§ Proof

Public records, methods, and controlled evaluations.

Twenty-one public vulnerability findings, four accepted Linux mainline fixes, and a proof library that keeps independently verifiable public research separate from generalized methodology writeups and project-reported benchmark results.

§ 01 — Public research

Twenty-one public findings. Fifteen full writeups.

Fourteen CVE records and twelve published repository advisories — five cases carry both, so the identifier totals overlap. Each links to the detailed SecHive writeup and the independent public advisory record. Six cases carry identifier and status only until their coordinators release the technical detail.

21
Distinct Public Findings
A paired CVE and GHSA counts once.
14
Publicly Disclosed CVEs
12
Published Repository GHSAs
Five also carry a CVE.
4
Accepted Linux Mainline Fixes
Two authored, two reported-by.

Public records rechecked 5 September 2026. Additional work remains in coordinated disclosure; identifiers and technical details are withheld until the respective advisories are released.

RECORD.0121 findings

CVEs & advisories

The complete identifier index — project, ecosystem, failed invariant, severity, and links to both the writeup and the source advisory.

RECORD.024 commits

Linux kernel

Accepted mainline contributions across net/sctp and fs/smb/server, with patch authorship and reporter credit recorded exactly as upstream states them.

RECORD.03Methodology

Proof library

Generalized research-method writeups and project-reported benchmark evaluations, kept clearly separate from the independently verifiable public record.

§ 02 — Highest impact

Six writeups worth starting with.

Sandbox escapes, a code-generation boundary, a filesystem canonicalization failure and a restricted-evaluator write path. Each page carries the failed invariant, the negative controls, the fix and the stated limits of the claim.

  1. 9.9vm2 NodeVM builtin allowlist escapeAn allowlisted node:test module exposed a method that launches a Node process with caller-controlled execArgv.GHSA-qhwx-74w5-xhxq
  2. 9.8Apache Fory registration bypassStrict class registration rejected an unregistered capture class and admitted the equivalent interface.CVE-2026-64606
  3. 9.8vm2 host prototype mutator escapeA fixed-depth wrapper guard recognized the shape of an invocation, not the capability behind it.CVE-2026-47698
  4. 9.1python-statemachine write-side traversalDunder reads were denied; the assignment path walked the same segment and mutated shared class state.GHSA-v3qq-3xvg-m77g
  5. 8.8isomorphic-git NTFS alias bypassPath validation reasoned about the input spelling rather than what the destination filesystem resolves.CVE-2026-77355
  6. 8.8DynamicExpresso late-bind reflection bypassReflection policy applied to statically typed nodes but not to the dynamic call sites that reached the same runtime capability.GHSA-5jx8-p6q2-455g
Where the rest are. The remaining findings — TypeBox, fast-xml-parser, fast-jwt, OpenEXR, the second python-statemachine advisory, and the WordPress CVEs — are listed with identifiers, severities and links on the CVEs & advisories index.
§ 03 — Linux kernel

Four accepted mainline fixes.

Public kernel history records the upstream changes, the credit trailers, and the stable-tree backports. Two patches authored; two accepted fixes reported.

  1. AUTHORsctp: validate Adaptation Indication parameter lengthAuthored and signed off by Charles Vosburgh; acked by Xin Long. Mainline Linux 7.2, backported to six stable releases.Commit
  2. AUTHORsctp: validate chunk length in the inqueue parserAuthored and signed off by Charles Vosburgh; acked by Xin Long. Applied to netdev/net.git on 2026-08-30.Commit
  3. REPORTksmbd: preserve VFS inherited POSIX ACL maskReported-by: Charles Vosburgh; patch authored by Namjae Jeon. First in Linux 7.2-rc5.Commit
  4. REPORTksmbd: enforce signing required by the sessionReported-by: Charles Vosburgh; patch authored by Namjae Jeon, tested and reviewed by ChenXiaoSong. First in Linux 7.2-rc5.Commit
  5. STABLESCTP stable-tree backportsSix verified backports, all of the first authored SCTP fix. The second authored fix and both reported-by fixes have none.6.18.y · 6.12.y · 6.6.y · 6.1.y · 5.15.y · 5.10.y
Coordinated disclosure. Additional vulnerabilities are currently being handled through coordinated disclosure with affected maintainers. Identifiers and technical details are withheld until their respective public advisories are released.
§ 04 — Proof library

Writeups and benchmarks.

Generalized methodology pieces and internal evaluations remain clearly separated from the independently verifiable public record above.

  1. CS.01Runtime authorization replayA signed one-time action that executes more than once. The proof shape is what makes it credible.CVSS High/Critical
  2. CS.02Validation boundary / denylist bypassA policy enforced on one route, skipped on another. Outbound checked, inbound released.Critical impact
  3. CS.03Cross-domain / cross-asset logic abuseAn authorization for object A, rebound to object B through hook or forwarding data.Critical impact
  4. CS.04XBOW-style benchmark campaign104 cases, 98 black-box wins, 104 white-box wins.benchmark
  5. CS.05Coordinated-disclosure methodologySix research-method families and three generalized writeups, with no claim that internal workflow items are confirmed vulnerabilities.methodology
  6. CS.06Juice Shop reference reportFull unredacted reproducible benchmark report.benchmark
  7. CS.07GOAD evaluation frameworkActive Directory lab methodology and scoring framework.framework
  8. CS.08OWASP WrongSecrets evaluation frameworkSecrets-management challenge taxonomy and scoring framework.framework
Benchmark boundary. XBOW-style, Juice Shop, GOAD, and WrongSecrets material is project-reported internal evaluation work. It is not independent certification or part of the public CVE total.
§ 05 — Trust and safety

Authorized research and human control.

  • I limit research to systems I own, locally reproduced open-source targets, or systems for which I have explicit testing authorization.
  • I handle findings through responsible coordinated disclosure.
  • I keep potentially harmful artifacts in segmented testing environments, including no-egress environments where appropriate.
  • I require human review before disclosure, external communication, patch submission, or any elevated action.
  • Advanced AI access is reserved for my internal, authorized research. It is not sold, shared, proxied, embedded for customers, or exposed through public SecHive services.
  • I do not provide customers or third parties with access to my AI accounts or credentials.
  • I use local-first handling, reproducible evidence, patch validation, snapshots, and rollback for sensitive research workflows.