Remodeling / documented here, maintained here

A home for the work.
A record of its care.

The projects directory now tracks what needs doing, what is documented, and whether the site’s basic health and security checks are passing.

1 · Project evidenceThe existing catalogue, project pages, briefs and README presence.
2 · Remodeling agentOne persistent cluster service. Shared progress and monitoring history in SQLite.
3 · Visible follow-throughThe public tracker, private operations, and status changes on Simon’s Fedora screen.

What this project delivers

  • A project tracker with progress, owner, next step, filters and personal flags.
  • Documentation coverage: project page, useful agent brief, README and guide. Each is present, missing or not checked.
  • A separate review date. A file existing, or being regenerated, does not prove its contents are current.
  • A persistent monitoring agent with run history, a shared MCP interface and Agent Link reports.

Projects without a recorded progress review start as unreviewed. A “live” listing describes availability, not completion.

Now: a private workspace for the work itself

Each project connects a goal to tasks, named owners, blockers and recorded results. Simon can see five initial projects together, add another from the existing catalogue, or change what is in focus.

Finishing a task records what changed, why, evidence, limitations and the next useful step. That one operation preserves the result, creates any follow-up task and flags documentation for an actual review.

Agent presence is labelled as recent bus activity. It does not prove that an agent is working. Build names are shown with their snapshot date, rather than being presented as a live deployment check.

The journal stays private. Separately authored case-study fields can appear on the Work & outcomes page: the problem, Simon’s contribution, an observed outcome and public evidence.

Open Now on the tailnet → · Agent workflow →

Private work uses the existing SQLite volume, revision checks prevent overwritten edits, and result retries are idempotent. Nightly consistent backups are exported outside the application volume. No private workspace route is exposed by the public website.

Site health and security checks

Loading the running agent’s schedule…

  • The public homepage, tracker, this guide and the project-context index respond with the expected content.
  • The cluster’s site origin responds independently of the public edge.
  • The HTTPS certificate validates and has at least fourteen days remaining.
  • Content Security Policy, MIME sniffing protection, HTTPS policy and referrer policy are checked.
  • Three fixed sensitive-file paths are rejected: environment file, Git configuration and SQL backup.

Results distinguish passed, failed, warning and unknown. These are baseline checks, not a penetration test or a guarantee that the site is secure. The current inline-script policy is reported as a warning.

Remodeling reports changes and a daily quiet digest to Simon through Agent Link. It performs no automatic repairs and makes no paid model calls. The existing five-minute uptime script remains separate.

How to keep tracking useful

  1. After work on a project, tell the connected agent to update Remodeling with the project, stage, owner and one concrete next step.
  2. After reading and checking the documentation, record the actual review date. Reviews older than thirty days are flagged for attention.
  3. Update the project’s README and brief at their source, then publish the site. The documentation inventory refreshes on the next monitoring run.

Example instruction:

Update Remodeling: Screen Pointer is active, Codex owns it, and the next step is to verify a Chrome capture. Leave its documentation review date unchanged.

Agents use remodeling.update_project through MCP Registry. Fields are slug, stage, owner, next and reviewed. Progress is stored on the cluster and shared across devices. Row flags are deliberately personal to each browser.

Everything in the public tracker is public. Keep credentials, patient information and private operational details out of project notes.

Implementation and boundaries

Remodeling runs as one Python service in the Kubernetes cluster, with a persistent SQLite volume. It is a deterministic agent: a schedule invokes a fixed set of tools, stores evidence, compares findings and reports changes. It does not need a language model for these checks.

The public site exposes only read-only project and monitoring summaries. Operations, history and MCP tools are available on the tailnet. A failed connection is shown as unavailable, and the tracker can display its last published snapshot.

Open operations on the tailnet → · Agent brief →

Delivery log

31 October 2026 · First deployment
Published the project tracker and persistent cluster agent. Browser checks passed on desktop and phone widths. The first monitoring run completed with fourteen passing checks and one warning about the existing inline-script policy. Recurring checks remain disabled until Simon chooses the interval.

The current agent status is shown above; a missing status is never treated as a successful check.