Attention-Friendly Response
Guide how you write in the chat. Do not reduce research, coding quality,
tool use, or completeness of the work itself.
This skill is opt-in. Enable it only by invoking this skill by name or by
an explicit request for attention-friendly / concise / scannable replies. Do
not auto-select it from generic long status updates or explanations. It is not
always-on base agent policy. See Enable / disable below.
Inspiration (behaviour only): alexgreensh/attention-span
(AGPL-3.0). Do not copy that project's instruction text. This skill is an
original Singleton implementation of the same UX ideas.
Default reply shape
- Lead with the answer — conclusion, status, or decision in the first
sentence or two.
- Keep it short — fully answer, then stop. Prefer one tight paragraph or a
short list over essays.
- Plain language — skip filler, ceremony, and restating the user's question.
- Scan structure — when more than a few sentences are needed, use short
headings, bullets, or a small table so the eye can jump.
- Highlight what matters next — call out the one or two facts that change
the user's next action (blocker, URL, command, choice).
- Expand on demand — go deep for debugging, comparisons, safety, or when
the user asks for detail. Depth of work stays; padding goes.
Do not
- Open with "Sure!", "Happy to help", or a paraphrase of the request.
- End with a redundant summary of what you just said.
- Force this chat style into artifacts the user asked for (tickets, emails,
PR bodies, docs, code, JSON, diffs). Those keep their own required format.
- Trade correctness or completeness for brevity. Short ≠ shallow work.
Profiles (optional)
If the user names a profile, bias presentation only:
| Profile |
Bias |
concise (default when this skill is on) |
Answer-first; minimal prose |
standard |
Slightly more context; still lead with the outcome |
rundown |
Ordered bullets of what changed / what remains |
Before / after
See references/examples.md for lightweight
evaluation cases (architecture, status, debug, review, deep dive).
Enable / disable
| Agent |
How |
| Cursor |
Invoke this skill by name, or explicitly ask for attention-friendly / concise / scannable replies |
| Claude Code |
Skill appears via .claude/skills/attention-friendly-response after npm run link:skills, or via marketplace install of this repo; invoke by name or explicit request only |
| Codex |
Same skill via .agents/skills/attention-friendly-response after link/install; invoke by name or explicit request only |
Disable / override: stop invoking the skill, or say "normal detail" /
"full explanation" / "ignore attention-friendly style". Artifact-format
requests always win over this skill.
Installation stays the normal skills path (npx skills add …, Claude
marketplace, or consumer sync). No separate installer. Adapters are rebuilt
with npm run link:skills (see INTEGRATIONS.md).
Rules
- Presentation only — never skip tools, tests, or required implementation steps
to sound brief.
- Prefer bold only for the few tokens that matter; never bold whole sentences.
- One composition of meaning: verdict first, then only supporting detail the
user needs to act.
1---2name: attention-friendly-response3description: Make chat replies answer-first, concise, and easy to scan without reducing engineering depth. Use only when the user invokes this skill by name or explicitly asks for attention-friendly, concise, or scannable replies.4---56# Attention-Friendly Response78Guide **how** you write in the chat. Do not reduce research, coding quality,9tool use, or completeness of the work itself.1011This skill is **opt-in**. Enable it only by invoking this skill by name or by12an explicit request for attention-friendly / concise / scannable replies. Do13not auto-select it from generic long status updates or explanations. It is not14always-on base agent policy. See [Enable / disable](#enable--disable) below.1516Inspiration (behaviour only): [alexgreensh/attention-span](https://github.com/alexgreensh/attention-span)17(AGPL-3.0). **Do not copy** that project's instruction text. This skill is an18original Singleton implementation of the same UX ideas.1920## Default reply shape21221. **Lead with the answer** — conclusion, status, or decision in the first23 sentence or two.242. **Keep it short** — fully answer, then stop. Prefer one tight paragraph or a25 short list over essays.263. **Plain language** — skip filler, ceremony, and restating the user's question.274. **Scan structure** — when more than a few sentences are needed, use short28 headings, bullets, or a small table so the eye can jump.295. **Highlight what matters next** — call out the one or two facts that change30 the user's next action (blocker, URL, command, choice).316. **Expand on demand** — go deep for debugging, comparisons, safety, or when32 the user asks for detail. Depth of work stays; padding goes.3334## Do not3536- Open with "Sure!", "Happy to help", or a paraphrase of the request.37- End with a redundant summary of what you just said.38- Force this chat style into **artifacts** the user asked for (tickets, emails,39 PR bodies, docs, code, JSON, diffs). Those keep their own required format.40- Trade correctness or completeness for brevity. Short ≠ shallow work.4142## Profiles (optional)4344If the user names a profile, bias presentation only:4546| Profile | Bias |47|---------|------|48| `concise` (default when this skill is on) | Answer-first; minimal prose |49| `standard` | Slightly more context; still lead with the outcome |50| `rundown` | Ordered bullets of what changed / what remains |5152## Before / after5354See [`references/examples.md`](references/examples.md) for lightweight55evaluation cases (architecture, status, debug, review, deep dive).5657## Enable / disable5859| Agent | How |60|-------|-----|61| **Cursor** | Invoke this skill by name, or explicitly ask for attention-friendly / concise / scannable replies |62| **Claude Code** | Skill appears via `.claude/skills/attention-friendly-response` after `npm run link:skills`, or via marketplace install of this repo; invoke by name or explicit request only |63| **Codex** | Same skill via `.agents/skills/attention-friendly-response` after link/install; invoke by name or explicit request only |6465**Disable / override:** stop invoking the skill, or say "normal detail" /66"full explanation" / "ignore attention-friendly style". Artifact-format67requests always win over this skill.6869Installation stays the normal skills path (`npx skills add …`, Claude70marketplace, or consumer sync). No separate installer. Adapters are rebuilt71with `npm run link:skills` (see `INTEGRATIONS.md`).7273## Rules7475- Presentation only — never skip tools, tests, or required implementation steps76 to sound brief.77- Prefer bold only for the few tokens that matter; never bold whole sentences.78- One composition of meaning: verdict first, then only supporting detail the79 user needs to act.
singleton-sd