AST rules catalog (arch-01 … arch-16)

Gate 2 ships sixteen built-in, Semgrep-compatible architectural rules. They check the structure of a diff — things a regex cannot reliably see. A rule at ERROR severity that matches makes the RFC BLOCKED.

Reference, not source of truth

This catalog summarises the intent of each built-in rule. The authoritative definitions are the .semgrep.yaml files in meridian/core/ast-spec/ in the repo. Exact patterns and supported languages may evolve between versions — check the source for specifics.

The rules

ID	Focus	What it flags (intent)
arch-01	Authentication	Route/handler defined without an authentication check on a path that should require one.
arch-02	Tenant scoping	Database query in a multi-tenant context missing the tenant filter (e.g. WHERE tenant_id = ?).
arch-03	Command execution	Shell/process execution (exec, spawn, system) with interpolated or untrusted input.
arch-04	Unsafe deserialization	pickle, yaml.load without SafeLoader, native deserialization of untrusted data.
arch-05	Dynamic code execution	eval / Function / dynamic require/import on request-derived data.
arch-06	SQL construction	SQL built by string concatenation/interpolation instead of parameterized queries.
arch-07	Secrets in code	Credentials/keys/tokens assigned as literals (structural complement to Gate 1 regex).
arch-08	TLS / cert verification	Disabling certificate verification (rejectUnauthorized: false, verify=False).
arch-09	Authorization checks	Sensitive operation lacking an authorization/role check.
arch-10	Input validation	Endpoint consuming request input without validation/sanitization.
arch-11	Path traversal	File access using unsanitized user-controlled path segments.
arch-12	SSRF	Outbound request to a URL derived from request input without allow-listing.
arch-13	Logging hygiene	Logging of secrets/PII (tokens, passwords, personal identifiers).
arch-14	Error handling	Swallowed errors / empty catch blocks that hide failures.
arch-15	Crypto misuse	Weak or misused cryptographic primitives (e.g. ECB mode, weak hashing for secrets).
arch-16	Dangerous defaults	Insecure defaults (permissive CORS *, debug enabled, open bind addresses).

How a rule reports

A matched rule appears in the RFC under gates.ast.findings:

{
  "gates": {
    "ast": {
      "status": "fail",
      "findings": [
        {
          "id": "arch-02",
          "severity": "high",
          "message": "DB query missing tenant scope filter",
          "line": 42
        }
      ]
    }
  }
}

Severity and blocking

• Rules authored at Semgrep ERROR map to blocking findings → RFC BLOCKED.
• Rules at WARNING are advisory and do not, by themselves, block.
• A genuine false positive should be handled via override, not by deleting the rule.

Extending the catalog

To add arch-17 and beyond, follow the Custom AST rules guide. Remember the current limitation: there is no runtime MERIDIAN_AST_RULES_DIR, so new AST rules require a build step or a plugin (gaps).

Next: Comparisons