SkillSpector 2.13.2

This patch release builds on the published v2.13.1 maintenance release and
backports three fixes for false incomplete-analysis reports. It preserves the
existing maintenance fixes, third-party dependency lock, and provider defaults.

Fixes

• #779: Recognize command wrappers
  with no wrapped command at a proven clause boundary. Markdown table cells and
  ordinary source references to timeout, sudo, nice, and xargs no longer
  cause false parser-limit reports. Redirections, unfinished fragments, and
  commands in subsequent clauses retain their existing checks.
• #917: Treat sudo -n as a flag
  and parse descriptor redirections such as 2>&1 as complete operators. Align
  the MCP TP4 batch limit with its existing 128-file limit so supported inputs
  are not prematurely truncated after 64 batches. Byte, token, declaration,
  file, runtime, and provider-error limits remain enforced.
• #921: Recognize direct template
  messages on verified built-in Node assertion imports, avoiding false shell
  substitution limits in JavaScript helpers. Classify .mjs and .cjs files
  consistently, including archive members. Raw security findings and their
  source locations are preserved. Ambiguous syntax, assertion aliases or
  mutations, effectful interpolation, and exceeded bounds remain conservative;
  TypeScript retains its existing parser path.

Scope and upgrading

The JavaScript correction is a bounded assertion-message proof, not a general
JavaScript parser or dataflow analysis. It assumes ordinary built-in assertion
semantics and does not prove the absence of arbitrary runtime mutation or
object getters. Existing findings and independent inspection limitations remain
visible; these fixes do not suppress malformed provider responses.

Regenerate and review exact version-bound finding baselines for 2.13.2 rather
than relabeling baselines from another scanner version. The GitHub release
provides a wheel and source distribution; the publication workflow does not
publish to PyPI.

Validation and known limitations

This release preserves every v2.13.1 commit and includes #779, #917 and #921. All six candidate CI jobs passed, including 12,395 unit tests. Fresh local packaging checks, 72 offline acceptance scans and 22 additional boundary checks also passed.

Live acceptance completed all 28 planned cases: 26 passed and two large-skill TP4 cases failed under the unchanged 600-second workflow budget. The candidate planned 118 batches, but completed 65 from source and 66 from the installed wheel; all remaining ranges were explicitly reported as runtime-limited. The old 64-batch planning cap is removed, but complete live inspection of this large input is not established. Live LLM coverage used one configured model through the NVIDIA gateway; other provider routes were not qualified. The CAD checks exercised preparation plus TP4, not the entire CLI analysis workflow.

The batch cap is now 128, so a reported 145-batch workload still exceeds it. Large-skill runtime completion, remaining obfuscated-instruction limitations, evaluator binary-scope admission policy and deduplication limits, and general malformed-provider-response failures remain unresolved follow-up work. Independent content findings and missing-reference limitations remain visible; upgrading does not guarantee that a downstream admission gate passes.

See release PR #922 for validation details.