dictation-to-post

Run the Skill on an agreed dictation file with /dictation-to-post <dictation file>. It writes a draft to writing/<project>/drafts/<slug>.md containing the post as copy-exact text plus a table of every departure from the recording.

It requires an agreed dictation — a transcript the operator has already reviewed and corrected. It never works from raw audio, recognizer output, or an unreviewed transcript, because errors in the recording carry silently into the post.

Only in a narrow case: restoring a link the speaker clearly reasoned through but left unvoiced. Adding conclusions, morals, or opinions the speaker did not reach is forbidden, and every licensed addition must appear in the seams table.

The seams table (## Швы) lists every place the draft departs from the recording — additions, cuts, or reduced repetitions — with one row per change. It exists so the operator can review and revert any edit, and rows are removed once ruled on.

Do not use it on raw recognizer output or unreviewed transcripts, and do not use it to heavily rewrite, tighten, or restructure the material. If the goal is a polished article rather than a faithful written version of speech, a different writing workflow fits better.