This report records a HostLatch 0.2.0 snapshot scan performed on 2026-09-30. It is evidence about scanner behavior on a convenience sample, not an independent security certification or a claim that HostLatch detects every malicious path.
| Measure | Result |
|---|---|
| Repositories | 12 |
| Files inspected | 2,359 |
| Findings | 77 |
| High / medium findings | 2 / 75 |
| Decisions | 2 block, 8 review, 2 allow |
| Manual rule-fit review | 77 of 77 matched documented rule intent |
| Truncated files | 9 ordinary files; no truncated control surfaces |
| Total wall time | 21.4 seconds |
| Median repository time | 1.02 seconds |
The 77 findings are not 77 confirmed vulnerabilities. Seventy-five identify files that a trusted consumer can load later and therefore deserve separate review. The two high-severity matches identify concrete activation patterns: a write-enabled CI deployment workflow and a package installation lifecycle command. Both appeared intentional, but both accurately cross the trust-handoff boundary HostLatch is designed to expose.
The corpus contains the 12 most recently active, non-fork repositories visible to the repository owner when the run began:
Private repository names, commit identifiers, file paths, contents, and individual findings are not published. The raw private manifests were held only in a temporary local validation directory and deleted after aggregation.
hostlatch scan <path> --snapshot --output <manifest> --fail-on never compared the current tree with Git’s empty tree.The per-file Git process fan-out discovered during the first run was fixed before the recorded run. Git index and tree modes are now fetched in batches.
| Repository | Pinned commit | Files | Findings | Decision |
|---|---|---|---|---|
| dev-agent-autopilot | cb1c6ba71132d82b5f53d1c81127b871f8f566f1 |
73 | 7 | review |
| iammurtaza53.github.io | fe42e265e675168413e014bb88dee2aee4c2a0bd |
31 | 3 | block |
| hostlatch | 486d51c0ca06d5e8f94366b8208ea3f6d89236e8 |
33 | 3 | review |
| dev-agent-autopilot-demo | 4e35de10fde1f99bf113f3d29904a28e0bc08194 |
10 | 4 | review |
| n8n-faq-chatbot-langchain | 91d4830fd1c9da89a5fce0f13b75cdee63f4a416 |
8 | 0 | allow |
Public subset total: 155 files, 17 findings, one block, three review, and one allow decision.
To reproduce a public result after checking out HostLatch v0.2.0:
git clone https://github.com/iammurtaza53/dev-agent-autopilot.git corpus/dev-agent-autopilot
git -C corpus/dev-agent-autopilot checkout cb1c6ba71132d82b5f53d1c81127b871f8f566f1
node ./bin/hostlatch.js scan corpus/dev-agent-autopilot --snapshot --fail-on never
The seven anonymized private repositories contributed 2,204 files and 60 findings: one high and 59 medium. Their decisions were one block, five review, and one allow.
| Category | Findings |
|---|---|
| Package control plane | 36 |
| Agent control plane | 23 |
| CI control plane | 7 |
| Environment control plane | 6 |
| Git control plane | 4 |
| IDE control plane | 1 |
Manual rule-fit review found no mismatch between a reported finding and its documented rule criteria. This narrow observation must not be presented as 100% vulnerability-detection precision: the review measured whether HostLatch described the intended control surface, not whether the repository was exploitable.
The total recorded wall time was 21.373 seconds. Median repository time was 1.024 seconds; the slowest repository took 7.988 seconds.
allow decision means no documented HostLatch rule matched. It does not certify a repository as safe.Machine-readable aggregate results are available in validation/2026-09-30.json.