68 tests passing. Public evaluation logs and synthetic fixtures included.
I'd love feedback from developers using coding agents for security reviews.
https://t.co/5WB0pYaWyX
@OpenAIDevs@AnthropicAI
Your API checks login. Does it check who owns the invoice?
I built Verified Code Security Audit, an open-source skill for coding agents that guides reviews of tenant isolation, IDOR, backend permissions, secrets, and XSS
@OpenAIDevs@AnthropicAI
Each finding must include exact source lines, an exploit path, impact, and a testable fix.
The workflow records what was actually reviewed, then uses validated JSON to generate PDF reports and GitHub issue drafts in English or Portuguese.
@gitsecops@AnthropicAI@OpenAIDevs Pre-merge. recheck takes --rev, so PR CI runs it against the PR head and exit 1 blocks the merge — the stale finding never reaches main. Nightly is only the fallback for repos you don't gate. Same command either way; where you put it is the policy.
Coding agents invent security findings. So I made evidence mandatory.
Every finding needs path, exact lines, snippet and an exploit path, or vcsa validate rejects the report. Coverage is checked, not declared.
Agent Skill, MIT:
@AnthropicAI / @OpenAIDevs#AppSec#DevSecOps
@gitsecops@AnthropicAI@OpenAIDevs Fair point, so I shipped it: vcsa recheck compares every recorded snippet against a revision — intact, moved, stale, unverifiable. Exits 1 on stale, so a report expires in CI when the code moves under it.
Schema already required metadata.revision. Now something reads it.
Coverage is validated, not declared.
Claim exhaustive and it checks discovered == reviewed. Partial review must be recorded as sampled or limited.
No reviewing 4 of 30 endpoints and calling it clean.
https://t.co/5WB0pYaWyX
#AppSec#DevSecOps
@ml_ajuda 3 dias tentando pagar minha fatura do mercado credito! Porem o dinheiro esta na conta, ficou retido para ser cobrando e nao cobra. Já tentei contato com diversos atendentes, ngm resolve!