test(tools): comprehensive test suite for search_exploit_db, search_packetstorm, query_threat_intelligence_feeds, obtain_cves

[manus-use]

Summary

Adds tests/test_exploit_search_tools.py — a comprehensive, fully-mocked test suite covering four tools that previously had zero dedicated test coverage:

Tool	Tests added	Notes
search_exploit_db	21	TOOL_SPEC contract, input validation, HTML parse, error paths, toolUseId echo
search_packetstorm	22	TOOL_SPEC contract, input validation, HTML parse, error paths, result structure
query_threat_intelligence_feeds	19	TOOL_SPEC contract, input validation, CVE found/not-found, case-insensitive match, per-feed error degradation
obtain_cves	19	TOOL_SPEC contract, _get_all_cves_from_nvd (single/paginated/empty), _get_all_cves_from_github (single/pagination/error/no-cve-id), _filter_cves_by_epss (pass/block/percentile/empty/missing), tool entry point (merge, dedup, chunking, error handling), _enrich_with_cisa_kev
Cross-tool (vi_agent wiring)	6	Confirms all three vi_agent exploit-search tools referenced + importable
Parametrised TOOL_SPEC contract	20	4 tools × 5 invariants each

Total: 107 new tests (902 → 1019 passing, 0 failures, 3 pre-existing deselections unchanged).

Why these tools?

These four tools are wired into the core vi_agent pipeline (manus-agent analyze) and the vd_agent discover loop. Despite being actively used in production pipelines, they had no dedicated test file. Any regression in HTML parsing, EPSS filtering, or pagination logic would go completely undetected.

Test design

• 100% mocked HTTP — unittest.mock.patch on requests.get; no real network calls.
• Edge cases covered: empty query, whitespace-only, None, integer type coercion, missing keys, HTTP 4xx/5xx, ConnectionError, Timeout, generic Exception, empty response bodies, pagination boundary.
• Behavioural tests: CVE deduplication across NVD+GitHub sources, EPSS dual-threshold logic (absolute score > 0.05 OR percentile > 0.5), chunked EPSS calls (≤100 per request), CISA KEV enrichment correctness, toolUseId round-trip echoing.
• Living-doc tests: vi_agent wiring checks confirm tool module paths remain importable and referenced in the system prompt.

Existing open PRs checked for overlap (none)

#51, #53, #54, #58, #60, #64, #65, #67, #74, #75, #76, #77, #78, #79, #80, #82, #83, #85, #86, #87, #88, #89

No open or merged PR covers search_exploit_db, search_packetstorm, query_threat_intelligence_feeds, or obtain_cves test suites.