A documentation refresh and quality-of-life patch. No code changes to detection logic — your scan results are identical to v1.2.0.

What's new

• 📒 CHANGELOG.md — proper release history covering v1.0.0 → v1.2.1, in Keep a Changelog format
• ⚡ Faster GitHub Action — inkog-io/inkog@v1 now downloads pre-built binaries instead of building from source. CI scans run ~10× faster (~60s → ~5s)
• 📝 Better contributor docs — CONTRIBUTING.md rewritten with a Universal IR primer and clearer guidance for adding framework support
• 🛡️ Stronger security policy — SECURITY.md now points to GitHub Security Advisories, lists supported versions, and covers the MCP server and GitHub Action
• 🤖 Dependabot enabled — weekly automated updates for Go modules and GitHub Actions
• 🐛 New issue template — dedicated framework-support template with checklists for detection categories and example agent code
• ✨ Refreshed bug report, feature request, and PR templates

Upgrade

Pin in your workflow:

- uses: inkog-io/inkog@v1   # always tracks latest 1.x
# or
- uses: inkog-io/inkog@v1.2.1

CLI:

brew upgrade inkog
# or
go install github.com/inkog-io/inkog/cmd/inkog@v1.2.1
# or
npx -y @inkog-io/cli scan .

Full Changelog

See CHANGELOG.md or the v1.2.0...v1.2.1 diff.

Highlights

Deep Scan — Run deeper behavioral analysis on your agent code with inkog -deep .. Deep scan layers AI-powered semantic analysis on top of the core static engine, catching subtle logic flaws that pattern matching alone misses.

Skill & MCP Scanning — Two new commands to audit the tools your agent uses:

• inkog skill-scan <path> — Audit SKILL.md packages for tool poisoning, command injection, and excessive permissions
• inkog mcp-scan <path> — Check MCP servers for data exfiltration, privilege escalation, and unsafe tool definitions

MLBOM Generation — Generate a Machine Learning Bill of Materials documenting your agent's components, tools, data flows, and dependency chain.

Inkog Red (Preview) — inkog red launches adversarial testing against running agents. Probes for prompt injection resilience, jailbreak resistance, and tool misuse under real-world conditions. Early access — join the waitlist.

New Features

• AI provider secret detection — Client-side detection now catches Anthropic, Google Gemini, Groq, and HuggingFace API keys alongside existing providers
• AG2 framework support — Full detection support for AG2 (formerly AutoGen) agent patterns, including multi-agent topology extraction
• NO_COLOR support — Respects the NO_COLOR environment variable for CI/CD and piped output
• Upload progress bar — Visual progress indicator during code upload to the analysis engine
• Multi-platform CI/CD templates — Ready-to-use pipeline configs for GitHub Actions, GitLab CI, Azure DevOps, and Jenkins (thanks @monssefbaakka)

Improvements

• Improved secret detection precision (v5–v12): multi-layer filtering for entropy, context, docstrings, variable names, PEM constants, and test fixtures
• Branded scan output as "Inkog Core" alongside "Inkog Deep" for clarity
• Better JSON output: empty arrays instead of null for findings
• Agent name derived from source path when not specified (-agent-name flag also available)
• .gitignore support and -max-files 500 limit for large repos
• Server-side policy support for governance filtering

Install / Upgrade

# Homebrew
brew upgrade inkog

# Go
go install github.com/inkog-io/inkog/cmd/inkog@v1.2.0

# npx (no install)
npx -y @inkog-io/cli scan .

# Binary
# Download from the assets below

New Contributors

• @monssefbaakka made their first contribution in #12

Full Changelog: v1.1.0...v1.2.0

Inkog v1.0.0

Static security scanner for AI agents.

Features

• Multi-framework support — LangChain, CrewAI, n8n, Flowise, and more
• OWASP LLM Top 10 — Prompt injection, insecure output handling
• EU AI Act compliance — Article 12, 14, 15 checks
• SARIF output — GitHub Security tab integration
• Diff mode — Only flag new findings in PRs

Quick Start

- uses: inkog-io/inkog@v1                                                                                                                                                                
  with:                                                                                                                                                                                  
    sarif-upload: true                                                                                                                                                                   
                                                                                                                                                                                         
Documentation                                                                                                                                                                            
                                                                                                                                                                                         
https://docs.inkog.io 

The first big iteration on detection precision. Volume of false positives reduced dramatically across multiple secret-detection patterns through 12 successive passes (V1 → V12), plus core scan ergonomics for large repositories.

Scan precision (huge FP reduction)

• 4-layer FP filtering pipeline for entropy-based secret detection (entropy, context, docstrings, variable names)
• Multiple precision passes targeting: PEM constants, Algolia search keys, hex hashes, SRI hashes, base64 placeholders, default DB passwords
• File-skip rules for cookbook/, samples/, tutorials/ directories
• Narrowed nonce_seed regex to remove noise

Repo handling

• .gitignore support — scanner now respects repo .gitignore rules during file walk
• -max-files flag (default 500) — caps file count on large repos with agent-relevance prioritization
• HTTP client timeout increased to match backend processing time
• Pre-upload size validation

Distribution

• Homebrew install — brew tap inkog-io/inkog && brew install inkog
• Pre-built binaries for darwin/linux/windows on each release with checksums

Server integration

• --policy flag now passes through to the backend for governance filtering
• Compliance count accuracy fixes
• Governance policy filter behavior corrected

Fixes

• URLs no longer flagged as hardcoded credentials
• Relative path handling fixes

Full Changelog: v1.0.0...v1.1.0

The initial public release of the Inkog CLI — a security pre-flight check for AI agents.

What's in 1.0

Static analysis for AI agent code mapped to EU AI Act, NIST AI RMF, and OWASP LLM Top 10. Catches the things code review misses because they look like normal code:

• Infinite / unbounded loops
• Token bombing / context window exhaustion
• Prompt injection (user input → LLM call without sanitization)
• Missing human oversight (Article 14 controls)
• Missing rate limits, missing authorization checks
• Hardcoded secrets (detected and redacted locally — never uploaded)

Frameworks supported at launch

LangChain · LangGraph · CrewAI · AutoGen · OpenAI Agents · Semantic Kernel · LlamaIndex · Haystack · DSPy · Phidata · Smolagents · PydanticAI · Google ADK · n8n · Flowise · Langflow · Dify · Microsoft Copilot Studio · Salesforce Agentforce

CLI features

• Five security policies: low-noise, balanced (default), comprehensive, governance, eu-ai-act
• Output formats: text, json, sarif, html
• Diff mode (-diff -baseline) — show only new findings since baseline, with regression detection for CI gates
• Cross-platform builds: darwin (amd64, arm64), linux (amd64, arm64), windows (amd64)

GitHub Action

inkog-io/inkog@v1 — composite action with SARIF upload to GitHub Security tab and PR comment integration.

Privacy model

The CLI is a dumb client by design. Secrets are detected and redacted locally before any code leaves your machine. All vulnerability detection runs server-side at api.inkog.io (you bring your own API key).

Install

go install github.com/inkog-io/inkog/cmd/inkog@v1.0.0
# or download a pre-built binary from the assets below

Get a free API key at app.inkog.io.