MindFluence is a prompt-only skill. It does not execute code, collect analytics, or store user data. The security surface is therefore mostly about what the model is instructed to generate and how the repository is handled.

Please do not open a public issue for a security problem.

Use GitHub's private vulnerability reporting:

1. Open the repository's Security tab.
2. Select Report a vulnerability.
3. Include the affected version, the prompt or file, the expected behavior, and the observed behavior.

If private reporting is unavailable, open a regular issue that says only "security report — please contact me privately" and wait for a maintainer to respond. Do not include exploit details, secrets, or personal data in a public issue.

• Prompt injection / instruction bypass that makes the skill ignore its refusal policy, ethical boundaries, or provenance gate.
• Fabricated evidence produced by the skill's own instructions — for example, a check that rewards invented numbers, names, testimonials, studies, journals, or years.
• Unsafe marketing guidance that violates the refusal policy (tobacco, gambling, weapons, minors, fraudulent health claims, MLM, hate, exploitation of vulnerable groups).
• Secret exposure in the repository, examples, logs, or local tooling.
• Supply-chain risks in scripts used to build, test, or ship the skill.

• The truthfulness of third-party statistics or quotes that a user supplies.
• Marketing performance claims (conversion rates, CTR, revenue) — this skill does not guarantee outcomes.
• Legal or regulatory compliance in a specific jurisdiction. Users are responsible for verifying claims, disclosures, and consent requirements.
• Vulnerabilities in third-party LLM providers or agent platforms.

• Treat every generated name, number, quote, and citation as unverified until you check it.
• Keep placeholders such as [N], [Customer name], and [Source, Year] in drafts until real data is available.
• Never paste secrets, API keys, or personal data into a prompt.
• Do not use the skill to fabricate testimonials, scarcity, authority, or health claims. The skill's ethical boundaries prohibit this, and the provenance gate is designed to catch it.

• _ship/ is local release automation and is intentionally ignored by git. It must never be committed. It can read a GitHub token from the local credential store; treat any machine that runs it as a trusted environment.
• Do not commit .env files, tokens, customer data, or private prompt logs.
• If a secret is committed, rotate it immediately and remove it from git history before publishing.

We aim to acknowledge reports within 7 days and to publish a fix or mitigation within 30 days for confirmed issues. Credit is given to reporters unless they prefer to remain anonymous.