Bring what you have

Bring the rules you already have

Standard ownersEngineering leads6 min read

Your team already has rules. They sit in agent files, lint configs, CODEOWNERS, handbooks, wiki pages and review comments. Groundrule reads those sources and turns what it finds into proposals in the review inbox. This page lists every source, what each one produces, and how they differ in what they read, what they send, and what they cost.

Nothing you import changes your rulebook on its own. Every proposal waits until someone with an authoring role (admin, platform admin, or standard owner) accepts or rejects it.

The sources

Source	Where you start it	What it produces	Page
Repository scans	`npx @groundrule/cli scan --upload`, in a repository	Instructions from agent files, tool settings that match catalog rules, and owners from CODEOWNERS	Scan repositories
Documents	Inbox → Import a document → Upload a file or Paste text	Rules found by AI, each citing its heading, paragraph or page	Import documents
Connected apps	Inbox → Import a document → From an app	Rules found by AI in a Notion page, Google Doc or Confluence page	Import from Notion, Google Docs and Confluence
Pull-request reviews	Inbox → Import a document → PR reviews, or `/groundrule rule` in a review comment	Rules reviewers keep asking for, citing the comments	Rules from pull-request reviews
Developers and coding agents	`npx @groundrule/cli propose`, or the `propose_rule` tool in the MCP server	One proposed rule, with who proposed it and where	Rules from developers and coding agents

How they compare

Source	What Groundrule reads	What leaves your machine or app	AI	Cost
Repository scan	Every file the scan’s rules look at, on your computer	Results per rule, short redacted snippets around findings (optional), redacted instructions from agent files, tool settings, CODEOWNERS entries	No	Free
Document upload or paste	The file or text you provide	The text, redacted, is stored until reading ends; passages are sent to the model	Yes	Shown before you confirm; a 300-word handbook cost about $0.06
Connected app	Only the page you pick	Same as a document	Yes	Same as a document
PR review import	Inline review comments by people (not bots) in the repositories and period you choose	Comment text, redacted, with its repository, pull request, file, line and reviewer	Yes	Shown before you confirm
`/groundrule rule` in a review	That one comment, or the comment it replies to	The rule text, redacted	No	Free
`propose` and `propose_rule`	Only what you or the agent send	The rule, and the optional reason, example, file and line	No	Free

AI features need an admin to turn on Use AI in this workspace in Settings → AI. They count against the workspace’s monthly AI budget. AI and your data says exactly what each feature sends.

Everything lands in the inbox

Every source ends in the same place: Inbox, as proposals. Each proposal shows:

its kind: Instruction, Tool setting, or Ownership;
its category, and the owners of that category;
where it came from: a file and line, a document heading or page, or a pull-request comment;
an AI label with a confidence, when AI wrote it.

The same rule found twice is one proposal. An instruction in ten repositories is one proposal with ten sources. A rule proposed by three developers is one proposal, “proposed 3 times”. A document imported again merges its rules into the proposals already there.

Accepting a proposal does the real thing: it creates a standard, adopts or turns off a catalog rule, or assigns an owner. See The review inbox.

Where to start

Scan one repository. It takes seconds, needs no AI, and shows what your agent files and tools already say. Start with Scan repositories.
Import your main engineering document, such as a handbook or a security policy. See Import documents.
Ask your team to propose rules as they come up, from the terminal or their coding agent. See Rules from developers and coding agents.
Review the inbox once a week. See The review inbox.