Small, focused tools that catch AI-coding failure modes at the exact moment they'd otherwise slip through.

A secret firewall for AI coding prompts.

Hooks directly into Cursor's and Claude Code's prompt-submission events and scans every prompt locally before it leaves your machine. If it spots something that looks like a secret — an API key, token, or .env value — it blocks the prompt before it's ever sent.

• Runs 100% locally — no network calls, nothing stored.
• Detects AWS, OpenAI, Anthropic, GitHub, Stripe, Slack, npm tokens, JWTs, private keys, and high-entropy strings.
• Fails open by default; --strict mode fails closed for security-critical setups.

npx @khushwant.r/promptguard install

Catch hardcoded secrets before they reach a commit.

A VS Code / Cursor extension that watches for hardcoded secrets while you code and flags them on file open and on every save — long before they slip into git history.

• Flags known credential formats as errors (AWS, GitHub, GitLab, OpenAI, Anthropic, Stripe, Google, Slack, SendGrid, Twilio, npm, private keys, JWTs).
• Always shows secrets truncated (e.g. sk-a…2f) — never prints the full key back.
• Supports inline secretguard-ignore comments for false positives.

Stop hallucinated and typosquatted package imports.

AI assistants confidently invent packages that don't exist, or you fall for a typosquatted package planted by an attacker. ImportGuard checks every import against what's actually declared in package.json (or installed in your Python env) and flags anything that doesn't line up.

• Works across JS/TS/Angular/JSX/TSX and Python.
• Understands scoped packages, subpaths, and monorepos (walks up to the nearest package.json).
• Runs automatically on file open/save, plus on-demand workspace/file scans.

Real-time observability for LLM and agent calls.

A learning project (and likely future open-source release): a real-time event pipeline for observability into LLM/agent calls — latency, cost, and failure tracking — built on Kafka-compatible streaming and an OLAP query engine.

• Kafka-compatible streaming via Redpanda.
• Analytical queries with DuckDB, moving toward ClickHouse.
• Exploring partitioning, backpressure, and delivery guarantees.