VibeCert
2.0/3
claude-skills, feat/add-vibecert · 11 commits · 6000 changed lines · 99% generated
| Metric | Score | Adjusted | Evidence |
|---|---|---|---|
| 1. Follow-up questions | 3/3 | 3/3 | Questions such as 'Which metric is check_info for?', 'What's this checkinfo change?' and 'Is 07a1c042 wrong too?' reference specific commits, identifiers and prior claims, showing the output was read and interrogated |
| 2. Coding guidelines | 2/3 | 1/3 | All four applicable guidelines were in context and compliance is broad, but the rule preferring direct name imports is violated throughout, with import pathlib and pathlib.Path at the call site where the same files correctly do from collections import Counter |
| 3. Output amendments | 1/3 | 1/3 | 293 of 5954 generated lines were rewritten after the author challenged them |
| 4. Prompt granularity | 2/3 | 1/3 | Median of 56 generated lines per producing prompt, inside the 50 to 200 band, with prompts naming discrete units of work such as 'Now implement point 2 of Milestone A' rather than whole outcomes |
| 5. Secondary tool usage | 3/3 | 3/3 | behavioural checks configured and ran |
| 6. Generative comments | 3/3 | 3/3 | 0 comment lines over 4078 lines of code, against a repository baseline of 0.0 |
Weight
Carried weight 3, being the highest of length 3, criticality 3 and risk 3.
Security findings
- S603: `subprocess` call: check for execution of untrusted input
- S607: Starting a process with a partial executable path