[plugin mimosa] path-traversal call-site false positives on test fixtures + PreToolUse heredoc ergonomics

Environment

• Plugin: mimosa@zcode-plugins-official 1.0.3
• Context: ZCode review gate deep scan (mimosa scan) over a Python repo with security-test fixtures; macOS

Problem 1: path-traversal rule family — call-site matching false positives on test fixtures

When new test files enter the scan's project graph, the path-traversal family reports ~10 highs on open() calls inside test fixture helpers (stubs like fake_orca / fake_ps that simulate CLI tools) plus a few stock open() sites in production sentinel scripts. Same pattern as the SSRF family (see #300): the verdict is call-site-based with no taint/flow modeling — the helper's path argument comes from the test, not from user input, but there is no way to express that. Combined with the missing suppression channel (#300 / #613 / #526), these findings re-fire on every scan and drown the signal.

Ask: same as #300 — verdict-level allowlist + sanitizer/wrapper awareness; or at minimum a scan-scope/profile knob that excludes designated fixture directories from entry-point rules.

Problem 2: hook ergonomics — writing files via heredoc in Bash is misreported as "Bash writes source code directly"

The PreToolUse hook flags git commit -F <file>, PR bodies, or any heredoc that mentions in-repo paths as "Bash directly writing source files", even when no source file is being written at all (the target is a temp file outside the repo, or the string is just a path reference in prose). False-positive ergonomics aside, the message trains users to ignore the hook.

Ask: tighten the heuristic (only fire when the redirect target is actually a repo-tracked source path), and/or record the matched pattern in the denial so users can see exactly what tripped it.

Start by reproducing both cases with mimosa scan: fixture-helper open() calls in the Python project graph and PreToolUse heredoc commands such as git commit -F <file>. Trace which path-traversal and Bash-write patterns produce the findings, then verify that designated fixtures and non-source heredoc targets no longer trigger false positives while genuine violations remain visible.