Blog Audio Narrator

The documentation discloses the environment variable, external Gemini TTS data flow, FFmpeg conversion, and missing-key fallback; the script does not print the key and uses an isolated virtual environment. Points are deducted because generation has no explicit user or cost confirmation, article text is sent to Google without sensitive-content or privacy guidance, a requested output path may be overwritten because FFmpeg uses -y, and dependency and runtime network trust boundaries are not fully specified within this skill.

Commands, parameter validation, voice validation, empty-input checks, missing-file errors, API-error handling, rate-limit guidance, and long-text guidance are present, and the main workflow is mostly internally consistent. Points are deducted because this is a static review with no verification of the Gemini SDK response shape, audio conversion, or venv installation; the skill-specific requirement files are not provided, automatic setup may fail on dependency or network issues, and retry and failure diagnostics remain limited.

The audience, trigger phrases, modes, input forms, and primary outputs are reasonably clear, and multilingual support is claimed. Points are deducted for missing explicit non-fit cases, semantic trigger boundaries, and sensitive-content limits; Chinese support is asserted only through the generic 80+ language claim with no Chinese example or evidence; the core function depends entirely on Google Gemini TTS, with no mainland-China reachability assessment.

The skill has frontmatter, MIT licensing, version metadata, argument hints, quick reference, prerequisites, workflow steps, embedding examples, voice references, an error table, and on-demand reference loading. Points are deducted for lacking a skill-specific changelog, clearly assigned maintenance responsibility, and an explicit update path; the supplied material does not include the skill's dependency manifests; and setup is described as configuration although the implementation mainly checks, guides, and performs a dry run.

The skill specifies directly usable narration files, duration, voice, cost estimate, and HTML embed code across summary, full, and dialogue modes. Points are deducted because no representative successful artifact or skill-specific execution test is supplied; cost and duration are estimates, multilingual quality is unverified, and users may still need to audition audio, inspect paths, and review prepared text.

The supplied scripts expose auditable validation, cost-estimation, error paths, and output logic, while repository CI and test signals provide some maintenance evidence. Points are deducted because the shown tests are primarily for other repository scripts and do not demonstrate real blog-audio execution, Gemini API mocking, audio validity checks, or third-party corroboration; claims about 30 voices, 80+ languages, and pricing are largely author-provided.