Sitebase — database-backed website development and management

Sitebase builds and runs content-driven websites from a database. It models the content a site needs, writes the migrations and the data layer its pages read from, and works directly with the people who write the content: staging drafts, showing them the change, and publishing only with their explicit go-ahead.

Who it's for

Content creators who want their site to keep pace with their material, and the operators who support them. It suits a site whose content lives in a database and is maintained by someone who is not a developer — a publication, a portfolio, a catalogue, a listings site — where the hard part is rarely the page and usually the hand-off between the person with the words and the system that renders them.

What's included

Path	What it is
SOUL.md	The agent's identity: goal, style, authority.
skills/content-schema-design/SKILL.md	Modelling content in a database: real examples first, the columns every content table earns, draft-to-publish as a state, reversible migrations.
skills/website-data-layer/SKILL.md	The only code that touches the database: parameterised queries, idempotent writes, cache invalidation, expand-migrate-contract deploys, verified backups.
skills/creator-collaboration/SKILL.md	The human loop: intake, preview links, one-round-one-list reviews, publication approval, plain-language reporting, privacy boundaries.
skills/agent-reporting/SKILL.md	ARP v1.2.1 reporting profile, so the agent can report on its own work. Ships with its validator, template and the vendored schema.
config/config.template.yaml	Provider, the environments the agent may touch, and optional memory wiring.

Before you start

• Hermes Agent — tested on v0.21.3 (September 2026 release). The config template is written for the present schema: one nested model: block. An older install carrying root-level provider:/model: keys gets hermes doctor complaints ("Stale root-level config keys", "Could not validate model/provider config"); convert the config rather than replacing it, and the template shows the shape.
• A database the agent may write to — PostgreSQL 13 or newer (the skills' examples assume it); SQLite is fine for local work. Give it a connection string in the environment rather than a credential in the config, and name which environment is production.
• A site checkout, and ideally a preview URL, so staged drafts can be looked at before they are live.
• Node.js 18 or newer — needed only by skills/agent-reporting/scripts/validate.mjs, the report validator (a Node ESM script; checked on v22). Hermes itself runs on Python, and nothing else here needs Node.
• No account anywhere. Everything in this pack works against your own site and your own database; nothing in it calls a hosted service.

Quickstart (Hermes)

1. Create a profile: hermes profile create sitebase

2. Copy SOUL.md into the profile as SOUL.md.

3. Copy the skills/ folders into the profile's skills directory.

4. Copy config/config.template.yaml to config.yaml, set your provider and model, and uncomment the environment block so the agent knows which database and checkout it may touch — and which environment is production.

5. First tasks that exercise the whole loop:

   • "Here are five articles; model the content and show me the schema before you build anything."
   • "Stage yesterday's three posts as drafts and send me a preview link."
   • "Add an image-credit field to articles — migration, data layer, and the editor form."

How the work runs

Step	What happens	Skill
Intake	Real examples, not descriptions; who publishes, who approves, what silence means.	creator-collaboration
Model	Nouns and relations, then types; the columns every content table earns.	content-schema-design
Migrate	One intent per migration, stated rollback, additive first.	content-schema-design
Build	One module owns the queries; parameterised, bounded, cached with a version key.	website-data-layer
Stage	Everything lands as a draft with a preview link.	creator-collaboration
Review	One round, one list, state both sides can see.	creator-collaboration
Publish	Explicit go-ahead recorded, live URL confirmed, provenance kept.	creator-collaboration

Memory

Hindsight is recommended and optional: with it the agent keeps durable notes between sessions (which content models it has agreed, what a creator asked for, what is staged and waiting); without it everything works within a single session. Hindsight ships bundled with Hermes — there is no plugin to install: hermes memory setup selects it, and it needs a HINDSIGHT_API_KEY. It is MIT-licensed and developed upstream at https://github.com/vectorize-io/hindsight, documented at https://hindsight.vectorize.io. The config template ships the block commented out.

Pack record

• Pack: Sitebase · v0.1.0 · 2026-10-06
• Runtime declared: Hermes Agent (SOUL + skills + config template) — https://hermes-agent.nousresearch.com/docs
• Reporting: ARP v1.2.1 profile — a periodic report the operator can ask for on demand; it is additive and changes nothing about how the other skills work.
• Classification (OASF v1.1.0 snapshot, leaves only):
  • skills: software_engineering/web_development/fullstack_development, data_engineering_analytics/data_modeling_design/schema_design, data_engineering_analytics/databases/relational_database_operations, data_engineering_analytics/data_modeling_design/database_migration
  • domains: telecommunications/internet_services, media_and_entertainment/publishing
• Licence: MIT for this pack's own content (see LICENSE).
• Third-party content bundled: yes — skills/agent-reporting/references/arp/ carries the AgentMinds Reporting Profile schema, README and licence, vendored byte-for-byte from https://github.com/agentmindsdev/profile at commit 51e335aca582460cf14a409ccbe18780c8c78e9f (© AgentMinds, CC-BY-4.0, unmodified; its licence travels with it at references/arp/LICENSE). The vendored schema's own title field reads "ARP v1.2.1" while that commit is the repository's v1.3.0 release — the commit is the pin, and neither version string on its own describes these bytes. Nothing else third-party is bundled.
• Archive pin: this pack cannot contain its own sha256, so the archive's hash, byte size and file count are recorded in the platform upload record for this version (the changeNote), and every per-file hash can be re-derived from the archive itself.
• Hygiene: built from generic templates. The scrub ran twice — over the source tree and over the unpacked archive — with per-file sha256 diffed between them, so the archive provably equals what was scrubbed. Removed for this archive: machine paths and usernames; credentials and any token-shaped strings; operator and third-party identifiers; internal task references; and the fleet-specific examples and namespacing inside the reporting skill were rewritten generically for this release. Nothing personal, no client material.
• Provenance: authored for release as a general-purpose pack. The publishing account's credit line is carried on the platform's own record for this upload.