Dashi PPT Skill — AI-Powered Presentation Generator

Evidence: README claims zero content upload, all generation, editing, and export run locally, with only npm dependency installation and a silent version check touching the network, and the local preview server is LAN-only for viewing, export endpoints local-only. This shows data flow transparency, but lacks user confirmation mechanisms (e.g., explicit prompts before install or export), and no specific least-privilege details (e.g., required file system access scope). Dependency security lacks a dependency list or security audit info. External effects (network requests) are described but no user control options. Rollback mechanism not mentioned. Source attribution via copyright and GitHub link, but publisher unverified. Thus, data_flow_transparency scores 2, others 0-1 due to insufficient evidence.

Evidence: README is consistent in feature descriptions, no contradictions found, but no tests or verification results provided. Dependency availability requires Node.js 20+ and Chrome, but no dependency list or version pinning. Failure messages only mention PPTX export failure requiring local Chrome, but no detailed error handling or user guidance. Thus, self_consistency scores 2, dependency_availability and failure_messages each 1.

Evidence: README clearly defines suitable scenarios (industry research, funding reviews, etc.) and non-fit (pixel-perfect hand-crafted design), and lists supported platforms (Claude Code, Codex, etc.) with requirements (Node.js, Chrome). Capability boundaries clear, but trigger precision lacks specific trigger words or command examples beyond install command. Environment fit requires Node.js 20+ and Chrome, but no other environment variables or config. Thus, audience_and_scenarios, capability_boundaries, environment_fit each 2, trigger_precision 1.

Evidence: README has clear structure with install, usage, FAQ, license sections, good information architecture. Install notes detailed with command and parameters. Naming stability not mentioned (no version naming rules or API stability). Examples and FAQ provide usage examples and common questions. Known limitations clearly state non-fit scenarios and customization limits. License is AGPL-3.0 with proprietary component exception, note. Versioning changelog not provided. Maintenance responsibility: copyright to author, but no maintenance commitment or contribution guide. Thus, information_architecture, install_notes, examples_and_faq, known_limitations, license each 2, naming_stability and maintenance_responsibility each 1, versioning_changelog 0.

Evidence: Output usability: README claims editable HTML and PPTX with export, but no actual output samples or user reviews. Marginal value: provides 12 themes, 1020 layouts, 8576 tunable controls, feature-rich, but no comparison with other tools. Cost-benefit: mentions ~100k tokens for 10 pages, but no detailed cost analysis or comparison. Thus, output_usability and marginal_value each 2, cost_benefit 1.

Evidence: Claims (e.g., theme count, layout count) lack verifiable tests or data sources, only images. Cross-source corroboration: no third-party verification or independent testing. Fact-inference separation: README distinguishes feature descriptions and suitable scenarios, but does not explicitly separate facts from inferences. Thus, claim_traceability 1, cross_source_corroboration 0, fact_inference_separation 1.