Strategy Visualization Workbench

The security policy says there is no install-time or background behavior, core rendering makes no network requests and needs no credentials, and writes go only to user-selected output paths. It also constrains interactive source links at build time, supporting high scores for least privilege, data-flow transparency, and external-effects controls. Users are warned not to share sensitive material with agent environments they do not control, but there is no detailed classification or handling workflow, so sensitive-data handling is partial. Users control writes through explicit destinations, and scaffolding avoids overwriting non-empty directories unless forced; however, there is no general confirmation flow or explicit rollback plan. Dependency notes distinguish the standard-library core from optional browser and PDF tools, but provide no dependency vulnerability review. Source fields and design records help attribution, though a complete provenance chain for every output is not shown.

The README, CI configuration, and test files describe validation, including interactive cases and hand-calculated expected values. Static materials cannot establish that these checks passed at the target revision, and the coexistence of a stable release and a preview release adds version interpretation overhead. The core uses the standard library, while browser and PDF workflows require additional Playwright, browser, and Poppler dependencies, so availability is only partly addressed. Validation and test commands are documented, but evidence for user-facing failure diagnosis and recovery is limited.

The documentation covers board updates, strategy, project status, market entry, sales proposals, and Japanese business documents, and distinguishes static, animated, interactive, and report outputs. Boundaries are explicit: 22 patterns render directly while others produce specifications and image prompts; the English example is not a general business model, and unverified platforms are listed. Trigger guidance distinguishes ordinary animation from explicitly requested dynamic interaction. Installation examples focus on terminals and particular agent environments, so the claimed support for many agents is not fully demonstrated.

The README is organized into quick start, workflow, gallery, templates, reports, speaker scripts, and validation, with extensive examples and operating instructions. Installation steps are concrete but span several paths, and readers must interpret the stable-versus-preview release relationship. Limitations and unverified environments are documented in detail, and the MIT license text is complete. A changelog and release information are present, but maintenance responsibility is not clearly assigned beyond issue and security-report channels. Unverified publisher identity is not treated as a deficiency.

The skill turns inputs into editable SVG, HTML, print-ready PDF, reports, and speaker scripts, with templates, chart patterns, and multilingual support that fit common workflows. The core renderer needs neither extra dependencies nor network access, but browser-based generation of new content requires the user's Anthropic API key, and PDF and browser acceptance workflows add dependencies, so cost-benefit is not fully established.

README claims have traceable leads in scripts, specifications, examples, CI workflows, and test files, and the documentation distinguishes directly renderable patterns from patterns that only produce specifications. Multiple file types corroborate the stated design, but this static review cannot confirm runtime results or test success at the specified revision. The documentation labels sample data as fictional and separates design decisions from public references, though not every claim is tied to specific supporting evidence and fact-inference separation is only partial.