Skip to content

Security: waitdeadai/minmaxing

Security

SECURITY.md

Security Policy

minmaxing is a workflow harness that can shape powerful autonomous agent systems. Treat issues involving credentials, runtime authority, generated agents, memory seeds, audit logs, and customer data as security-sensitive.

Supported Versions

Security fixes target the latest main branch unless a maintained release branch is explicitly listed in this file.

Reporting A Vulnerability

Do not open a public issue for vulnerabilities involving secrets, private customer data, bypassable approval gates, unsafe command execution, agent escape, runtime privilege escalation, or kill-switch failure.

Preferred reporting path:

  • Use GitHub private vulnerability reporting for this repository when enabled.
  • If private reporting is unavailable, contact the maintainer privately through the GitHub profile and include only the minimum detail needed to coordinate a safe disclosure path.

Include:

  • A concise description of the issue and affected files.
  • Reproduction steps using dummy data.
  • Expected impact and whether credentials, customer data, external side effects, or system-of-record writes are involved.
  • Suggested mitigation, if known.

Do not include:

  • Real API keys, tokens, passwords, session cookies, customer exports, CRM records, audit logs, or private Hermes memory seeds.
  • Exploit automation against third-party systems.

Security Baseline

Public contributions and generated examples must preserve these rules:

  • No real secrets in git.
  • Least-privilege tool grants for Hermes agents.
  • Runtime actions must declare authority, approval gates, audit events, idempotency keys, and kill switches.
  • REVCLI/Revis-facing agents must route side effects through the governed runtime control plane instead of direct unmanaged system-of-record writes.
  • Public examples must use fake data and safe simulated runtime evidence.

Runtime Policy Matrix

minmaxing supports multiple runtime profiles. Do not blur them together. This clone defaults the committed Claude Code project settings to the trusted-local bypassPermissions posture because the operator explicitly works that way. Treat that as local operator authority, not as a safe default for teams, clients, public forks, or unfamiliar machines.

Profile Intended Use Permission Mode Network / Secrets Required Proof
project-default Operator-owned private local loop bypassPermissions local operator responsibility; secrets still denied by settings valid JSON, explicit warning, secret deny rules, governance hook smoke
solo-fast Trusted single-operator local loop bypassPermissions local operator responsibility; secrets still denied by settings valid JSON, secret deny rules, governance hook smoke
team-safe Shared project and collaborator work acceptEdits narrower allowlist; no broad side effects by default valid JSON, governance hooks, no bypass default
ci-static Public CI and pull requests static only no secrets, no external network requirement shell syntax, static smokes, eval pack, diff hygiene
ci-runtime Authenticated runtime validation isolated test workspace dedicated test credentials only, redacted logs explicit runtime smoke with no production secrets

bypassPermissions is not the recommended team default. It is the explicit trusted-local default in this operator workspace for a single operator who understands the repo and accepts the local risk. Team and CI contexts should prefer team-safe, ci-static, or ci-runtime depending on whether runtime credentials are intentionally present.

Out Of Scope

  • Vulnerabilities in private REVCLI deployments that are not present in this public repository.
  • Social engineering, physical attacks, denial-of-service tests, or attacks against third-party services.
  • Reports that require real customer data or production credentials to verify.

There aren't any published security advisories