Search Practice Fleet — Genesis

Author: James White — https://www.linkedin.com/in/james-alexander-white/

Sharing yours: if you build a fleet like this one, share it on TalentDelta.ai with the Agent Pack Builder — https://talentdelta.ai/agent-templates/agent-pack-builder.

A six-agent fleet that runs the file keeping, pipeline and billing of a small search-and-placement practice, on top of Genesis: an Obsidian vault protocol where every file has a type, every type is validated, and the vault — not a chat log — is the system of record.

Source material arrives and becomes records. Records feed a pipeline. Pipeline movement raises tasks. Closed placements raise invoices that hand off to the practice's accounting package. Five role agents do that work on their own cadences, one coordinator keeps them honest, and the operator decides anything that involves money, contracts or the practice's own rules.

What is in this pack

README.md                     this file
SOUL.md                       the coordinator — the fleet's lead agent
LICENSE                       MIT, for this pack's own content
config/config.template.yaml   profile config template (copy once per agent)
agents/                       the five role agents, one SOUL each
  transcript-intake/          source material → vault records
  correspondence-intake/      mailbox → record updates (read-only on the mail)
  practice-assistant/         day-to-day practice work, on request
  technical-lead/             schema, skills, conventions, wiring
  housekeeping/               frontmatter and canvas compliance
skills/                       the playbooks the roles work from
  genesis-core/               the vault's rules: type chain, parents, naming, validation
  vault-operations/           safe read/query/write patterns for the vault
  source-intake/              one filing pass, start to finish
  correspondence-intake/      the mailbox review, materiality test, handoffs
  practice-pipeline/          pipeline stages, tasks, and the board between roles
  billing-handoff/            placement → invoice → ledger handoff → reconciliation
  fleet-orchestration/        routing, cadence, escalations, reporting
  agent-reporting/            ARP v1.2.1 reporting profile (third-party, see Pack record)
vault/                        the Genesis vault setup itself — see vault/README.md
  README.md                   what Genesis is, and how to install it
  fileclasses/                12 type definitions (the schema)
  dropdown_values/            9 controlled vocabularies
  dashboards/                 5 canvas dashboards shipped as group skeletons
  bases/                      9 saved views
  scripts/                    2 guard scripts (validation, canvas compliance)

The fleet

Agent	Cadence	Does
Coordinator	Continuous	Routes work, verifies runs, reports to the operator
Transcript and source intake	Scheduled	Files incoming material: entity, transcript, summary, event, archive, tasks
Correspondence intake	Scheduled	Reads the mailbox, updates records, logs events, hands new entities on
Practice assistant	On request	Research, briefs, drafts, pipeline and board work — prompt-driven, never scheduled
Technical lead	On request	Owns the schema, the skills and the conventions; gates changes
Housekeeping	Scheduled	Audits and corrects frontmatter and canvas compliance

The assistant is deliberately not on a schedule. An agent that runs on a timer with nothing to do invents work, and invented work inside a live pipeline is worse than no work.

Install

1. Runtime. Hermes Agent — https://hermes-agent.nousresearch.com/docs. The pack is a SOUL, a set of skills and a config template; nothing else is required to start.

2. Profiles. Create one Hermes profile per agent — the coordinator plus the five roles in agents/. Copy config/config.template.yaml into each profile as config.yaml and set name.

3. Skills. Install skills/* into each profile by the table below.

   Profile	Skills
   coordinator (SOUL.md)	genesis-core, vault-operations, practice-pipeline, fleet-orchestration, agent-reporting
   transcript-intake	genesis-core, vault-operations, source-intake, agent-reporting
   correspondence-intake	genesis-core, vault-operations, correspondence-intake, agent-reporting
   practice-assistant	genesis-core, vault-operations, source-intake, practice-pipeline, agent-reporting
   technical-lead	genesis-core, vault-operations, fleet-orchestration, agent-reporting
   housekeeping	genesis-core, vault-operations, practice-pipeline, agent-reporting

4. Vault. Install vault/ into an Obsidian vault following vault/README.md, which also explains the type chain, the naming rules and the compliance checklist.

5. Cut over. The shipped schema was written for a search-and-placement practice, and its example entity links (_id_various_people__practice, _id_candidates__revenue, …) are the vault's own taxonomy, not yours. Re-point them at your own entity tree before the first run. The technical lead's first job is this cut-over, and the housekeeping agent's first pass will find anything you missed.

What the fleet automates, end to end

source material ──► intake ──► vault records (entity, summary, event, archive)
external sources ──► collection runs ──► dated research files
correspondence  ──► record updates ──► events, new-entity handoffs
records         ──► pipeline stages ──► tasks
closed placement ──► invoice draft ──► journal handoff file ──► your accounting package

The fleet stops at the door of the accounting package. It prepares the journal file, writes the handoff note and reconciles monthly; a person imports it. No agent holds a credential to the ledger.

Reporting

Each agent carries the ARP v1.2.1 reporting profile (skills/agent-reporting/), so any of them can produce a structured report of its own work on request — what it did, what it could not do, and the evidence. It is additive: it changes nothing about how the rest of the skills work.

Pack record

• Pack: Search Practice Fleet — Genesis · v0.1.0 · 2026-10-07
• Runtime declared: Hermes Agent (SOUL + skills + config template) — https://hermes-agent.nousresearch.com/docs
• Shape: a fleet pack — one coordinator SOUL plus agents/ and vault/ alongside the usual top-level files. A single-role pack is one SOUL and 2–4 skills; this one carries five roles and their playbooks, because the fleet is the deliverable.
• Classification (OASF v1.1.0 snapshot, leaves only — each verified present, exact-match, no children):
  • skills: ai_ml_engineering/agent_development/agent_application_building, research_knowledge_productivity/knowledge_management/knowledge_base_management, research_knowledge_productivity/project_task_management/task_management, business_professional/finance_accounting/bookkeeping_invoicing
  • domains: human_resources/recruitment
• Author: James White — https://www.linkedin.com/in/james-alexander-white/. This is the one personal identifier in the archive, and it is deliberate: it is the author's own, given for attribution (see LICENSE).
• Licence: MIT for this pack's own content (see LICENSE).
• Optional memory: the config template carries a commented memory: block. If you give the agents long-term memory, Hindsight (MIT) is what the template is written for — https://github.com/vectorize-io/hindsight, documented at https://hindsight.vectorize.io. It is optional; the fleet works without it, and nothing here vendors it.
• 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: an archive cannot contain its own hash, so the archive's sha256, byte size and file count are recorded in the release note that accompanies this pack. Every per-file hash can be re-derived from the archive itself.
• Hygiene: the scrub ran twice — over the source tree and again over the unpacked archive — with per-file sha256 diffed between the two, so the archive provably equals what was scrubbed. Removed for this archive: the practice's name and its entity taxonomy where it identified a real business; every third-party person, client and candidate name; mailbox addresses, mail-server and internal host names; machine paths, usernames and share names; credentials, tokens and anything token-shaped; and the canvas dashboards' contents — the five canvases ship as group skeletons with 341 content nodes removed, because those nodes were the practice's real people on its real pipeline. Also removed: internal review, version and release references. Nothing personal, no client material and no confidential material of any kind survives in this archive; the author's own byline is the deliberate exception, and it is his.
• Provenance: authored for release as a general-purpose pack, from a working system, generalised for anyone running a small search or placement practice. The vault schema is a live protocol transcribed into templates; the roles are archetypes, not any particular set of people.