| name | tone-modes |
| title | tone-modes |
| description | Switch between lawyer mode and casual mode, apply the read-aloud test, and use scripts as scaffolds rather than oracles — triggers whenever Claude is about to draft user-facing language or is about to treat a script's output as final. |
| author | EvanOchsner |
| author_url | https://github.com/EvanOchsner/personal-advocacy-toolkit/tree/main/.claude/skills/tone-modes |
| license | Apache-2.0 |
| version | 0.1.0 |
| execution_mode | open |
| jurisdiction | general |
| practice | general |
| language | en |
tone-modes
Two registers, one test, one rule about scripts. This skill is a
portable codification of a pattern: keep the internal reasoning
candid, keep the outbound language defensible, and never let a script
pretend to be a lawyer.
When this skill fires
- About to draft any user-facing artifact: letter, complaint,
position paper, email, exhibit cover.
- About to paraphrase something the user said into paperwork.
- About to present the output of an intake / triage / packet script
as the answer rather than as a scaffold.
The two modes
Lawyer mode. The register for anything that leaves the private
workspace. Properties:
- Claims are stated as positions, not as feelings. Every factual
assertion is traceable to an exhibit or a cited rule.
- Adversarial language ("they're trying to re-underwrite the policy
after the loss") is rephrased ("the insurer's conduct is
inconsistent with the agreed-value endorsement at policy
inception").
- Hedges where warranted — "the record shows," "the file reflects,"
not "obviously" or "clearly."
- Tight. The read-aloud test (below) is the pressure check.
Casual mode. The register for internal reasoning, comments in
drafts, strategy discussions, and the user's own notes. Properties:
- Honest about weak legs of the argument. If the specialist-labor-rate
premium is the weakest line item, casual mode says so.
- Uses the user's own idiom, not courthouse idiom.
- Free to name what the counterparty is doing in plain language.
- Must not leak into the final artifact. Anything written in casual
mode needs a translation pass before it goes outside.
The read-aloud test
Before any lawyer-mode paragraph ships, read it out loud (literally —
subvocalize the words). Two filters:
- Would a hostile reader quote this sentence back at you? If the
sentence contains a word that reads as snide, speculative, or
emotionally loaded, it's a gift to opposing counsel. Rephrase.
- Does every noun and verb carry its own weight? Adverbs
("grossly," "blatantly") and intensifiers ("clearly," "obviously")
almost always weaken the sentence they inhabit. Cut them.
If a sentence passes both filters out loud, it's ready.
Scripts as scaffolds, not oracles
The toolkit's scripts — situation_classify.py,
authorities_lookup.py, pii_scrub.py, build.py — are
deterministic rules engines against YAML tables. They are good at
what they do. They do not:
- Know whether the user's framing is correct.
