LESSON 2 OF 10 · AGENTS IN THE SDLC · 3 min read
Where coding agents touch the SDLC: the six control points
Short answer
By The Pipeline Desk · Best Agentic AppSec · Published 2026-09-28 · Updated 2026-10-11
Why map control points at all?
When a person writes code, most AppSec controls wait for the commit. When an agent writes code, a lot happens before the commit: the agent reads files, installs packages, runs commands and calls tools. Mapping the points where a control can act shows which risks your current tools cannot see.
The six points
- Agent. Controls that act on the coding agent itself: which agents and MCP servers are allowed, what commands they may run, what they may install. Legit Security's VibeGuard 2.0 works here, on the developer endpoint. Semgrep Guardian works inside individual agents through hooks that fire on every file write.
- IDE. Checks in the editor or before commit. Cycode's AI Guardrails, launched on 17 February 2026, intercept secrets in prompts, file reads and MCP tool calls before they leave the IDE.
- Pull request. Scanning and review of the change itself. This is where most SAST, SCA and secrets scanning already runs, for agent-written and human-written code alike.
- Pipeline. Policy across CI/CD, source control and the software supply chain. GitHub, for example, added a ruleset rule in public preview on 9 September 2026 that blocks a pull request from merging while secret scanning alerts remain open.
- Triage. Deciding which findings matter. Endor Labs, for instance, builds triage on reachability across dependencies, containers and first-party code.
- Fix. An agent or AI feature that writes the change. Aikido Security's AutoFix opens pull requests, and its free plan includes 10 AI AutoFixes a month.
What does each point catch?
- The agent point catches actions: a risky command, an unapproved MCP server, a package install, a secret pasted into a prompt.
- The IDE point catches problems while the developer is still watching.
- The pull request point catches what reached the repository, regardless of who wrote it.
- The pipeline point catches posture: unprotected branches, unreviewed workflow changes, secrets outside code.
- Triage decides the order of work. Fix decides how much of that work a person has to do.
How to use the map
Draw the six points for your own organization and mark the tools you already run at each one. Most teams find the pull request point well covered and the agent point empty. On this site every platform profile has a pipeline strip with the same six points, built from the vendor's own pages. A filled node means the vendor describes a control there; it does not measure how well the control works.
The next lesson looks at the first three points in detail and how to choose where to enforce agent policy.
Sources
Source: Legit Security: VibeGuard 2.0 press release · Read 27 Sep 2026
Source: Cycode blog: AI Guardrails in the IDE · Read 27 Sep 2026
Source: Semgrep Guardian · Read 27 Sep 2026
Source: GitHub Changelog: block pull requests with exposed secrets (9 Sep 2026) · Read 27 Sep 2026
Source: Endor Labs homepage · Read 27 Sep 2026
Source: Aikido Security pricing · Read 11 Oct 2026
Read next
Lesson 3: Choosing the enforcement point
Agent policy can be enforced on the developer endpoint, in the IDE, through hooks inside one agent or at the pull request. What each point covers and misses.
Pipeline strips
Which control points each vendor documents.
How the pipeline strip works
The six control points and what a filled node means.