Explain with evidence

Skill dlwlgus9125/EZPowers/skills/explain-with-evidence

Structure user-facing explanations and work-result reports in the user's active conversational language, grounded in observed evidence and honest limits. Use when explaining completed work, technical decisions, knowledge, concepts, numbers, or rationale beyond a one-line confirmation or status update; not to rewrite fixed-schema artifacts or machine-readable output.From its SKILL.md

Install

npx -y skills add dlwlgus9125/EZPowers --skill explain-with-evidence

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

12 stars12 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.7 KB, as published. Nobody here has run it

Explain With Evidence

Improve presentation only. Do not change the task, authority, evidence, or completion contract supplied by the user, repository, host, or another skill.

Match the user's language

Infer the response language from the latest substantive natural-language user message. An explicit language instruction overrides inference.
If that message is too short or genuinely mixed, use the dominant language of the recent conversation.
Ignore code, commands, paths, identifiers, model names, quoted material, and required proper nouns when inferring or translating. Preserve them as needed.

Choose the smallest useful shape

Result report: outcome, reason when useful, material change, observed verification, then remaining limits.
Deep explanation: context, alternatives actually considered, useful discovery sequence, decisive finding, observed evidence, then remaining uncertainty or the next check.

Use the result shape for ordinary task completion. Use the deep shape only when the user requests a detailed explanation, rationale, retrospective, or similar narrative. Follow a user-requested format instead of either shape. Omit stages that add no information; do not repeat an opening conclusion at the end.

Ground every stage

Include an alternative, objection, reversal, or discovery sequence only when it was actually considered or observed and helps explain the result.
Distinguish direct observation from inference and unverified possibility.
Never invent a measurement. Tie numbers and PASS/FAIL claims to commands, artifacts, sources, or other evidence actually inspected.
Prefer concrete cases and exact evidence over analogy. Use analogy only when it materially helps an unfamiliar audience.

Preserve fixed contracts

Do not reshape code, commands, specs, plans, JSON, tool output, receipts, certificates, or other fixed-schema artifacts. Explain around them when useful. Never soften or narratively reinterpret exact states such as PASS, FAIL, BLOCKED, READY, or CERTIFIED. This skill adds no tools, workflow transition, write authority, retry policy, or completion authority.

Adaptation notice: this file is a substantially modified EZPowers adaptation of j-explain-style; see LICENSE in this skill directory.