Sightglass watches your AI coding agents (Claude Code, Codex, Cursor) and tells you exactly what they decided — every package installed, every tool invoked, every file touched. It classifies decisions as training recall vs. active research, flags supply chain risks (like known CVEs in auto-installed packages), and builds a community intelligence layer so teams can learn from each other's agent behavior. One CLI command to set up. Dashboard to explore. Open source CLI.

memories.sh — One layer for memories, skills, and rules across any agent

One layer for memories, skills, and rules across any agent

Hey Product Hunt! 👋 I built Sightglass because I realized I had no idea what my AI coding agents were actually doing. Claude Code was installing packages, making architecture decisions, choosing libraries — and I was just hitting "approve" without really thinking about it. When I finally audited one project, I found that 95% of package install decisions were pure training recall — the AI wasn't researching or comparing options, it was just going with whatever it remembered from training data. One of those packages had a known CVE. Sightglass is the tool I wish I had from day one. It passively watches your agent sessions, classifies every decision, and flags risks before they become problems. The CLI is open source, setup takes one command, and there's a dashboard at sightglass.dev to explore your data. Would love your feedback — especially from anyone else who's been vibe-coding and wondering what their agents are up to behind the scenes!

memories.sh — One layer for memories, skills, and rules across any agent

One layer for memories, skills, and rules across any agent

Sightglass

See what your AI coding agents actually decide

See what your AI coding agents actually decide