openai/codex-security/threat-model-visualize
Claude Code and agent skill by openai/codex-security
- vendor
Checking that the SKILL.md file is still in the repository.
Turn an existing threat model into an interactive architecture and threat-path explorer. Resolve simple requests for a target's last scan; do not perform a new assessment.
Install
npx vskill@latest install openai/codex-security/threat-model-visualizeSKILL.md
The same SKILL.md was published earlier as openai/codex-security/threat-model-visualize.
Threat Model Visualize
Create a standalone explorer that shows application teams and leadership what can go wrong, how it could happen, why it matters, and where controls would stop or limit it. Use the shared workbench and role palette; do not use Mermaid or run a new scan.
A request such as “Visualize the threat model generated in the last scan of X” is sufficient. Resolve the target, latest assessment and snapshot. If the latest artifact is missing, disclose that before using an older fallback. The default input is the existing threat-model artifact, not an automatic import of a complete scan. Keep inputs immutable and do not silently combine snapshots. Do not inspect application code or invent topology, authority, controls, reachability or validation.
Capture source-supported deployment context: exposure, trust, tenant and process boundaries, identities, and conditional deployment assumptions. Do not treat repository defaults as proof of deployed configuration. Preserve uncertainty and the distinction between hypotheses, findings, declared controls and observed enforcement. For several explicitly supplied artifacts, pass one derived document to --assessment. Label verbatim excerpts with each source’s identity, snapshot and original finding status. Distinguish findings from hypotheses, identify local IDs, and preserve conflicting accounts and the original artifacts.
For a fresh candidate or regeneration, author a new view model; do not patch or copy a previous candidate model or HTML. Ordinary edits may reuse the existing model. Renderer-only polishing is a display update, not fresh generation.
1. Enumerate and validate the source
Complete this phase before planning any visualization, choosing display groupings or routes, assigning geometry, or authoring the explorer JSON. Save source-inventory.md in the workspace. Validate fidelity to the supplied assessment, not the application implementation; do not run a new assessment to fill gaps.
Record the assessment identity, exact file hash and scope. Give inventory rows semantic IDs and supporting source spans or exact quotations, without assigning visual node or edge IDs. Enumerate:
- Components and roles: actors, inputs, application components, stores, external dependencies, trust boundaries, platform-specific instances and administrative contexts. Distinguish trusted operators, assumed compromise, controlled input, owned assets and enforcing components.
- Relationships and capabilities: each supported caller → operation → receiver relationship, meaningful return, independently sufficient authority or entry condition, prerequisites and platform/deployment conditions. Enumerate alternatives per operation: an account session and an issued capability may each authorize the same operation. Preserve known logical capabilities when transport or order is unknown; do not import a prerequisite from another entry point.
- Controls and qualifications: each protective mechanism → enforcing owner → constrained component → protected operation, with conditions and evidence. Record an unmapped enforcer explicitly; configuring or being subject to a restriction does not establish enforcement ownership. Classify existing mechanisms, proposals, security objectives, configuration behavior and caveats separately. Ownership, expanded permissions, disabled protection or documentation advice alone is not an existing safeguard.
- Threat-model coverage: substantive threats, supplied findings, hypotheses and consequential configurations, linked to the relevant components, relationships and controls. Retain source confidence, potential impact and unresolved questions.
Review the inventory against the complete assessment before proceeding. Check that evidence supports the same object, caller, operation and owner; distinguish configuration delivery from protected operations and assets from their managers. Do not transfer a validation rule between different inputs or infer platform endpoints from a group label. Account for every substantive source concern; reconcile aliases and record contradictions without inventing a resolution. Correct unsupported assertions, misclassified controls and unexplained omissions now.
Finish the inventory with source-reconciliation corrections and unresolved source gaps. Keep later model-repair notes and review verdicts in correctness-review.md, not this source inventory. It is ready for composition when its claims are supported or explicitly unknown and coverage has been reconciled. Source uncertainty may remain; it must not silently become a visual relationship or safeguard. Do not plan the visualization while performing this review.
2. Compose from the validated inventory
Read the view-model contract and presentation guidance, then build a source-backed JSON model and geometry. The shell supplies interaction, colors and exports.
Default to standard: compose a readable architecture overview before adding threats. Show the actors and entry points, trust boundaries, application or library components, identity and authorization, stores, and external dependencies needed to explain the assessment. Include administrative, automation and build contexts when relevant. Connect them with meaningful request, data and authority relationships. Match platform relationships to the service endpoints named by the source; do not infer group membership or a UI handoff. Use grouping, clear boundaries, aligned flows and short labels while preserving distinctions that change a threat or control. Do not collapse the whole system into one threat's pipeline or move substantive sources and control owners off-map merely for sparsity.
Keep the primary architecture visible in muted colors when a threat is selected. Emphasize its complete source-supported route, including proxies, other transit components, handoffs and meaningful returns. Highlight active transit with fill while retaining its neutral outline; transit alone is not a weakness. Distinguish conditional transitions from ordinary flows. A partial or unmapped view needs an evidence-based scope explanation. Secondary connections may appear under More connections, but the basic architecture must stand on its own.
Use the validated inventory as the coverage ledger. Map each substantive operation and its supported authority alternatives to selectable relationships and controls, or an explicit source-supported scope exclusion. Showing an actor in another threat does not cover its access to this operation. Shared nodes and edges must be valid for every alternative that uses them; an unordered union may retain alternative-specific relationships when their membership is explicit. Prose such as “otherwise authorized” cannot broaden an account-only path. An embedded quotation is not selectable coverage. Label a user-requested subset incomplete. Optional authoring fidelity is simple for a more grouped summary, standard by default, or detailed for useful secondary connections and facets; it does not change claims or require a UI selector.
Bind each case's mechanism, priority, prerequisites, impact and evidence to its own substantive source section. Verify that each quotation matches and supports the same object, operation and enforcing owner as the claim. Do not turn ambiguous source language into a more specific existing safeguard. Do not invent a sequence, authority gain or successful transition. Where order is unsupported, show related components with an explicit scope note. Model the actual starting actor or input and least required authority. Trusted operator or administrator activity remains authorized context, even when it changes configuration or consumes resources; make it a threat source only when the assessment explicitly models its abuse or compromise. An added assumption of malicious intent is not source evidence. Classify a caller-controlled value as untrusted input only when the source supports that trust boundary. Keep alternative footholds distinct and scope each selection to its relevant relationships and controls, with other architecture muted.
Expose consequential supplied controls and proposals at their actual operations and enforcing owners. Split controls with different owners, operations or effects, including deployment reachability and protocol authentication. If an enforcing owner is unmapped, disclose the gap rather than assigning the control to its recipient. Label proposals Proposed control and their effects conditionally as “Would block” or “Would limit.” Show where and why a control acts, its conditions and residual effect; mark the effect unknown when it is not established. Preserve arrival at the decision owner where required; blocking a sensitive result does not necessarily stop ordinary traffic. Do not present a proposal as deployed or validated.
3. Adversarial correctness review
Finish this review before starting a preview server, opening the explorer in a browser or testing exports. Static build and structural validation may run while review is pending; browser work may not. Try to disprove the saved JSON's claims, not confirm the author's intent. Use a fresh-context reviewer when delegation is available: explicitly disable history inheritance (fork_turns: "none" for spawn_agent) and send the review task and artifact paths, without the author's conclusions or screenshots. Omitting the history option can inherit the author's work. Wait for that review and its fix verification. Use a separate self-review only if delegation is unavailable or an attempted delegation fails; record the concrete reason and independence limit. Pending review is not grounds for that fallback. Do not inspect application code or perform a new assessment.
Have the reviewer read the complete assessment and view-model contract first, retrieving any portions omitted by truncated tool output, and save a compact set of source-derived obligations in correctness-review.md before inspecting the inventory or JSON. Include explicitly trusted roles, authorized operations and scope exclusions alongside substantive concerns, least starting authority, supported operation and result/recipient/asset consequence, deployment conditions, and controls with their enforcing owners. Then compare the inventory and actual JSON with those obligations. Neither a validated inventory nor the author's coverage ledger is proof that the JSON preserves the source.
Resolve each distinct source/route selection's effective highlighted relationships and applicable controls, including inherited defaults, overrides, grouped members, branches and control effects. Record the source span, starting authority, effective node/edge IDs, operation and consequence, control owners, and any discrepancy in the review. Equivalent selections may share a row if their equivalence and differences are explicit. Check the resolved IDs against their definitions, not only scenario prose: an override can remove a consequence, a source can impose an unnecessary prerequisite, and an owner marker can contradict an unmapped-owner note. Challenge these concrete failure modes:
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files. Certified 14 Nov 2026.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingplugins/codex-security/skills/threat-model-visualize/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/openai/codex-security/threat-model-visualize)