Imported from MohamedTyr/counselle (skills/deep-research/SKILL.md). Install upstream with npx skills add MohamedTyr/counselle --skill deep-research. Copyright stays with the author.

Deep Research

Briefly establish the decision being made and the material research axes before collecting evidence. Use Counselle's facts store first for covered school facts.

Keep the investigation bounded to the decision. For a two-school comparison, choose at most three material axes, read only the domains or sources needed for those axes, and synthesize once the evidence is sufficient. Do not keep searching for completeness after the decision-relevant picture is clear.

If tool budget or time is becoming tight, stop collecting evidence and write the best bounded synthesis from the sources already gathered. Never make the final answer only a tool-budget apology when cited evidence is already available.

When the source-routing contract applies, load counselor-research. Use official school sources for current policy and program facts, broad web for discovery or context, and Reddit only for clearly labeled lived-experience signals.

Triangulate important claims, resolve or disclose conflicts, and distinguish fact from inference. Look for unknown unknowns that could invert the recommendation.

Lead the final synthesis with the answer or recommendation, then organize the evidence by decision axis. Be comprehensive, but never pad; length must be earned by material axes.

If the scope is outcome-changing and genuinely unspecified, ask one focused natural-language question and wait for the student's next ordinary message before expensive research. Do not call or imply a clarification widget.

Never claim multi-agent or GPT-Researcher execution unless that subsystem ships and this skill is deliberately revised.

Preserve all honesty, citation, source, authorization, read-only, and value-reading rules.