What problem does it solve? Running an AI-driven security scanner like Vercel's deepsec involves expensive LLM calls, complex credential setup, and multi-stage workflows (scan, process, triage, revalidate, export). This Skill orchestrates the entire deepsec lifecycle cost-consciously, preventing runaway spend and misconfigured CI security gates. ## Core Features & Use Cases - Workspace Bootstrap: Installs the .deepsec/ workspace, configures AI credentials (AI Gateway, OIDC, direct provider keys), and writes a project-specific INFO.md that improves finding precision. - Cost-Aware Scanning: Enforces calibration passes (--limit 50 --concurrency 5) with cost extrapolation before any full process run, and resumes safely after quota or network interruptions. - PR Security Gating: Scaffolds a two-job CI workflow using process --diff with --comment-out, keeping PR-controlled code in a no-write job. - Custom Matchers & Triage: Authors project-specific matchers in .deepsec/matchers/, runs triage and revalidate to cut false positives, and exports findings to markdown or JSON. - Use Case: A team wants to add an AI security review to every pull request. This Skill sets up deepsec, calibrates cost on a sample of files, wires the two-job GitHub Actions workflow, and surfaces only revalidated HIGH+ findings as PR comments. ## Quick Start Ask the agent to set up deepsec in this repository, run a calibration scan on 50 files, and report the estimated cost of a full scan before proceeding.
