Back to blog

Semgrep in the Coding Agent: Source Detection and Runtime Proof

3 min read
Semgrep in the Coding Agent: Source Detection and Runtime Proof

Sources checked: . FailSafe Research.

Application security is moving from the pull-request gate into the coding-agent session. Semgrep launched an Agentic AppSec Platform on 12 November 2026. It includes controls applied as agents generate code, agentic vulnerability-discovery and remediation workflows, organizational context and policy inside the agent loop, an MCP server usable from Codex, Claude Code, Cursor and other clients, plus automated triage, fix generation and ticket creation. Semgrep says its Guardian controls have been deployed to more than 10,000 developers.

Those adoption and effectiveness figures are vendor-reported. The architectural shift is the part worth planning for. Defects can be prevented or fixed before they enter a repository, which reduces some classes of work for runtime testing. It also increases reliance on agent-facing policy, MCP connectivity and automated remediation.

Source-time prevention and runtime proof answer different questions

A source rule can show that a dangerous pattern exists in code, or that a generated change avoids it. Runtime testing shows what the deployed system actually does with a real identity, configuration and data path. Neither replaces the other. A missing source pattern does not prove exploitability, and a reproduced exploit does not always point to the exact source change that removes it.

What each layer can establish
LayerStrengthLimit
Source rule in the agent loopPrevents or flags a known pattern while code is written.Only covers patterns someone encoded; configuration and deployed behavior can differ.
Static analysis in CIConsistent review across a repository and history.Cannot show whether a reachable path produces an effect.
Runtime attack-path testReproduces a consequence through the deployed identity and data path.Requires access and time; covers the tested configuration only.
Retest after fixConfirms the deployed change denies the original case.Does not cover unrelated paths or later changes.

A workflow that keeps both claims honest

The useful division of labor is sequential. An authorized runtime test reproduces a path and records the source-aware details an engineer needs. A source tool such as Semgrep proposes a fix in the repository under normal review. The fix is merged. The runtime test repeats the original case on the deployed build. Each step produces its own artifact, and the finding reference connects them.

That sequence avoids two common failures. A source finding marked "fixed" without a runtime retest can leave the path reachable through another entry point. A runtime finding handed to engineering without source context can produce a patch that closes one route and leaves the underlying pattern. Keeping the claims separate makes it clear what has been shown.

Treat Semgrep as a partner, not only a competitor

Semgrep operates at the prevention and remediation layer. FailSafe tests the running system across applications, APIs, cloud, identity and AI tools. A combined workflow is more useful to a customer than a scorecard that pretends one layer replaces the other. The black-box to source-aware remediation guide covers the handoff, and the co-development practice covers embedding fixes with engineering teams.

Buyers should still validate vendor-reported figures independently. Ask for the policy set, the agent clients supported, how conflicts between generated and reviewed policy are resolved, and whether a runtime retest is part of the remediation claim. A tool that reports "three times more likely to be fixed" is reporting its own measurement until an independent party repeats it.

Questions & answers

Questions about this topic

Direct answers to common questions about this page.

No. Source rules find patterns. A runtime test shows what the deployed system does with a real identity, configuration and data path.

Keep one finding reference from reproduced impact through the proposed fix, the merged version and a runtime retest on the deployed build.

The deployment figures are vendor-reported. Confirm the scope and method before treating them as a market measurement.

It occupies the prevention and remediation layer. FailSafe tests runtime behavior. The two can be connected in one remediation workflow.

Put this guidance to work

Review how reproduced findings become reviewed code changes and a retested closure.

Startup offer: $999 USD for 250 credits and one authorized black-box run. Sign in to SWARM before Stripe checkout. Custom scope and sovereign deployments are separate.

Connect testing to remediation