Ethics charter template
This is the template for a white-label instance's ethics charter. It is adapted from the reference instance's ETHICS.md, which records a real decision made 2026-10-02: the managed-services front-door model is legitimate MSP-style work, and the work proceeds within the bounds below.
How to use this template #
- Copy this file to
ETHICS.mdin your instance repo (replacing the
reference instance's charter — that's the reference operator's decision, not yours). 2. Fill in every [BRACKETED] field. Don't leave placeholders. 3. Make the binary decision yourself: read the "What this is" / "What this is not" sections, confirm they describe your actual business, date it, and sign it (your name / operator identity). 4. Keep it where your agents can read it. Point them at it the way this template points at the standing rules — the charter's job is to end relitigation, not to start it. 5. Read the adaptation guide at the bottom before you edit the substance.
Ethics Charter — [YOUR BRAND]
The call ([DATE]) #
[This project is / This project is not] an ethical business, and the work [proceeds / does not proceed].
[1–3 sentences: what the business is, in plain language. Example: "The front-door provisioning system exists so a Linux professional can offer managed infrastructure services: helping real, consenting business owners provision, run, and troubleshoot their own Muse instances."]
The assisting agent acts as the operator's tool. The operator — [OPERATOR NAME / IDENTITY], the [PIN holder / trust root] — is the responsible professional; the agent executes within the bounds below.
What this is #
- Managed services. Clients hire the operator to [YOUR SCOPE: e.g.
set up and maintain their Muse containers] — the same relationship as any MSP managing servers. The client owns the account; the operator provides the expertise. - Consented administrative access. The "front door" (shared VM, per-machine reverse-SSH tunnels, per-machine subdomains) is the support channel the client is paying for. It exists so the operator can diagnose and fix the client's systems when something breaks. - Per-client separation. Every machine generates its own credentials (SSH keys, passwords, cookie secrets). Nothing secret is ever copied between machines or between clients.
What this is not #
- Not account farming. No accounts are created deceptively, in bulk,
or on anyone's behalf without their knowledge. Every account belongs to a real client who asked for the service. - Not ToS circumvention. The operator has reviewed the platform's Terms of Service. Nothing here evades platform limits, identity verification, or access controls. - Not non-consensual access. The front door is installed only where the client has engaged the operator for managed service. No system is touched without its owner's consent.
Public display of agents and work #
No agent identity, work summary, or client name appears on any public surface (the /.well-known/manifest.json, the landing page, marketing) without that client's explicit consent, recorded per docs/CONSENT.md. Consent is per-client, scoped, and revocable; when it is revoked, the public display is removed.
Standing rules #
- One client, one set of credentials. Never share keys, passwords, or
secrets across machines or clients. 2. This charter approves the business as described above. If the model ever drifts toward the "is not" list, stop and re-evaluate — do not silently stretch the approval. 3. If the platform Terms of Service change in a way that touches this work, re-read them and amend this file. 4. The agent does not relitigate this call per task. If the facts change, amend the file; otherwise, proceed.
Why this file exists #
The operator asked for a binary decision, once, in writing — because re-establishing the ethical framing in every session is exhausting and erodes trust. The decision is [yes/no]. Future agents: read this before asking whether the work is okay. It is, within the bounds above.
Signed: [OPERATOR NAME], [DATE]
Adaptation guide (read before editing the substance) #
What you may change: the bracketed fields, the business description, the scope of managed services, the operator identity. The chat addendum below may be dropped if your instance doesn't run the chat room — or extended if you add surfaces with the same trust properties.
What you must not remove:
- The "What this is not" list. It is the load-bearing part of the
charter. If your business can't honestly keep all three bullets, the answer to the binary decision is "no" — don't edit the bullets to fit. - Standing rule 2 (the drift clause). The charter approves a described model; the clause is what catches the model changing under you. - Standing rule 3 (the ToS review trigger). Platform terms change; the charter must be re-checked against them, not grandfathered. - The public-display consent section. Your clients' trust in you doesn't transfer to your agents' public listings — that's a separate consent (see docs/CONSENT.md).
When to amend: the model drifts, the ToS changes, you add a new public surface that displays client or agent information, or you take on a client whose situation doesn't fit the template. Amend the file with a dated entry; don't silently stretch the old approval.
The chat addendum (adapt or drop): if your instance runs the chat room, keep an equivalent of the reference addendum — private channels are per-business, digests respect the boundary, no cross-client tasking without consent, the never-paste-secrets rule is enforced server-side. If you don't run chat, delete this paragraph, not the principle: any surface where one client's agents can see another client's information needs the same boundary written down.