Yellow-Control

Author: F.M. Robert Vergnes / robert.vergnes@yahoo.fr Assisted-by: ChatGPT: GPT-5.5 Thinking; Codex

Yellow-Control is a public-safe, Hermes-compatible governance layer for classifying and gating persistent autonomous-agent actions.

Status

Item	Status
Version	v0.1.3
Scope	Narrow Yellow-Control governance skill and human-facing documentation
Hermes compatibility	Compatible skill layout; not officially endorsed by Hermes or Nous Research unless accepted by them
Runtime state	Private, operator-managed, and outside this public repository

What this is

Yellow-Control defines reusable governance doctrine for agents that may change systems, use external services, handle confidential material, or operate persistently. It provides classification rules, policy gates, backup and rollback expectations, external-access governance, secrets-handling rules, repository workflow guidance, and telemetry expectations.

The canonical classifications are:

Acronym	Expansion	Use
ADAL	Agent Delegated Administration Level	Host, identity, administrative, and system-impact authority.
CDEL	Container Delegated Execution Level	Containers, runners, sandboxes, sockets, and delegated execution.
ESAL	External Service Authority Level	APIs, SaaS, repositories, dashboards, servers, accounts, and webhooks.
PCL	Project Confidentiality Level	Information read, written, logged, transmitted, or published.

Why this matters for SysOps, CyberOps, and SSH-based work

Yellow-Control is designed for Hermes runtimes that may connect to external servers through SSH, APIs, repositories, dashboards, or service accounts and perform real operational work.

Yellow-Control keeps a human-controlled governance plane around:

• external server contact;
• SSH access and credential custody;
• delegated elevation such as sudo, constrained wrappers, service restarts, package actions, and recovery paths;
• backup and rollback readiness;
• allow/defer/block decisions before mutation.

Possession of a shell, SSH key, token, dashboard session, or sudo-capable command is not authority by itself. Yellow-Control separates technical reach from accountable authority.

What this is not

Yellow-Control is not a runtime installer, package manager, secret store, project-management framework, skill registry, incident archive, or operational register repository. It does not contain private hosts, private paths, credentials, logs, real register entries, or Oscar-specific runtime material.

Who this is for

Yellow-Control is for maintainers and operators who want a public, installable governance skill for persistent autonomous agents while keeping live operational state private. It is also useful for reviewers who need readable policy mirrors under docs/.

How it works

1. Classify the requested action with ADAL, CDEL, ESAL, and PCL.
2. Check authority, scope, confidentiality, backup, rollback, external-access, persistence, repository, and telemetry gates.
3. Allow only when required evidence and approvals are present.
4. Defer when evidence is incomplete.
5. Block when a request violates policy, exposes private state, self-grants authority, or bypasses governance.

flowchart LR
  A[Request] --> B[Classify ADAL CDEL ESAL PCL]
  B --> C[Evaluate policy gates]
  C --> D{Decision}
  D -->|Pass| E[Allow]
  D -->|Missing evidence| F[Defer]
  D -->|Policy violation| G[Block]

Loading

Repository layout

Path	Purpose
skills/yellow-control-governance/	Installable Hermes-compatible skill package.
skills/yellow-control-governance/SKILL.md	Skill entry point and Hermes metadata.
skills/yellow-control-governance/references/	Packaged operational doctrine used by the installed skill.
skills/yellow-control-governance/templates/	Public-safe YAML templates and schema starters for private runtime records.
skills/yellow-control-governance/references/hermes-backup-snapshot-policy.md	Packaged governance for native Hermes backup and snapshot readiness.
docs/	Human-facing mirrors and explanations for repository readers and maintainers.
CONTRIBUTING.md	Contribution expectations.
SECURITY.md	Public-safe security reporting guidance.
CHANGELOG.md	Version history.

Installable Hermes skill

skills/yellow-control-governance/ is the installable skill package. Copy or install that directory into a Hermes-compatible skills location according to the runtime's skill-install workflow.

skills/yellow-control-governance/references/ contains the packaged operational doctrine so the skill remains self-contained after installation. The top-level docs/ directory mirrors relevant explanations for humans, but installed skill behavior must rely on the skill-local references.

Yellow-Control uses Hermes-compatible skill metadata and non-secret metadata.hermes.config settings for runtime register locations. It does not declare required environment variables and does not require secrets to install.

Runtime registers and templates

The public skill contains templates and schema only. Live runtime registers and decision records must remain private and outside the public skill package and public repository. Runtime register paths are configurable.

Template	Private runtime target
skills/yellow-control-governance/templates/external-access-register.template.yaml	External-access register at yellow_control.external_access_register_path.
skills/yellow-control-governance/templates/governance-decision-record.template.yaml	Decision records under yellow_control.governance_decision_log_dir.
skills/yellow-control-governance/templates/server-first-contact-record.template.yaml	Server-first-contact records under yellow_control.server_first_contact_dir.

Default private runtime locations are configurable through Hermes skill config:

Config key	Default
yellow_control.external_access_register_path	~/.hermes/yellow-control/registers/external-access-register.yaml
yellow_control.governance_decision_log_dir	~/.hermes/yellow-control/decision-records
yellow_control.server_first_contact_dir	~/.hermes/yellow-control/server-first-contact

Do not commit live register files, decision records, first-contact records, private evidence, or local runtime paths to the public repository.

Prerequisites

Requirement	Notes
Hermes-compatible runtime	Must already be installed and operating. Yellow-Control does not install Hermes.
Git	Required for repository workflows.
GitHub, GitLab, or similar	Required only when repository automation is used.
Dedicated agent account or fork	Recommended for repository automation instead of borrowing a human maintainer session.
Backup/checkpoint capability	Required before risky privileged, external, persistent, or runtime-governance actions.
Optional integrations	Telegram, OpenWebUI, gateway, and dashboard integrations are optional runtime integrations, not Yellow-Control requirements.
Optional package layer	External package management is a separate optional implementation layer for target-specific server packages.

Native Hermes backup governance

Yellow-Control does not implement or wrap Hermes backup. Local rollback/checkpoint hygiene uses native hermes checkpoints and the default governance recommendation hermes checkpoints prune --retention-days 30 --max-size-mb 500. Hermes runtime backup archives use native hermes backup, optionally hermes backup -o <path> or hermes backup --quick --label <name> when appropriate. Hermes updates use hermes update --backup or the native updates.pre_update_backup: true setting when owner policy requires full pre-update backup by default.

Checkpoint pruning is not backup archive retention. Backup repositories and artifact stores are for private backup archives, manifests, and metadata, not raw ~/.hermes/checkpoints/ shadow stores. Backup archive retention is mandatory because unbounded archive growth can fill disk or repository storage. Yellow-Control only governs whether a backup or checkpoint is required, whether native or owner-approved mechanisms exist, whether targets are private, whether archive retention is approved, and what public-safe telemetry must be recorded.

Deploying on a real Hermes runtime

This public repository provides the installable skill package, governance doctrine, and public-safe templates. Real runtime deployment still requires a private runtime overlay for local AGENTS/SOUL adaptation, private registers, private evidence, and owner-approved local enforcement implementation.

Use docs/runtime-deployment-pattern.md for the deployment phases, governance boundaries, and validation checklist. Do not commit runtime registers, decision evidence, private paths, or other runtime-private operational material to this public repository.

Safety model

Yellow-Control is conservative by default. Unknown authority, unknown confidentiality, missing external-access registration, missing rollback evidence, or unclear operator custody causes defer or block rather than silent execution.

Public documentation and packaged references must use fictional identifiers, role names, placeholders, and redacted examples. Credentials, tokens, private keys, private hostnames, private addresses, private runtime paths, raw logs, and real register entries must stay out of the public repository.

Community / contributions

Contributions should patch the existing governance layer rather than redesign it. Keep documentation in English, preserve metadata lines, keep examples generic and public-safe, and maintain alignment between docs/ and skills/yellow-control-governance/references/.

Authorship and AI assistance

Yellow-Control is authored by F.M. Robert Vergnes. This repository records AI assistance from ChatGPT / Codex in the document metadata lines. Compatibility with Hermes does not imply official approval, endorsement, or acceptance by Hermes or Nous Research.