APPSEC TERMS
Agentic AppSec terms explained
Short answer
A
Agent activity log
A record of what a coding agent did: commands run, files touched, packages installed, tools called. OX Security lists an agent activity log among VibeSec's features.
Learn more in the AppSec course
Source: OX Security docs index · Read 27 Sep 2026
Agent inventory
A list of the coding agents, models, MCP servers and skills in use across an organization. It is the first step of agent governance.
Agent policy
Rules about what coding agents may run, install, call and send, usually written as allow, steer or block.
Agentic AppSec
Application security for a software process in which AI agents take part: governing coding agents, and using AI agents to triage, prioritize and fix findings.
Agentic remediation
An AI agent writes the fix for a security finding, ideally validates it, and lands it for review, usually as a pull request.
Agentic triage
Automated or AI-assisted ranking of findings by reachability, exploitability, runtime context or business context.
AI-BOM
An AI bill of materials: an inventory of the models, AI services, agent rule files, MCP servers and AI API keys a software project uses. Cycode, Apiiro and Legit Security describe AI-BOM or AI inventory features.
Source: Cycode: AI security · Read 27 Sep 2026
Anti-tampering
Protection that stops a user or process from switching off or bypassing a security control. Legit Security lists anti-tampering among VibeGuard 2.0's features.
Source: Legit Security: VibeGuard 2.0 press release · Read 27 Sep 2026
ASPM
Application security posture management: a platform that collects findings from many scanners, de-duplicates and prioritizes them, and tracks posture across the SDLC.
AutoFix
A feature that generates a code change for a finding. Aikido Security and Apiiro use the name AutoFix; GitHub calls its feature Copilot Autofix.
C
Coding agent
An AI tool that writes and changes code and can take actions such as running commands, installing packages and calling tools. Examples include Claude Code, Cursor, GitHub Copilot, Windsurf, OpenAI Codex and Kiro.
Control point
A place in the software process where a security control can act. This site uses six: agent, IDE, pull request, pipeline, triage and fix.
Cyber Resilience Act (CRA)
EU regulation on products with digital elements. From 11 September 2026, manufacturers must report actively exploited vulnerabilities and severe incidents through the CRA Single Reporting Platform, with an early warning within 24 hours and a notification within 72 hours.
Source: European Commission: CRA reporting obligations · Read 27 Sep 2026
D
Dependency cool-down
A waiting period before a newly published package version may be installed. OX Security's Dependency Vetting can apply cool-down periods to new packages that AI agents try to install.
Source: OX Security blog: VibeSec Dependency Vetting · Read 27 Sep 2026
E
Endpoint agent
Security software that runs on the developer's machine rather than inside one editor, so it can see several coding agents at once. Legit Security built VibeGuard 2.0 as a desktop app rather than a plugin.
Source: Legit Security blog: what we learned building VibeGuard 2.0 · Read 11 Oct 2026
Excessive agency
An AI agent with more functionality, permissions or autonomy than its task needs. OWASP's Top 10 for LLM Applications lists excessive agency as LLM06.
Exploitability
Whether a vulnerability can actually be used by an attacker in a given application. Cycode's AI Exploitability Agent checks whether a CVE is reachable.
Source: Cycode homepage · Read 27 Sep 2026
F
Fix acceptance rate
The share of agent-written fix pull requests that reviewers merge without rewriting. A practical measure of a fix agent's usefulness.
H
Hook
A point inside a coding agent where custom code runs before or after an action. Claude Code documents a PreToolUse hook that can block a tool call before it runs.
Source: Claude Code docs: hooks · Read 27 Sep 2026
I
Install scripts
Scripts a package runs during installation, such as preinstall and postinstall in npm. npm v12 will not run dependency install scripts unless they are approved.
Source: GitHub Changelog: upcoming breaking changes for npm v12 (9 Jun 2026) · Read 27 Sep 2026
Instruction file
A file in a repository that gives a coding agent standing instructions, such as .github/copilot-instructions.md for GitHub Copilot or project rules in .cursor/rules for Cursor. Instructions steer an agent; they do not enforce.
Source: GitHub Docs: Copilot repository instructions · Read 27 Sep 2026
M
Malicious package
A package published to a registry with harmful code. Several platforms here block known-malicious packages at install time.
Managed settings
Configuration an organization enforces on every installation of a tool. Claude Code supports managed policy settings, including an option to allow only managed hooks.
Source: Claude Code docs: hooks · Read 27 Sep 2026
MCP (Model Context Protocol)
An open protocol that lets AI applications call external tools and data sources through MCP servers. In the 2026-07-28 revision each request declares its protocol version.
Source: Model Context Protocol: versioning · Read 27 Sep 2026
MCP proxy
A component that sits between an agent and MCP servers to inspect or control the calls. Backslash Security lists an MCP proxy for dynamic protection.
Source: AWS Marketplace: Backslash Vibe Coding Security Platform · Read 11 Oct 2026
MCP server
A service that exposes tools or data to an AI agent over MCP. Several AppSec vendors publish MCP servers so agents can query security findings.
P
Prompt injection
Input crafted to make an AI model ignore its instructions or take unintended actions. OWASP lists prompt injection as LLM01:2025.
Source: OWASP: LLM01:2025 Prompt Injection · Read 27 Sep 2026
Push protection
A control that blocks a push to a repository when it contains a detected secret. Part of GitHub Secret Protection.
Source: GitHub Docs: about GitHub Advanced Security · Read 27 Sep 2026
R
Reachability
Whether vulnerable code is actually called by the application. Reachability analysis lowers the number of findings that need action.
S
SAST
Static application security testing: analysis of source code for vulnerabilities without running it.
SBOM
Software bill of materials. CISA describes an SBOM as a nested inventory, a list of ingredients that make up software components.
Source: CISA: Software Bill of Materials · Read 27 Sep 2026
SCA
Software composition analysis: finding known vulnerabilities and license issues in open source dependencies.
Scanner orchestration
Running or ingesting several scanners and merging their findings into one de-duplicated queue.
SSDF
Secure Software Development Framework, NIST SP 800-218: practices in four groups, Prepare the Organization, Protect the Software, Produce Well-Secured Software and Respond to Vulnerabilities.
Source: NIST CSRC: Secure Software Development Framework · Read 27 Sep 2026
V
Vibe coding
Building software mostly by prompting an AI agent and accepting its output with little line-by-line review. It raises the importance of controls on the agent and checks at the pull request.