Skip to content

RUNG_TOOL_DESCRIPTION_SUFFIXES

const RUNG_TOOL_DESCRIPTION_SUFFIXES: Readonly<Record<ApprovalRung, string | null>>

Defined in: _worktrees/docs-release/gaunt-sloth/packages/core/dist/config/tool-descriptions.d.ts:62

§4.5’s table, verbatim. The wordings are normative copy the model reads; do not paraphrase, re-punctuate or “improve” them.

  • manual and write share one sentence: at both rungs a non-granted call goes to the human, so the user’s approval is a certainty, not a possibility.

  • assisted softens will to MAY: the rater approves what it rates safe, so only some calls reach the user.

  • auto has its own sentence, and the difference it states is real: [[EXT-29]]’s negotiation hands a call the rater will not clear back to the model, with the reason, so the first consequence of an unsafe-looking call there is a refusal it can answer rather than a person being asked — what changes is who answers first, and that is the one thing a model choosing a tool can act on. A shared wording would hide it in the model’s own tool-selection input.

    Its MAY therefore sits outside the contrast the three bullets above draw: it qualifies “be refused by the auto-rater”, not “require the user’s approval”, so this sentence says nothing either way about whether a person can be asked. That silence is deliberate — the string is §4.5’s row verbatim — and its cost is recorded here rather than left to be discovered: catastrophic escalates to a human on the FIRST hit with no rounds at all, so a person can be reached at auto and this input does not say so. Silent is not false, which is why the wording stands and why the premise quoted above still holds; a wording claiming the user is never asked would be false.

  • bypass appends nothing to anything: no gate, so no sentence could be true.