Use this skill when the input is already good content that needs to fit a different audience: a research note repurposed for an executive summary, a marketing draft made less salesy, a friend's casual message turned into a professional intro, an academic abstract rewritten for a blog.
This skill does NOT:
- add or remove substantive claims (use
writer / essay-write / prose-edit to rewrite content);
- generate viral hooks (use
viral-text);
- check translation parity (use
translation-sync);
- audit canon (use
canon-check).
ROLE
Read input + target register → output the same content rewritten in the target register → run the writer 4-layer cleanup so the result is shippable.
PIPELINE
Detect source register. Read the input and identify its starting register (see references/registers.md). If the user hasn't named the source, infer it from sentence length / vocab / hedge density.
Confirm target register. The user must name the target. If unclear, ask: "Which register? casual / friendly-professional / business-formal / academic / technical / plain-explainer."
Identify deltas. Look up references/transformation-rules.md for the source→target pair and gather the specific changes: contractions on/off, sentence length, vocab swaps, hedge density, jargon level, person (1st / 2nd / 3rd).
Apply transformations clause-by-clause. Don't paraphrase the whole text in one shot — that loses content. Walk through each sentence and apply only the deltas that fit. Preserve original sentence boundaries when possible.
Pass through writer. Final 4-layer clean: typography, anti-slop regex categories, structural-prose, ru-calques. This is the same final pass every other wrapper runs — see writer/SKILL.md.
Diff-report. Show before/after side-by-side with a 1-line note on which deltas were applied. The user reviews; they can ask for a different register or a partial revert.
MODES
tone-shifter <text> --to <register> — shift to named targettone-shifter <text> --to <register> --from <register> — explicit source (skips detection)tone-shifter <text> --to <register> --conservative — apply only safe deltas (vocab + contractions); skip restructuringtone-shifter <text> --analyze — return only the detected source register + suggested targets, do not rewritetone-shifter <text> --profile <profile.json> — apply a custom brand-voice JSON profile (vocabulary, avoidWords, hooks, ctaPhrases) on top of register shifttone-shifter --infer-profile <samples...> — read 2-5 sample texts and output a brand-voice JSON profile for reviewtone-shifter <text> --verify-profile <profile.json> — read-only check: does the passage match the profile? Returns structured report.
REFERENCES (load on demand)
| File |
When to load |
| references/registers.md |
When detecting source or naming target — defines the 6 registers and their detection markers |
| references/transformation-rules.md |
After source+target known — the specific deltas to apply for each pair |
| references/brand-voice-profile.md |
When user provides a custom brand-voice profile (JSON), or asks for one to be inferred from samples — overlays registers with concrete vocabulary / banned words / hooks / CTAs |
EXAMPLES
See examples/before-after.md for 4 calibration pairs covering the most-common shifts.
CONSTRAINTS
- Do not change facts. If the input says "we raised $4M in 2023", the output cannot say "we raised several million" or "we raised in the early 2020s".
- Do not change structure unless register demands it. Casual register may merge two short academic sentences into one. Academic register may split a casual run-on. But the order of points stays.
- Do not invent transitions. The transformation is local. If the input has rough transitions, they stay rough — the user can run
prose-edit for that.
- Preserve quoted speech verbatim. Quoted material does not get tone-shifted; only narration around it.
- Person changes (1st → 3rd, you → reader) only if the target register strictly requires it. Otherwise preserve original person.
INVOCATION HINTS
When the user says any of:
- "make this more casual / formal / professional / friendly / academic / accessible"
- "rewrite for a younger / older audience"
- "turn this into [email / Slack / LinkedIn / blog / academic] tone"
- "less salesy", "less corporate", "less jargon", "less formal"
- "give me a plain-English version"
Use this skill. If the user wants a viral version (hook + CTA), use viral-text. If they want fiction voice, use prose-edit.
1---2name: tone-shifter3description: Rewrite text in a different register without changing meaning — formal↔casual, business↔academic, technical↔friendly, plain-explainer. 6 named registers + transformation deltas. Wraps `writer`. Use when the user says 'make this more casual / formal / accessible / business-like', 'rewrite for younger audience'.4license: MIT5---67<objective>8Tone-shift a passage from one register to another while preserving the underlying claims, facts, structure, and information density. This is a re-voicing pass, not a content rewrite — the reader should learn the same things, but in a different "voice".910Use this skill when the input is already good content that needs to fit a different audience: a research note repurposed for an executive summary, a marketing draft made less salesy, a friend's casual message turned into a professional intro, an academic abstract rewritten for a blog.1112This skill does NOT:13- add or remove substantive claims (use `writer` / `essay-write` / `prose-edit` to rewrite content);14- generate viral hooks (use `viral-text`);15- check translation parity (use `translation-sync`);16- audit canon (use `canon-check`).17</objective>1819## ROLE2021Read input + target register → output the same content rewritten in the target register → run the `writer` 4-layer cleanup so the result is shippable.2223## PIPELINE24251. **Detect source register.** Read the input and identify its starting register (see `references/registers.md`). If the user hasn't named the source, infer it from sentence length / vocab / hedge density.26272. **Confirm target register.** The user must name the target. If unclear, ask: "Which register? casual / friendly-professional / business-formal / academic / technical / plain-explainer."28293. **Identify deltas.** Look up `references/transformation-rules.md` for the source→target pair and gather the specific changes: contractions on/off, sentence length, vocab swaps, hedge density, jargon level, person (1st / 2nd / 3rd).30314. **Apply transformations clause-by-clause.** Don't paraphrase the whole text in one shot — that loses content. Walk through each sentence and apply only the deltas that fit. Preserve original sentence boundaries when possible.32335. **Pass through `writer`.** Final 4-layer clean: typography, anti-slop regex categories, structural-prose, ru-calques. This is the same final pass every other wrapper runs — see `writer/SKILL.md`.34356. **Diff-report.** Show before/after side-by-side with a 1-line note on which deltas were applied. The user reviews; they can ask for a different register or a partial revert.3637## MODES3839- `tone-shifter <text> --to <register>` — shift to named target40- `tone-shifter <text> --to <register> --from <register>` — explicit source (skips detection)41- `tone-shifter <text> --to <register> --conservative` — apply only safe deltas (vocab + contractions); skip restructuring42- `tone-shifter <text> --analyze` — return only the detected source register + suggested targets, do not rewrite43- `tone-shifter <text> --profile <profile.json>` — apply a custom brand-voice JSON profile (vocabulary, avoidWords, hooks, ctaPhrases) on top of register shift44- `tone-shifter --infer-profile <samples...>` — read 2-5 sample texts and output a brand-voice JSON profile for review45- `tone-shifter <text> --verify-profile <profile.json>` — read-only check: does the passage match the profile? Returns structured report.4647## REFERENCES (load on demand)4849| File | When to load |50|---|---|51| [references/registers.md](references/registers.md) | When detecting source or naming target — defines the 6 registers and their detection markers |52| [references/transformation-rules.md](references/transformation-rules.md) | After source+target known — the specific deltas to apply for each pair |53| [references/brand-voice-profile.md](references/brand-voice-profile.md) | When user provides a custom brand-voice profile (JSON), or asks for one to be inferred from samples — overlays registers with concrete vocabulary / banned words / hooks / CTAs |5455## EXAMPLES5657See [examples/before-after.md](examples/before-after.md) for 4 calibration pairs covering the most-common shifts.5859## CONSTRAINTS6061- **Do not change facts.** If the input says "we raised $4M in 2023", the output cannot say "we raised several million" or "we raised in the early 2020s".62- **Do not change structure unless register demands it.** Casual register may merge two short academic sentences into one. Academic register may split a casual run-on. But the order of points stays.63- **Do not invent transitions.** The transformation is local. If the input has rough transitions, they stay rough — the user can run `prose-edit` for that.64- **Preserve quoted speech verbatim.** Quoted material does not get tone-shifted; only narration around it.65- **Person changes (1st → 3rd, you → reader) only if the target register strictly requires it.** Otherwise preserve original person.6667## INVOCATION HINTS6869When the user says any of:70- "make this more casual / formal / professional / friendly / academic / accessible"71- "rewrite for a younger / older audience"72- "turn this into [email / Slack / LinkedIn / blog / academic] tone"73- "less salesy", "less corporate", "less jargon", "less formal"74- "give me a plain-English version"7576Use this skill. If the user wants a viral version (hook + CTA), use `viral-text`. If they want fiction voice, use `prose-edit`.
Mikefluff