Skip to content
New

DeepSeek V4.1 Flash free for 7 days: 10M tokens a day after a top-up of $10 or more. Until Nov 30.

How it works
aeondave/cve-searchReporting and compliance

cve-search

⁨Build evidence-backed CVE inventories and assess whether reported vulnerabilities match an asset's product, package, build, and configuration. Use for vulnerability lookup, asset review, and remediation prioritization.⁩

v2026.11.13-21cd5683 installs+3 in 7 days39 StarsMITsynced 11 hours ago

Install

No NotHoly CLI yet? Install it first.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/cve-search

Writes to ⁦~/.notholy/skills⁩ after checking the signature and the SHA-256.

SignatureSigned, ed25519
cM70KLc4bo…P17nHcCw==The CLI checks the SHA-256 and the signature before it writes a file.
Packagefc9c3ce4…68a55039
fc9c3ce4a084…9ad668a550391 file · ⁦2.8 KB⁩
ScanMedium2 findings
  • injection.no_confirmationSKILL.md:17
  • Links to outside domains
See the full scan
Sourcegithub.com/AeonDave/malskill/tree/main/knowledge-offensive/cve-search@21cd568

Turn vulnerability records into an applicability decision with known coverage. A public record or matching banner does not establish that an asset is affected.

Identify and discover

  1. Capture the product/vendor, ecosystem/package, edition, version/build, platform and relevant configuration from available evidence. Keep asset identity and observation date. If identity is uncertain, search plausible candidates without silently treating the guess as confirmed inventory.
  2. Match the research window to the question. An installed-asset review can require older CVEs and newly updated advisories; a recent-publication filter alone misses them.
  3. Discover candidates through the relevant vendor/distro advisories, CVE/CNA records, and ecosystem sources such as OSV or GHSA. Use upstream issues, releases and patches to resolve gaps. Preserve publisher attribution when aggregators combine CNA, ADP or NVD information. Search summaries are leads; inspect the underlying records.
  4. Follow documented pagination and record filters, retrieval time, source failures and coverage limits. Deduplicate using verified aliases or explicit duplicate/replacement links. Keep relevant advisories without a CVE as separately identified leads; do not invent an ID or discard them.
  5. Stop when the requested lookup or bounded inventory has been covered, or state which sources/records remain unresolved. Zero hits, missing enrichment or an incomplete response do not establish a clean asset.

Use available source APIs/connectors for structured records and web search for discovery. Tavily can find advisories when available; follow the actual tool schema. Do not mistake ranking, snippets, or generated synthesis for verified applicability.

Check lifecycle and applicability

  • Record whether the CVE is reserved, published or rejected. Treat a disputed tag/statement separately; it is not a lifecycle state or a ruling on which party is correct. A rejected ID is not a separate active vulnerability; read the reason and any replacement. Reserved details remain unavailable unless supported by another identified source. See the CNA Rules.
  • Match exact products and required conditions. Compare versions using the ecosystem/vendor's ordering, not lexical strings or assumed SemVer. In CVE JSON, account for version type, inclusive/exclusive range bounds, status changes and default status. Missing status is not evidence of unaffectedness.
  • Check backports, package revisions and installed build evidence. An upstream version alone can misclassify a vendor-fixed package; require that vendor's advisory, changelog or patch/build record linking the installed revision to the fix. See Red Hat's backport explanation.
  • Preserve complete affected-configuration logic, including required platform/component combinations. NVD CPE data is a matching aid, not a local inventory measurement; unresolved criteria stay unresolved.
  • Check optional features, privileges, deployment mode and exposure separately. A WAF, firewall, failed probe or inaccessible route may block an observed path without removing the vulnerable component.
Applicability Evidence required
matched Product/package, affected version/build and required conditions align with reliable local evidence. This does not establish successful exploitation.
not matched Reliable evidence excludes the affected product/range or a required condition, or identifies the applicable fix in the installed build. State the reason and assessed conditions.
unresolved A consequential identity, version, configuration or source conflict remains. Name the next evidence check.

A disputed record is unresolved only when the disagreement affects this applicability decision and the evidence cannot settle it. Keep credible conflicting claims visible.

Prioritize without conflating signals

  • Preserve CVSS version, vector, score and source when available; a Base score describes severity, not the asset's complete risk (FIRST CVSS).
  • KEV inclusion establishes CISA's catalog status, not exploitation of this asset. Absence from KEV is not absence of exploitation (CISA KEV).
  • EPSS estimates the probability of observing exploitation activity in the wild in the next 30 days. Record score and date; its percentile is relative rank, not probability. Neither is an asset-specific observation (FIRST EPSS FAQ).
  • Combine supported applicability, exposure, business impact, controls and threat evidence for prioritization. If a signal was not retrieved, report unknown/not checked, not zero or not listed.

Deliver the inventory or lookup

For each relevant CVE or separately identified advisory, report:

  • identifier/aliases, title, lifecycle and dispute status;
  • asset/package/build, applicability, evidence locator and required conditions;
  • severity and threat signals actually checked, with source/date;
  • vendor fix or mitigation and its limits;
  • unresolved evidence and next check.

For an inventory, also summarize sources, filters, pagination/coverage, exclusions and failures once rather than repeating them in every entry. Cite each affected range, condition and remediation claim. Do not imply completeness beyond the sources and scope checked.

Load poc-weaponization when reviewing an existing PoC artifact. Load deep-research-offensive when the question needs competing explanations, chained prerequisites or detection/mitigation analysis beyond a record lookup.

Files in this version

Permissions

PermissionDeclaredFound by the scan
Reads filesNoNo
Writes filesNoNo
Runs shell commandsNoNo
Uses the networkYesYes
Uses connectorsNoNo
Drives a browserNoNo

Scan findings

Reviewed on Nov 13, 2026

  1. MediumPrompt injection · injection.no_confirmationSKILL.md:17
  2. InfoLink · Links to outside domains

A clean scan lowers the risk but does not prove a package is safe. Read what you install.

VersionPublishedCommitSHA-256SignaturePermission change
v2026.11.13-21cd568Nov 13, 202621cd568fc9c3ce4Signed–

Ask anything your scope allows.

Create an account, add credits and ask the first question in the chat or the terminal.

  1. 1/3

    Create the account

    Individual or company. You can switch to a company later without losing balance or keys.

    IndividualCompany
  2. 2/3

    Add credits

    By card. Pay per token, no subscription.

    $5.00minimum deposit
  3. 3/3

    Ask

    In the chat, or install the CLI and run notholy login in your terminal.

    $ notholy login