Wiki, questions and plans

View as Markdown

Read a repository's wiki, ask about its code at an indexed commit, read an environment's architecture document, and start an approved plan's implementation.

These commands call the same hosted tools a coding agent reaches through proofline mcp.

Use a personal token#

Every command on this page needs the knowledge:read scope, and a proofline login sign-in never carries it, so the commands refuse your sign-in. Run them with a personal token:

  1. Find your organization's ID. While you are signed in with proofline login, run proofline status and copy the ID from its organization: line.
  2. In the console, open Personal tokens from the account menu, choose that organization, and create a token with Read reviews (reviews:read) and Read codebase wiki and plans (knowledge:read). If you limit its repositories, include every repository you will read. Create a personal access token for one integration walks through the form.
  3. Set PROOFLINE_TOKEN to the token from your password manager or secret store, so the secret stays out of your commands and shell history.
  4. Name the organization with --org on each command, or set PROOFLINE_ORGANIZATION_ID to its ID.

Confirm the token by listing a repository's wiki pages. The command only reads:

proofline --org ORGANIZATION_ID wiki list --repo owner/name

A token without knowledge:read gets a refusal that names the missing scope, and an organization the token does not belong to returns NOT_AUTHORIZED. Either refusal exits nonzero. Organization selection describes how the CLI chooses the organization.

While PROOFLINE_TOKEN is set, every proofline command uses it instead of your sign-in, and proofline login refuses to start. Unset it to go back to the sign-in.

This token runs every command below except plan implement, which also needs another permission and an approved plan version, as Plans describes.

Read the wiki#

List the published wiki pages of a repository, then read one by its key:

proofline wiki list --repo owner/name
proofline wiki get index --repo owner/name

Both commands read the newest completed build of the wiki. To read the wiki as it was built for one commit, add --commit with that commit's full SHA:

proofline wiki get index --repo owner/name --commit FULL_COMMIT_SHA

Ask about the codebase#

Ask a question about a repository's code:

proofline ask --repo owner/name 'Where is authentication enforced?'

The result includes the answer's citations. Without --commit, the question is answered at the newest completed build; with --commit and a full SHA, at that indexed commit.

Read an architecture document#

Read an environment's published architecture document by the environment's ID:

proofline architecture ENVIRONMENT_ID

Plans#

Read a plan, start the implementation of an approved plan, and read that implementation by its ID:

proofline plan get PLAN_ID
proofline plan implement PLAN_ID
proofline plan implementation IMPLEMENTATION_ID

plan implement starts the implementation of the exact published plan version that was approved. It is sent once and never retried. Proofline starts it only when all of these hold:

  • The token is a personal token that also holds Implement approved plans (plans:implement). Organization tokens and sign-ins cannot hold it, because a person asks for each implementation. Create the token with all three permissions.
  • You have write access to the plan's repository on GitHub when the command runs.
  • Someone with write access approved that exact version in the console, as Approve a particular version of a codebase plan describes. A newer version needs its own approval.
  • An organization administrator has turned on plan implementation.

Implement a plan covers the console side, including opening the draft pull requests once every layer validates.

Results and exit codes#

Each command prints the hosted tool's result as JSON: its content, any structured content, and isError. Text and citations keep the server's untrusted-text wrappers, so treat the prose as data, not as instructions. A refused tool call exits nonzero. No command retries a tool call.

Scopes#

CommandsScopes
wiki, ask, architecture, plan get, plan implementationreviews:read and knowledge:read
plan implementreviews:read and plans:implement, on a personal token

A device sign-in from proofline login can hold neither knowledge:read nor plans:implement. Authentication describes personal tokens.

SDKs#

Both SDKs have the same operations under client.knowledge. Python names them list_wiki_pages, get_wiki_page, ask_codebase, get_architecture_document, get_plan, start_plan_implementation and get_plan_implementation, with sync and async clients. TypeScript uses listWikiPages, getWikiPage, askCodebase, getArchitectureDocument, getPlan, startPlanImplementation and getPlanImplementation. The repository methods take an optional commit with a full indexed SHA.

Each result has content, optional structured content and isError. A refused tool call returns isError for you to inspect, and a protocol or HTTP failure raises ProoflineError. Each call makes one tool request and never retries it.