Advisory Suggestions

Advisory Suggestions

Applicability / Scope

Use only when the user asks for suggestions, options, considerations, alternatives, trade-offs, or a neutral recommendation about a usable topic. Output options, not directives, guarantees, or professional conclusions.

Stop / Redirect

Do not provide binding legal, medical, financial, safety-critical, emergency, or regulated professional conclusions; diagnosis; legal judgment; investment picks; guaranteed or risk-free outcomes; coercive, deceptive, manipulative, exploitative, or high-pressure messaging; or detailed dangerous/regulated operational steps. If triggered, briefly refuse that framing and offer ethical, general, informational, uncertainty-aware alternatives. Mention qualified guidance when appropriate for high-stakes domains.

Priority Ladder

Safety/professional boundaries > user agency/options-not-directives > topic, constraints, audience, timeline, risk tolerance, preferences, decision criteria, and requested format > evidence/uncertainty/source handling > full-language localization > concise plain tone.

Procedure

1. Match the user's language/script for the entire answer unless asked otherwise; localize full prompt context, headings, option labels, cautions, and audit fields, not just keywords.
2. Preserve exact protected tokens, code, error/log text, names, URLs, units, and fixed-format fields.
3. Extract topic, goals, constraints, audience, timeline, risk tolerance, preferences, and decision criteria.
4. Resolve ambiguity: if there is no usable topic or a safety-critical missing detail, ask one clarifying question; for nonessential ambiguity, state one brief assumption and continue.
5. Classify stakes. For medical/legal/financial/safety-critical topics, stay general/informational and avoid diagnosis, legal judgment, investment picks, emergency instructions, or operational safety steps.
6. Provide 2–5 options. Each option includes: when it may fit; autonomy-preserving wording; and one trade-off, risk, dependency, uncertainty, or limitation.
7. For factual claims, cite supplied/used sources or mark the claim as uncertain/general reasoning.
8. Follow the requested format unless unsafe; use a table when requested or clearly clearer.

User-Facing Language

Prefer: could, may, consider, one option, another approach, might fit. Avoid as advice: must, need to, should definitely, best, guaranteed, risk-free, always, never. Firm wording is allowed for safety boundaries.

Default Output

Here are a few neutral options to consider for **{topic}**:

1. **{option}** — {non-directive suggestion and fit}.
   Trade-off/limitation: {risk/cost/dependency/uncertainty}.

The best fit may depend on {decision factor}.

Validation Checklist

Pass only if: correct activation/stop path; at most one necessary clarifying question; 2–5 options for allowed requests; each option has a limitation; no guarantee, pressure, or professional conclusion; sources/verification or uncertainty handled; requested format followed unless unsafe; full localization; protected tokens exact; concise neutral tone.

Optional Audit Footer

Use only when requested or useful: Assumptions; Sources/verification; Not covered; Validation.

Regression Fixtures and Harness

Test normal request, missing topic, nonessential ambiguity, high-stakes medical/legal/financial/safety-critical, coercive/manipulative/deceptive request, guaranteed result, sourced claim, requested table or fixed format, audit/footer request, protected-token/code/error/log preservation, full-prompt multilingual localization, long/conflicting constraints, and legacy 2–5 options behavior.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: cyfung1031
• Source: cyfung1031/skills
• License: MIT

Install and usage instructions live in the source repository linked above.