Changelog

This page records changes to PlugVet's rules, grading behaviour and directory coverage. An entry is added whenever a rule or the way grades are worked out changes.

Grades are opinions formed by automated static checks under our published rubric. They are not a guarantee about any tool.

22 Oct 2026

One entry regraded: 3X Ui Admin Skill moves from D to B

• The Claude Code skill claude-code-3x-ui-admin-skill moves from D to B. It is the only entry whose letter changed because of this.
• The finding that made it a D came from the lines at check-pins.sh, lines 73 to 75. They are search patterns for a pipe to a shell, each paired with an example that uses example.invalid, an address reserved so that it can never resolve. The script writes those examples into a temporary test file as text and searches that file. Nothing is downloaded or run.
• The grader had read those lines as a script fetched from the network and run. That reading was wrong for lines like these. The entry now carries an Informational Note instead: installer patterns found as detection data only, not executed.
• The grader now recognises a piped download that appears only as a search pattern or as an example aimed at a reserved test address. We checked every entry against that change and no other letter moved.
• The entry keeps its Low findings and its environment variable Note. It was regraded after a manual check of the source at the commit we graded.
• Grade counts after this change: A 1,855, B 5,872, C 2,227, D 785, F 0.

Directory now lists 10,739 graded entries

• New entries were added and graded under Rubric v2.
• Grade counts: A 1,855, B 5,871, C 2,227, D 786, F 0.

21 Oct 2026

Grade lookup launched at plugvet.com/lookup

• Look up how a skill, plugin or MCP server is graded before you install it.
• Live from late afternoon.

Directory now lists 9,739 graded entries

• Grade counts: A 1,713, B 5,363, C 1,964, D 699, F 0.

Directory now lists 8,739 graded entries

• Released early that day. The rule tune below went live in the same release.
• Grade counts: A 1,592, B 4,787, C 1,743, D 617, F 0.

Rule tune clears 74 of 80 known false alarms

• We tested a narrower version of several warning rules against stored evidence for 96 warned entries.
• 74 of the 80 entries we had judged to be false alarms no longer show those warnings.
• None of the 11 warnings we judged to be genuine were lost, and none of the 5 we were unsure about were lost.
• Six entries still show warnings, because the wording or code involved cannot be told apart from a real concern without wider context.
• One grade changed as a result: mcp-mcp-server-zotero-dev moves from D to C. Every other entry keeps its letter.
• Some rules were deliberately kept broad: quoted ssh, scp and git commands still count as credential use, and example folders still count for the encoded payload rule.

20 Oct 2026

Rubric v2: severity tiers

• Every finding now has a tier: High, Medium or Low. Some findings are Notes.
• Any High finding, or two or more Medium findings, gives D.
• One Medium finding gives C.
• Only Low findings, or a security Note, give B.
• An entry reaches A only when its only findings are housekeeping Notes, such as a missing licence file or no commit in over a year.
• Hidden or minified code counts as Medium. It counts as Low only in a known harmless file type, such as images, lock files, source maps or data files, or where the code does not run. Where we cannot tell, it stays Medium.
• Grades were recomputed from findings already stored for each entry. Nothing was rescanned.

Rules that are never downgraded

Four findings are always treated as High, whatever else we find:

• Code that decodes and runs an encoded payload.
• An installer that pipes a download from a raw IP address, paste site or tunnel host, outside docs and tests.
• An instruction in an agent instruction file to ignore earlier instructions or to hide a step, without defensive wording.
• A network host taken from input together with a credential path in operative code.

Reason codes on every finding

• Each finding is shown with a reason code and the quoted evidence behind it, so you can see why an entry got its grade.
• F grades are not published automatically. An F grade, if one is found, is reviewed by a PlugVet engineer against the source code and the evidence we quote before it is published. No F grade has been published yet.
• The free directory grades are automated. Before release, a PlugVet engineer reviews every paid report and any grade we hold back, including all F grades, against the source code and the evidence we quote. This is a hygiene check, not a security audit or a guarantee.

395 entries move from D to B

• Under the severity tiers, 395 entries that were D because of lower severity findings, such as a documented install script or an unnamed network host, are now B.