[SizzleUnrlsd]

Part 2 of #240, first change: the import path. getattr(...)() follows in its own PR.

Summary

New sinks:

• sys.path.insert (second argument; the index is data), sys.path.append, sys.path.extend and site.addsitedir (sitedir) are CODE sinks with selects. Only a check that limits the directory to constants proves them.

How findings are reported:

• sys.path.* → hotspot, Untrusted import path: code runs only when a later import resolves to a module the attacker placed there.
• site.addsitedir → vulnerability, Untrusted site directory: the import lines of the directory's .pth files run during the call. The evidence states the condition: exploitable when the directory holds files the attacker controls.
• The environment and the command line reach these sinks, following the existing source policy.

Coverage is partial, as documented in docs/usage.md: writes to sys.path that are not calls (slice assignment, +=, rebinding) are not reported.

Verification

• tests/test_import_path_sinks.py, new: 23 tests; the 12 positive cases fail on main. They cover:

  • every sink and its argument;
  • non-directory arguments;
  • shape checks (hotspot);
  • constant checks (refuted);
  • environment and command line.

  No existing test changed.

• Full suite green, apart from the local test_packaging metadata issue. ruff check and mypy clean. Regression snapshots unchanged; no new finding on mitmproxy or healthchecks.