.svg)
ENTRY POINT
New listener
Review the external boundary and authentication requirement.
AUTOMATED THREAT MODELING · THREAT-INFORMED AI DEVELOPMENT
Give coding agents the security requirements for the files they change.
Map changed files to model elements and deliver their security requirements through MCP, the CLI and hooks. Detect selected architecture changes and let agents propose model updates for human review.
Add the webhook receiver in src/webhooks/receiver.ts.
Before edit · requirements
Authenticate requests before processing events.
After edit · design drift
New listener detected. Propose a model change for review.
Pull request
Model impact + requirement checklist + configured security checks
Give the agent the requirements for the file it is changing.
Heeler maps file paths to model elements. MCP, the CLI and pre-edit hooks deliver the relevant requirements before the agent edits; post-edit hooks check selected architecture changes.
Illustrative file mapping and hook result. Model and application permissions govern access.
CLI
heeler threat-model context --files src/events/retention.ts
Teach the agent your security workflow and standards.
Organization-specific skills tell the agent when to retrieve requirements, how to apply your security guidance and when to propose a model change. Developers can access the workflow through their coding agent, the CLI and VS Code.
Model and application permissions govern access. Installing a skill does not grant the agent access to additional systems.
Catch architectural changes while the agent works.
Post-edit hooks check the diff for selected signals and offer a model-change review when the addition is not represented.
INTEGRATION
New outbound host
Review the external system and the data crossing the boundary.
DATA AND CREDENTIALS
New store or credential read
Review the additional data handling and credential access.
Let the agent draft the model update. Keep approval with reviewers.
The agent can submit its diff and proposed components and flows through MCP. Reviewers inspect the change set, apply the changes they accept and approve the resulting version. Existing review locks and partial-apply rules govern agent proposals.
Illustrative webhook change
The agent adds a receiver and a Redis cache. It proposes the new components and flows; the reviewer checks their trust boundaries and requirements before applying them.
CONNECTED TO THE HEELER PLATFORM
The requirement reaches the agent. The platform checks the change.
Threat-model requirements enter through MCP and Agent Skills. Heeler’s broader controls assess agent activity, code changes and pull requests at their respective stages. Model-impact comments and requirement checklists give reviewers the design context.
Threat-model guidance complements the platform’s configured controls, from the agent session to the pull request.
Expose instructions that contradict the approved design.
Heeler cross-checks CLAUDE.md, .cursorrules and skills against model requirements. Conflicting instructions surface as findings alongside the requirements they undermine.
Illustrative workflow.
Explore Coding Agent SecurityTechnical questions
Does the agent approve its own model changes?
No. It submits a review proposal; people retain the model’s apply and approval decisions.
Does the post-edit hook replace scanning?
No. It checks specific architecture signals in a diff. Supported scans and code/deployment verification provide deeper evidence.
What happens if a hook cannot complete?
The hook uses the existing Block-fails-open behavior. An unavailable check does not establish that the change complies with the model.
Does an installed skill grant access to more models?
No. Model and application permissions still govern MCP and CLI access.
Next: verify what the agent implemented.
The webhook PR explains how it addresses authentication. Implementation verification checks the evidence at the analyzed revision and in the deployed environment.
AUTOMATED THREAT MODELING · FROM PROMPT TO RUNTIME
