agentic-startup-generalist

A generalist operator for an early-stage venture, packaged as a Hermes agent — runnable standalone or managed via HSM (Hermes Swarm Map).

In a venture with no specialists and no slack, this agent closes open loops: it researches, gathers and assesses evidence for product-market-fit questions, turns meetings into tracked work, keeps the board and the cadence honest, and supports the founder directly. Its job is not to make the consequential calls — it is to arrive at each call with the evidence assembled, the options framed, and the tradeoffs named, so a human can decide in five minutes what would otherwise take a week.

The methodology it runs is a single lifecycle applied to every unit of work, from a two-hour research question to a six-week PMF experiment:

FRAME → GATHER → ASSESS → RECOMMEND → HANDOFF

RECOMMEND is the phase that defines the agent and the one most likely to be skipped under pressure — it frames a gated decision for a human instead of walking through it. Per-duty skills (PMF; turning meeting transcripts into warm updates and candidate tasks) extend this spine and are added as they earn their place.

---

Open-core

The whole self-hostable package — the methodology skill, the core tools, and the agent's soul — is free and public. Run it yourself with your own keys and Docker via HSM. The hosted, managed agent and premium skills are the paid tier; nothing here is crippled to sell the upgrade. If you want to run it yourself, everything you need is in this repository.

---

Multiplayer by design

This package targets hermes-agent-mt — the multi-tenant Hermes runtime — not a single-user harness. One deployment can serve many users and contexts (Signal DMs, group chats, tenants) at once, which is the whole reason the privacy model has two levels rather than one: what your repository shows the world, and what one user of a running deployment can see of another. The instance overlay, the sanitization gate, and the glocal cross-context read floor are all here because multiplayer is the default, not an afterthought. If you only ever run one venture against one agent, this still works — you just won't need half of it. See docs/privacy-and-visibility.md.

The package layer itself is runtime-agnostic (a plugin + a skill + a soul + an optional engine), but the scaffolding — base image, instance bootstrap, HSM consumption — assumes the hermes-agent-mt image as its base. See docs/onboarding.md.

---

Two ways to run

Standalone — clone the repo, mount hermes-plugin/ into an existing Hermes instance, and start the agent. See docs/onboarding.md.

HSM-managed — register the package in HSM, which handles image builds, env injection, and lifecycle. See docs/onboarding.md.

Once source and surface are bound, the transcript → task pipeline runs on its own; see docs/transcript-pipeline.md.

---

What you get

• The methodology skill — hermes-skill/SKILL.md encodes the FRAME → GATHER → ASSESS → RECOMMEND → HANDOFF spine, with per-duty extensions for PMF and the transcript pipeline.
• The agent's soul — docker/SOUL.template.md carries the standing disposition: evidence over assertion, stage honesty, frame-the-gate-don't-walk-through-it, and the action policy that divides "drive freely" from "propose; a human disposes." docs/examples/SOUL.example.md is a filled-in example for one concrete persona.
• A transcript → task pipeline — turn meeting transcripts into warm updates and candidate tasks on whatever surface a venture uses (a markdown file, a tended board, or any external tracker via a config-declared adapter), with an inbox-lane safety property so automation never writes to a live board unattended. Sources are pluggable too: paste text, read a file, read a Notion page, or inject a reader for any remote system.
• hermes-plugin/ tools — seven real tools exposing the engine to the agent: transcript fetch and task routing, the stage-honest PMF read, retention-curve reading, experiment falsifiability checks, positioning drafts, and decision framing.
• Two-layer sanitization gate — deterministic regex (secrets, PII, API keys) as a hard-fail floor; an LLM-semantic scan for venture particulars routed to human review. Both run on every PR diff.
• Instance-overlay .gitignore — ventures/, instance/, .overlay/, and .env are private by construction. A venture's operational data cannot accidentally reach the shared package.
• CI workflow — .github/workflows/sanitization.yml runs scanner self-tests and the diff-mode gate on every PR.

---

The model

run privately under operational pressure
  → generalize + sanitize
  → contribute through the gate

You build under real conditions in your private instance. When a capability proves itself generic, you strip the particulars, run the gate, and open a PR. The gate catches what you missed. A maintainer makes the final call. That loop — not upfront design — is how the shared package stays both useful and clean.

---

Layout

Path	What it's for	Customize?
engine/	Domain logic: transcript sources, task surfaces, routing, pipeline types	Yes — the core
example-collectors/	Example data-source integrations; copy and adapt	Yes — add real collectors here
hermes-plugin/	Plugin register() + tool definitions for the agent	Yes — expose the engine as agent tools
hermes-skill/SKILL.md	Agent methodology: how to reason, grade sources, iterate	Yes — the reasoning patterns
docker/SOUL.template.md	Agent identity / soul (seeded by instance-setup.sh)	Yes — set role, defaults, tone
docs/examples/SOUL.example.md	Filled-in soul example (one concrete persona)	Copy/adapt — not wired
docker/Dockerfile	System-level deps for the agent image	Only if you need system deps
ventures/	Per-venture working data — gitignored, stays local	N/A — ignored by construction
instance/ .overlay/	Operator-local overrides, instance-specific config	N/A — ignored by construction
scripts/check_sanitization.py	The sanitization gate — runs in CI and via gardener	No — leave alone
.github/workflows/sanitization.yml	CI gate	No — leave alone
.gitignore	Instance-overlay privacy model	No — leave alone
sanitize.config.json	The one tuning surface — teach the gate your venture's particulars	Yes

---

Privacy

A venture's operational data is private by construction (.gitignore) and by automated scan (the sanitization gate). See docs/privacy-and-visibility.md.

---

Contributing

See CONTRIBUTING.md for the three ways to add a capability and the sanitization workflow.

When a capability is ready to promote from your private instance back to this shared package, see docs/promotion-and-upstream.md.