If you find a security issue in this collection — plugin configs, Skills, the bundled PostgreSQL MCP, credential handling, path handling, or read/write defaults — report it privately to the HiAI organization. Do not open a public issue for a vulnerability.

For vulnerabilities in upstream third-party MCPs or CLIs that a plugin integrates, please also report to the relevant upstream project. HiAI does not control first-party vendor MCP infrastructure (GitHub, Sentry, Figma, Cloudflare, Notion hosted endpoints).

• Plugin manifests (plugin.json, mcp.json) and their configs.
• Agent Skills (instructions and behavior).
• The bundled HiAI PostgreSQL MCP (plugins/postgresql/packages/postgres-mcp).
• Documentation accuracy around authentication and security.

• Never commit or share credentials: .env values, API keys, OAuth tokens, PATs, database passwords, private keys, credential-bearing DATABASE_URLs, or Authorization headers.
• Read/write defaults: most plugins are read/inspect-first; verify the posture in each plugin README before granting access.
• Remote MCP endpoints require host-managed OAuth — the plugin itself never embeds tokens.

See docs/TRUST_POLICY.md for the upstream trust model, and each plugin's docs/ for per-plugin security behavior (e.g. PostgreSQL server-side read-only enforcement and its security verification matrix).