What's new in Inkog v1.2.3: finding suppressions that actually work

Inkog v1.2.3 ships today. The headline is something that should have worked from the start: .inkogignore suppressions now apply to your scans.

Suppressions were silently dropped

If you set up an .inkogignore file to suppress a false positive, you may have noticed the finding kept coming back. That was not your configuration — it was a bug. The .inkogignore file was being read locally but was never included in the scan payload. The server never saw it, so server-side suppression rules never applied.

The same was true for inline # inkog:ignore comments when the server needed the file to evaluate them.

Starting with v1.2.3, .inkogignore is uploaded alongside your code on every scan. Server-side suppression rules apply. The workflow now works as documented.

# Suppress a known false positive
cat .inkogignore
# path/to/fixtures/mock_api_key.py
# src/tests/

inkog scan .
# ✓  Scanning your agent codebase...
# Suppressed 2 findings via .inkogignore
# 3 findings: 1 CRITICAL, 2 HIGH

If you were working around suppressions — deleting the rule, moving the file, filtering output in CI — you can simplify back to .inkogignore.

What else shipped

Agent Capability Surface. Every scan now prints a capability block showing the agents, tools, and control gaps Inkog found. The legacy Governance Status block is hidden when the surface is active. The dashboard surface view is linked directly from the scan output.

🔍 Agent Capability Surface
   Agents: 2 │ Tools: 11 │ Gaps: 4 (0 critical, 2 high)
   Governance Score: 87/100
   View full surface: https://app.inkog.io/dashboard/agents/...

The introduction post covers how the capability surface works in depth — the three-layer CAN/SHOULD/ENFORCED model, governance scoring, and compliance mapping.

OWASP GenAI LLM Top 10 2026. Finding references now use the current OWASP edition: LLM10:2026 and the OWASP Top 10 for Agentic Applications 2026 (ASI01–ASI10). If you are filing findings into a GRC tracker, the updated ids map to the current framework versions without a manual remap.

Six more AI provider secret patterns. The scanner now detects keys for Cohere, Mistral, Azure OpenAI, Vertex AI, and Stripe restricted keys (rk_live / rk_test scoped tokens, which have a different risk profile than sk_live). All patterns are applied client-side — credentials are redacted before any code leaves your machine.

Fixture notice. When Inkog classifies your uploaded files as test or example fixtures, the scan summary says so — rather than returning zero findings and leaving you to wonder why. If your agent code lives in a path that looks like a fixture directory, the notice tells you how to scan it directly.

Deep analysis warnings. Deep scan rules that could not be evaluated upstream are now reported as warnings instead of appearing in the strengths section. A Deep scan where every rule failed is reported as failed rather than clean.

-type flag. Override the Deep scan target: -type agent or -type copilot_studio. Useful when the scanner's inference gets the agent type wrong on a Copilot Studio project.

Why suppressions matter

Security scanning generates noise. The first scan of most codebases returns findings that are accurate but contextually unimportant — test fixtures with hardcoded keys, internal tools with intentionally broad tool access, known acceptable risk.

The way teams address this is suppression: mark the finding, explain why it is acceptable, move on. Suppression management is not security debt — it is security triage. It is how teams use a scanner productively after the first week.

When suppressions do not work, two things happen. Either teams stop suppressing and start treating the scanner as unreliable, or they suppress locally and forget that server-side results still show the finding — so CI gates and compliance dashboards look different from what the developer sees. Neither is a good outcome.

The MCP attack surface has made suppression hygiene more important, not less. As agents connect to more third-party MCP servers and tools, the volume of findings related to external tool trust grows. Teams need a way to mark which tool connections are intentional and reviewed. .inkogignore is one mechanism for that. It needs to work reliably. It does now.

Install

# Homebrew
brew upgrade inkog

# Install script
curl -fsSL https://inkog.io/install.sh | sh

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

Release binaries and checksums are on the v1.2.3 GitHub release page.

Inkog scans AI agent codebases for security vulnerabilities, governance gaps, and compliance issues. See pricing or book a demo.