Validates voice-dictated requests with readback checks and resolves ambiguous terms before execution.
npx skills add https://github.com/wkentaro/skills --skill hear-me-wkentaro
Please help me install this Agent Skill. Skill: hear-me Source: https://github.com/wkentaro/skills/tree/main/skills/productivity/hear-me Command: npx skills add https://github.com/wkentaro/skills --skill hear-me-wkentaro
ruff check from "rough check"). - Ambiguity Resolution: Marks unresolved phrases with credible candidates (e.g., [unclear: feature/cash-sync or feature/cache-sync?]) and uses conversation context and read-only checks to disambiguate before acting. - Persistent Dictation Mode: Stays active for the whole conversation once enabled, applying the protocol to every later message until explicitly turned off. - Use Case: You dictate "run rough check on the source folder" and the assistant replies with Heard: run ruff check (from "rough check") on the src/ directory, confirming the corrected interpretation before executing. ## Quick Start Use hear-me to turn on dictation mode and check this request before acting on it.High-intent search queries and answers about installing and using this skill.
Enable dictation mode and the assistant produces a Heard: readback restating the action, target, and constraints of each request before acting. Corrections are shown as code with the original transcript, so you can spot misheard commands, paths, or names immediately.
Ambiguous phrases are marked with their credible candidates, such as [unclear: feature/cash-sync or feature/cache-sync?], instead of guessing. The assistant resolves them using conversation context and read-only checks, and asks one focused question only when multiple interpretations remain.
Yes, once enabled the mode applies the readback protocol to every later message in that conversation, including typed text and skill invocations. Only an explicit off control disables it, and each new conversation starts with the mode off.
No, the protocol applies to every message while the mode is on; the only controls are on and off for the whole conversation. Sending off disables dictation mode entirely, and the assistant replies with only a state confirmation.
When two or more credible interpretations remain, the assistant asks one focused question and completes any independent parts of the request that do not depend on the ambiguous term. It only mutates state once both the action and its target are resolved.
Dimi82Capture a single voice input and return a text-only transcription.
LiuHarry1Execute literal shell commands directly in the terminal and report exit status and output.
TeamFlint-DevCorrect speech-to-text errors in dictated text using a project-specific glossary.
hulusi-tuncDesign voice interfaces with inclusive recognition and text alternatives.