md2wechat: AI-Agent CLI for WeChat Official Account Publishing

Evidence shows: README clearly distinguishes side-effect-free inspect/preview from explicit-side-effect convert (--draft/--upload), requiring explicit user request for upload/draft side effects, demonstrating user confirmation. However, there is no systematic statement on least privilege, e.g., whether the CLI requests more permissions than needed. Data flow transparency is insufficient: it does not detail what data is sent to which external services (e.g., API service, image providers). Sensitive data handling: mentions not outputting Secret, but does not explain how credentials are stored and protected. Dependency security: go.mod lists dependencies, but no vulnerability scanning or update policy. External effects: clearly states convert side effects require explicit request, but does not explain potential impact on WeChat accounts. Rollback: no rollback mechanism or undo operation. Source attribution: README and LICENSE provide author info, but publisher is unverified.

Evidence shows: README and docs describe command behavior consistently, e.g., inspect returns structured metadata, preview only writes successful conversion HTML, convert executes side effects only on explicit request, demonstrating self-consistency. However, dependency availability: relies on external API services (professional API mode) and image providers, with no availability guarantees. Failure messages: docs mention preview does not create file on failure, but no detailed error message examples or troubleshooting guide.

Evidence shows: README identifies target audience (AI agents, individual creators) and scenarios (Markdown to WeChat HTML, pre-publish checks, multi-account publishing), and provides adaptation notes for multiple agents. Capability boundaries: clear via discovery commands like capabilities, doctor, themes, layout. Trigger precision: command boundaries fixed, e.g., inspect has no side effects, convert requires explicit request. Environment fit: provides installation and configuration guides, but does not specify support for non-Linux environments.

Evidence shows: README provides clear information architecture including quick start, professional API, agent workflow, advanced layout, common commands, and doc links. Install notes: provides npm install and configuration guides. Naming stability: command names stable, but no version naming policy. Examples and FAQ: provides example commands and FAQ doc. Known limitations: mentions local layout validate cannot prove remote renderer deployment, but does not list all limitations. License: provides detailed Source Available License terms. Versioning/changelog: package.json has version, but no CHANGELOG content. Maintenance responsibility: provides contact and commercial licensing info, but no explicit maintenance commitment.

Evidence shows: Output usability: convert returns final HTML, inspect returns structured metadata, suitable for agents. Marginal value: provides unique features like AI image generation, multi-account management, advanced layout. Cost-benefit: free AI mode for experimentation, professional API mode requires payment, but no pricing info, making cost-benefit hard to assess.

Evidence shows: Claim traceability: feature claims (e.g., 68 layout scenarios, 53 syntax names) are supported by docs, but no test evidence. Cross-source corroboration: no third-party verification or user feedback. Fact-inference separation: docs distinguish local validation and remote rendering validation, but do not clearly separate facts and inferences.