Skip to content

Site pipeline

PULSE can run a company website’s content loop end to end: it writes blog posts and publishes them to the site’s Hugo repository, audits the live site every day, and opens pull requests that fix what the audit finds. Runink runs its own site this way. Each deployment points PULSE at its own site. There are no defaults that name Runink.

Blog posts to your Hugo site

How a post gets there

  1. A post is written. Either the social engine drafts one blog article per active campaign per day, or you approve blog-channel content yourself.
  2. It is scheduled. The social engine schedules its draft two hours ahead. That window is a chance to cancel it on the Schedule screen before the worker dispatches it.
  3. It is published over git. PULSE renders the content as a Hugo post at <blog dir>/<slug>.md and, by default, opens a pull request on branch pulse/publish-<slug>. Nothing goes live until someone merges it.

No third-party social API is involved. The post goes to your own repository with your own GitHub token.

Settings

VariableDefaultMeaning
PULSE_APPLY_ENABLEDoffMust be exactly true.
PULSE_AUDIT_APPLYoffAudit-fix apply/commit/pr only (not the blog). Must also be on for those modes.
PULSE_APPLY_ROOTnoneThe only directory an audit-fix repo_path may point into. Unset refuses every mode.
PULSE_PUBLISH_BLOGfollows the globalSet false to silence only the blog.
PULSE_PUBLISHER_USER_IDSempty (nobody)User ids whose content may publish to the site. Others are withheld, and their campaigns get no daily draft.
PULSE_SITE_REPOnone, requiredowner/repo of your site.
PULSE_SITE_BRANCHmainBase branch.
PULSE_SITE_BLOG_DIRcontent/blogWhere posts go.
PULSE_SITE_PUBLISH_MODEprpr opens a pull request. commit pushes to the base branch and is live at once.
PULSE_SITE_AUTHORomittedThe author in the post’s front matter.
PULSE_SITE_BASE_URLomittedYour site’s base URL, used for the canonical link.

The GitHub token comes from the GitHub connection of the user who owns the content. See Connections.

When it refuses

MessageCause
website publishing is disabled: ...The blog channel is not armed. The message names the layer that decided.
no github integration/token configured for user ...The content’s owner has no GitHub connection.
PULSE_SITE_REPO is not set — refusing to publish; set it to YOUR site repo as owner/repoNo target repository.
invalid site repo "...": expected owner/repoThe repository is malformed.
refusing to publish empty contentThe post has no title or body.

The daily blog drafting only runs while the blog channel is armed. A disarmed channel costs no model time.

The daily site audit

Once a day, at 03:47 UTC, the site audit engine runs the relevance audit against one URL and stores the full report. It also files a readable digest as content on the site_audit channel. That channel is not publishable: the digest exists to be read.

VariableMeaning
PULSE_APPLY_ENABLEDMust be exactly true, or the engine does not start at all. One autonomy switch covers the engine and the publishers alike.
PULSE_SITE_AUDIT_URLThe site to audit.
PULSE_SITE_BASE_URLUsed when PULSE_SITE_AUDIT_URL is unset.

Disarmed, the engine logs Site Audit Engine: NOT started — autonomy is disarmed (PULSE_APPLY_ENABLED is not "true"). Armed but with no target, it logs Site Audit Engine: NOT started — no target site; set PULSE_SITE_AUDIT_URL (or PULSE_SITE_BASE_URL).

The first audit after a deploy waits for the next 03:47 UTC slot. For a score now, start a New site audit on Brand Analysis; its full report is the site-score report. Each audit is limited to 10 minutes. A run that scored nothing, for example because the site was unreachable, is recorded as a failure rather than a result.

The engine audits and records. It never fixes. Fixing stays a human decision.

Audit-fix pull requests

From a relevance audit report, a person can apply the fix artifacts to a site repository (ApplyAuditActions). See Apply the fixes.

  • Every mode, plan included, needs PULSE_APPLY_ROOT set and a repo_path that resolves inside it. Anything else gets one generic refusal.
  • plan is a dry run. Like every ApplyAuditActions call, it needs an editor or admin (the RBAC write tier).
  • apply, commit, and pr need PULSE_APPLY_ENABLED=true, PULSE_AUDIT_APPLY=true, and a signed-in administrator. The role is read from the users table on every call, not from the token.
  • In pr mode with a GitHub connection, PULSE opens the pull request against the connection’s owner and repo, on branch seo/pulse-remediation by default, against main.
  • Copy and infrastructure changes are returned as manual and never applied.

Fixes proposed to the customer’s own repository

Without any server checkout, a connected website repository (Settings › Integrations, connected by an administrator with a fine-grained token for that one repository) lets an administrator press Propose as pull request on an audit finding. PULSE reads the repository through the GitHub API, places the fix, has the plan gate judge it, and opens one pull request from a new pulse/<audit>-<finding> branch. It never pushes to the base branch and never merges. It needs PULSE_APPLY_ENABLED=true and PULSE_AUDIT_APPLY=true, but not PULSE_APPLY_ROOT. See Website repository.