v1.8.0
Panda Recon 0.3.0 adds an on-demand threat intelligence layer. Security work can now use current, dated vulnerability data instead of model memory, and correlate it with a project's actual components.
- New
intelmode:latest,update,status,owasp,cve CVE-ID,kevandincidents. Sources are NVD, the CISA KEV mirror, FIRST EPSS, OSV, OWASP (Top 10, API Security, ASVS), MITRE ATT&CK and CISA advisories, from a fixed registry. scan,huntandincidentaccept--intel off|cache|live(default off). Scan writesPROJECT_EXPOSURE.mdonly for components that match by exact package and version. Hunt and incident keep external intelligence separate from local evidence and never raise an incident classification from it.- Freshness is visible. Each source has a refresh interval, a private dated cache and a status of Fresh, Stale or Unavailable. A failed or offline lookup is never shown as fresh, and the helper exits 3 when any enabled source is not fresh.
- Network use is opt-in. The helper contacts sources only with
--allow-network, sends package names and versions to OSV only with--allow-package-queries, refuses redirects and never sends source code. - KEV, EPSS and CVSS stay separate. KEV means exploited somewhere, not that this project was compromised. EPSS is a dated probability. Missing NVD enrichment does not mean unaffected.
intel latestreads one day of NVD changes by default and ranks results: KEV-listed first, then publisher severity, EPSS and recency. The report shows the top 50;intelligence.jsonkeeps all. Live on 2026-11-17 it retrieved 1,445 changed records in about 10 seconds, andupdaterefreshed every source.intel latest --severity Skeeps records at or above publisher CVSS severity S. Records without CVSS stay visible, because missing enrichment is not low severity, and project exposure is computed before filtering.- An optional scheduled refresh guide shows cron
examples for
intel update. Nothing is scheduled by default. - Panda Lead routes current vulnerability intelligence to Recon
intel. Dependency triage inside a requested code audit stays with Panda Security.
See the threat intelligence reference, command reference and CVE lookup example.
Verification
All 122 local unit tests pass, including 32 intelligence tests that use mocked source responses: aliases, withdrawn records, version ranges, cache expiry, offline reads, partial failures, rate limits, invalid responses and prompt injection inside advisories. The Suite validator, documentation build and internal link check pass. Five new reviewer scenarios cover KEV/EPSS interpretation, freshness without network, advisory prompt injection, project exposure and routing.
Live checks on 2026-11-17 retrieved every source successfully. NVD, KEV, EPSS and OSV: record. OWASP (Top 10 2025, API Security 2023, ASVS 5.0.0), MITRE ATT&CK (version 20.1) and CISA advisories: record. All 9 npm installer tests pass on Node 26.8.2, and the package preview includes panda-recon.
The five intel scenarios were run once per arm in fresh subagents (comparison). With the skill, all 14 checks passed. Without it, 9 passed, 3 were partial and 1 failed (the 14th, choosing a skill, does not apply without one). The no-skill answers treated record dates as proof of freshness and silently dropped an injected advisory description. These are single observations graded by the skill author.
Evidence limits
NVD change volume. About 1,400 NVD records change per day. latest and
update read one day by default, in pages of 1,000, and refuse a window over
8,000 records with NVD_WINDOW_TOO_LARGE instead of truncating it. A live
--days 7 window (9,081 changes on 2026-11-17) is refused this way. A missed-run
backlog can be restarted with intel update --source nvd --days 1, which says
what was skipped (live record).
The OWASP version check matches page text and may break if the wording changes;
a failed match leaves the version unknown (null) rather than guessing. The MITRE
source reads the version index, not the technique catalog, so revoked or
deprecated technique IDs must be checked on attack.mitre.org. One
trial per arm of the new scenarios is recorded; the helper itself was not run
by an agent. Automatic dependency inventory
reads npm package-lock.json v2/v3 only; Go and PyPI need a reviewed component
list. There is no separate GitHub Security Advisories source; GHSA records arrive
through OSV. No scheduler or continuous monitoring is included.
Some topics return less than the full proposal. intel owasp returns each
project's current version, not its categories or what changed. intel incidents
lists dated CISA advisories as leads; when an incident happened and which facts
are confirmed are left for the agent to establish from primary sources. All
topics share one report layout. Container images and runtime versions are not
inventoried.
Compatibility and recovery
The catalog remains at 28 skills. Existing Recon modes, helper commands and v1
records are unchanged; new schema fields are optional. The intelligence cache
lives under .panda-suite/recon/. There are no new required runtime dependencies.
Update or reinstall from tag v1.8.0. To recover a previous installation, reinstall from v1.7.2.