fix(security-guidance): add exclude_substrings to cut false positives

[zhang-liz]

Summary

The security-guidance plugin's content-based rules use plain substring matching, which produces well-known false positives for several common identifier and call shapes:

• eval( matches ast.literal_eval( (Python stdlib's safe parser) — see security-guidance plugin: eval_injection rule falsely flags ast.literal_eval() #55464
• exec( matches db.exec(...), stmt.exec(...), regex .exec(/.../), doc/code-fence mentions, etc. — see [BUG] security-guidance hook blocks markdown/doc writes containing "exec(" substring #46720
• new Function matches new Functionality, new FunctionComponent
• pickle matches pickled, pickle_jar, and bare import pickle / pickle.dump (serialization, not deserialization)

This PR adds an optional exclude_substrings field to the SECURITY_PATTERNS schema. A rule matches only if at least one occurrence of its trigger substring is not inside any excluded substring, so:

• Files containing only excluded shapes are skipped (no warning)
• Files containing both an excluded shape and a real risk shape still fire on the real one
• Rules without exclude_substrings behave exactly as before (backward compatible)

Schema follows the design proposed by the issue author in #55464.

Case matrix

Content	Trigger	Before	After
value = ast.literal_eval(raw)	eval(	flagged	not flagged
result = eval(user_input)	eval(	flagged	still flagged
ast.literal_eval(a); eval(b)	eval(	flagged	still flagged (on the real eval(b))
db.exec(schema)	exec(	flagged	not flagged
child_process.exec(cmd)	exec(	flagged	still flagged
\exec(query)`` in markdown	exec(	flagged	not flagged
new Functionality	new Function	flagged	not flagged
new Function('return ' + x)	new Function	flagged	still flagged
import pickle; pickle.dump(o,f)	pickle	flagged	not flagged
import pickle; pickle.load(f)	pickle	flagged	still flagged

Test plan

• python3 -m unittest plugins/security-guidance/hooks/test_security_reminder_hook.py — 25 tests, all pass locally
• Path-based GitHub Actions workflow rule still fires
• Existing content rules with no exclusions (dangerouslySetInnerHTML, document.write, innerHTML, os.system) unchanged

Closes / refs

• Closes security-guidance plugin: eval_injection rule falsely flags ast.literal_eval() #55464 (ast.literal_eval false positive)
• Refs [BUG] security-guidance hook blocks markdown/doc writes containing "exec(" substring #46720 (markdown exec( false positive — addressed at the substring level rather than via a file-extension allowlist, so source-file false positives are also covered)
• Alternative to fix(security-guidance): skip doc files for substring checks #47514, which scopes substring checks off for doc-like extensions. The two approaches are complementary; happy to defer or combine if a maintainer prefers that route.

🤖 Generated with Claude Code