starts empty. /neuronzai:init-profile fills it from the places your knowledge already lives — repositories, GitHub, Notion, internal docs — so that every agent working in that profile afterwards knows that material the way a colleague who has been there from the start does, without opening the sources again.

Running it

Name the sources in plain words:

/neuronzai:init-profile scan the whole billing repo here and everything on its GitHub, plus the complete Notion engineering wiki

Run it with no argument inside a git repository and it analyses that repository: its layout, how to build, run and test it, its architecture, conventions and gotchas, its recent history, and the docs it carries (README, CLAUDE.md, AGENTS.md, docs/). It writes nothing into the repository; everything it learns is stored in Neuronz.ai. Run it with no argument outside a repository and the agent asks what to read.

The records go to the profile the session is using. To fill a brand-new profile, create it first with /neuronzai:new-profile, which offers this command when it finishes.

What it does

1. Checks access first. For each source, the agent confirms it can reach it: the matching MCP server or CLI (GitHub, Notion, …) is installed and signed in. A source it cannot reach is listed straight away with what it needs, and the rest go ahead.

2. Reads everything, records what matters. The agent reads every reachable source in full and writes down what an experienced colleague would know: what the product is, who owns what, how releases, incidents and reviews work, the internal vocabulary, past decisions and why they were made, and the traps. It does not copy the sources. Status that changes from day to day, such as a build result or the state of one pull request, is left out. Secrets such as passwords, tokens and keys are never stored, even when a source contains them; the record says where they live instead.

3. Writes each thing to the right place:

   • single facts, including glossary terms and named systems, as facts;
   • runbooks, processes, architecture overviews and one overview per repository as knowledge docs;
   • colleagues and teams on the social map, with what they own and how they work;
   • every instruction the material gives the people working there ("always…", "never…") as a proposed rule, waiting for your approval. A sentence that only describes how the software behaves is recorded as a fact.

   Every record stands on its own: a link to the source may be attached, but the content is always written out.

4. Splits large sources. A source too large to read at once is divided into sections, each read by a subagent that writes its own records under the same instructions. Duplicates between sections are merged.

5. Proves the result with a quiz. The agent writes questions from the sources — who approves a release, what a term means, why something was dropped, how to run the tests — and a fresh subagent with no access to the sources answers them from Neuronz.ai alone. Every question it misses is a gap the agent fills, and the quiz repeats until every question is answered or shown to have no answer in the sources.

It writes without asking you to confirm each record first. Rules are the exception: they stay proposed until you approve them.

The report

When it finishes, the agent reports:

• the sources it read, and how much of each;
• the sources it skipped, why, and what would unlock them;
• the records it wrote, counted by kind;
• the quiz score, on the first pass and the last.

Running it again

Running it again on the same sources refreshes the profile. Each knowledge doc is keyed to where it came from, so it is updated in place rather than duplicated, and a fact that changed in the source replaces the old one.

Related

• Profiles — create a profile and choose which folders use it.
• Knowledge docs — the documents it writes.
• Rules — approving the rules it proposes.
• Harness parity — which harnesses run this command.