5 items, live from D1
Checklist
Every real, sourced verification/checklist fact, grouped by category unless filtered.
| item | source |
|---|
| Giving Claude a way to verify its work will markedly improve the quality of the final result; if Claude can close the feedback loop on its own, it will iterate until the output is right. |
Claude Cowork task list verification gate (web search, 2026-07-02) |
| Engineers give Claude abstract problems they're unfamiliar with, let it work autonomously, then review the roughly 80%-complete solution before taking over for final refinements. |
How Anthropic teams use Claude Code |
| Anthropic provides an AI-powered security review GitHub Action using Claude to analyze code changes for security vulnerabilities in context, producing detailed explanations, severity ratings, and remediation guidance. |
anthropics/claude-code-security-review (GitHub) |
| Anthropic's Claude Code security-guidance plugin runs a three-stage code review: a lightweight pattern check during file edits (no model call), a deeper review when Claude performs commits or pushes via the Bash tool that checks surrounding files/sanitizers/related code paths to reduce false positives, and a final catch for issues like injection flaws, unsafe deserialization, and insecure DOM APIs. |
Claude now reviews and fixes vulnerabilities as you write code (Help Net Security, 2026-05-27) |
| A verification-gate pattern should confirm a test/build/typecheck command exists for the detected project type and was actually run in the current session (an exit-code-0 log, not just "tests exist"), and should refuse to mark a task complete if changed files have no corresponding check run. |
agentic-dev-practices skill (this repo) |
Machine-readable version: GET /api/checklist-items?category=verification_gate