fix(ea1): bound wildcard-grant pattern to a single line and a standalone '*'

[rootkiller6788]

What

Fixes #405. The EA1 wildcard-tool-access pattern

r"(?:tools?|permissions?)\s*:\s*\[?\s*['\"]?\*['\"]?\s*\]?"

used \s* between the colon and the expected * value. Python's \s matches newlines regardless of re.MULTILINE, so the pattern could span a blank line and bridge two unrelated headings (For each tool:\n\n**Input Schema:** -> matched text tool:\n\n*), and it treated the first * of a markdown bold span (**API Coverage vs. Workflow Tools:** -> Tools:*) as a wildcard grant. Both surfaced as EA1/MEDIUM false positives on benign skills.

Fix

• Constrain the gap between the colon and the value to the same line ([ \t]* instead of \s*).
• Require the matched * to be a standalone token: add (?!\*|\w) so markdown bold (**) and word-adjacent * (e.g. a YAML anchor like *ref) no longer match.

r"(?:tools?|permissions?)\s*:[ \t]*\[?[ \t]*['\"]?\*(?!\*|\w)['\"]?[ \t]*\]?"

Genuine single-line wildcard grants still match: tools: "*", tools: [*], permissions: '*', tool:*.

Tests

• Added tools: * and permissions: '*' to the existing EA1-positive parametrization in tests/unit/test_patterns_new.py.
• Added TestExcessiveAgency::test_ea1_no_false_positive_on_markdown_bold_or_heading covering the cross-heading bridge, bold-markdown collision, and a bolded list of specific named tools.

Verified: pytest tests/unit/test_patterns_new.py (368 passed) and pytest tests/nodes/analyzers/test_static_patterns.py (151 passed); ruff check and ruff format --check clean.

[yashrajp22]

Closing as a duplicate: the identical production change already landed on main via #417 (fix(ea1): bound wildcard-tool-access match to a single line and a standalone asterisk, merged 2026-08-24), two days before this PR was opened — which is why this branch now shows merge conflicts.

Comparison against #417:

• The regex change is byte-identical to what fix(ea1): bound wildcard-tool-access match to a single line and a sta… #417 merged.
• The test cases added here (cross-heading bridge, bold-heading collision, bolded named-tool list, bare tools: *, permissions: '*') are all already covered by tests/nodes/analyzers/test_ea1_wildcard_line_boundary.py from fix(ea1): bound wildcard-tool-access match to a single line and a sta… #417.

Thanks for the contribution and the clear write-up regardless — the explanatory comment above the regex was a nice touch, and the underlying analysis matches what was merged. Remaining gaps in this pattern (footnote-legend false positive, YAML block-list / JSON quoted-key wildcard grants) are being tracked in follow-up issues.