feat(scan): add opt-in transitive reference scanning

[rodboev]

Summary

Adds opt-in transitive scanning for source references inside scanned skill content, including normalized trust-prefix checks, command-level traversal budgets, shared dependency result reuse across sibling roots, cached child LLM telemetry, and source-aware reporting.

Closes #97

Root cause

The current scan pipeline resolves one input into one local directory, builds context from that tree, and writes the graph result from that single invocation. Recursive mode only repeats that one-hop scan for immediate local child skill directories, and report emitters assume every finding belongs to the directly requested source.

The original transitive branch added follow-up scanning on top of that model, but traversal accounting sat too far outside the expensive work. Root scans could resolve, load context, and start analyzers without the shared budget. Recursive roots shared cached child results but not counters, deadlines, or byte accounting. Cached child scans also dropped llm_call_log, and summary counts were computed before report-time baseline suppression.

Diff Notes

• Add a shared transitive traversal module that extracts source-like external references from file_cache, filters out adjacent non-source URLs, canonicalizes source identities with dot-segment and unreserved percent-decoding normalization, enforces allow or deny prefix checks on normalized path boundaries, and owns per-root visited-state mutation.
• Add --transitive, --transitive-depth, and repeated allow or deny prefix controls to skillspector scan, then route both single-skill and recursive multi-skill entrypoints through a cached traversal helper.
• Enforce command-level target, byte, and time budgets during root graph execution, child resolution, context loading, graph execution, and transitive LLM calls.
• Share one traversal budget across recursive multi-skill roots while keeping per-root visited sets, so sibling skills can reuse dependency results without each receiving a fresh allowance.
• Reuse InputHandler for approved external targets so existing host allowlists, SSRF checks, clone or download handling, and archive protections stay authoritative, including archive URLs on allowed Git hosts.
• Preserve cached child llm_call_log records and merge them into the final report, so degraded transitive semantic calls remain visible even when a dependency result is reused.
• Count transitive findings from active post-baseline report findings, so baseline-suppressed dependency findings do not inflate summaries.
• Add transitive_depth and source_url to Finding, preserve the root cleanup path through both the merged and zero-depth transitive paths, keep baseline fingerprints scoped by provenance, and merge component coverage with source-aware keys so same-named dependency files do not collapse or appear skipped.
• Restore the report tests that had fallen out of collection, then add focused coverage for normalized prefix checks, bounded traversal truncation, sibling cache reuse, source-aware coverage, transitive failure isolation, baseline-aware transitive counts, root traversal budgeting, and per-call LLM timeout propagation.
• Replay the branch onto current main.

Scope

This change stays on source types already supported by InputHandler, normalized trust-prefix checks, traversal depth plus hard fan-out limits, allow or deny prefix controls, cache reuse across recursive siblings, and provenance in reports. It does not add a web crawler, new allowed hosts, MCP behavior, or any default behavior change when --transitive is absent.

Verification

• pytest tests\unit\test_cli.py tests\unit\test_transitive.py - pass, 41 passed
• pytest tests\nodes\test_llm_analyzer_base.py tests\unit\test_llm_utils.py - pass, 142 passed
• pytest tests\nodes\test_report.py tests\nodes\test_sarif_rules_and_empty_findings.py - pass, 74 passed
• ruff check src tests
• ruff format --check src tests
• Invariant enumeration gate - pass, 37 rows, state domains checked
• Focused follow-up adversarial review on root traversal sharing and per-call LLM timeout propagation - no findings
• CI Lint & Test (Python 3.12), Lint & Test (Python 3.13), and DCO Check - pending maintainer approval and rerun after push

[rng1995]

Requesting changes. Transitive prefix controls can be escaped through non-normalized URL paths, traversal has no fan-out budget, and shared visited state causes sibling skills to omit shared dependency findings. The merged report also misreports dependency coverage, several report tests are no longer collected, and the branch conflicts with current main. Please address the inline blockers and rebase while preserving current CLI, input-handler, and reporting behavior.

[rodboev]

Addressed the blockers on the transitive traversal path rather than patching each edge case in isolation. The branch now normalizes target identities before allow or deny checks, applies an explicit traversal budget, reuses cached child results per referring root instead of suppressing sibling findings behind one shared visited set, keeps dependency coverage source-aware in reporting, restores the report tests that had fallen out of pytest collection, and replays the whole change onto current main.

[rng1995]

[Automated SkillSpector Review]

Re-review: still requesting changes. The prior path-normalization, per-root cache/visited, source-aware coverage, and test-collection issues are addressed. Remaining blockers are that byte/time limits are checked only after a child scan, recursive roots do not share a command-level budget, child LLM degradation is dropped, baseline-suppressed transitive findings are overcounted, and the branch still conflicts with current main.

[rodboev]

Rebased onto current main as cb5ffa1. The update preserves the review-thread fixes already described above and removes the remaining merge conflict.