FAI Hooks
Event-driven security gates on every prompt. Search the catalog, inspect evidence and token size, compare Full and Lean variants, then copy, download, or install from the same bounded record.
Discover
Semantic and voice search with domain filters
Inspect
Quality, tokens, WAF, events, and Lean evidence
Install
CLI and native VS Code deep links
Evidence catalog
Find the right hook for the job.
Filter by operating concern, search by intent, and inspect the source contract before using it.
50,527 Full tokens · switch to Lean for exact per-card and filtered-view receipts
11/11 gatedLean ModeModel-aware cost estimation with per-session tracking, daily/weekly/monthly aggregation, anomaly detection, CSV export, and budget alerting — FinOps visibility for AI-assisted development.
Detects threat signals in user prompts across 7 OWASP LLM Top 10 categories — prompt injection, data exfiltration, privilege escalation, credential exposure, system destruction, insecure output, and model theft.
Scans dependencies for license compliance across npm, pip, NuGet, and Go ecosystems with SPDX classification, configurable allowlist/blocklist, and severity levels.
Validates LLM-generated code and config files against 9 checks: naming conventions, frontmatter schemas, JSON structure, content safety, hallucination markers, PII detection, format compliance, length constraints, and markdown structure.
Scans uncommitted code for personally identifiable information across 12+ PII categories with configurable redaction strategies (mask/hash/remove), severity filtering, locale-aware patterns, and GDPR/HIPAA compliance markers.
Detects leaked credentials, API keys, tokens, connection strings, and private keys in code changes using 40+ provider-specific patterns and Shannon entropy analysis.
Structured JSON Lines audit trail of Copilot session activity with log rotation, compression, rate limiting, and error recovery — without ever logging prompt content.
Per-model token budgets with sliding window tracking, grace periods, configurable alert thresholds, daily resets, and multi-tenant support — preventing runaway LLM spend before it happens.
Intercepts tool calls with allowlist/blocklist enforcement, URL pattern filtering, per-tool rate limiting, an audit trail, and sandbox mode — 7 threat categories covering file destruction, force pushes, database drops, permission abuse, network exfiltration, and infrastructure teardown.
Lightweight PostToolUse hook that logs every tool invocation (name, redacted input, truncated output, duration, success) to a JSONL trace file for observability and audit. Never blocks — always returns {}.
Validates code changes against all 6 Well-Architected Framework pillars with per-pillar scoring and configurable thresholds.