Case Law Research

Case-Law Research

Conduct a structured search of U.S. case law via CourtListener, read the opinion text, and produce a grounded research report. Every statement in the report must be traceable to retrieved text — do not assert what a case holds unless you have read the relevant passage.

When this skill applies

Apply when the user asks to locate, read, or cite court decisions on a legal question. Common triggers: finding controlling or persuasive authority, confirming whether authority exists, reading specific language from an opinion, assembling citations for a brief or memo.

Do not apply this skill when:

• The user asks about statutory text, regulations, or administrative guidance. Those are not in the CourtListener corpus; direct the user to the appropriate source.
• The user asks about non-U.S. law. CourtListener covers U.S. federal and state courts only; this skill does not extend beyond that coverage.
• The user needs to confirm whether a case is still good law (overruled, distinguished, limited). This skill is not a citator. It does not perform Shepardizing or KeyCiting. Always recommend the user validate any case's current status in Westlaw, Lexis, or a comparable citator before relying on it.

Inputs

Required: the question input — the legal question or issue to research. If the user has not stated a question clearly, ask before searching.

Optional: the jurisdiction input — a court or jurisdiction to focus on (e.g., "9th Circuit", "New York Court of Appeals", "SCOTUS"). When provided, weight searches toward that jurisdiction. When absent, begin with a broad U.S. search and note which courts produced the most relevant results.

If the user's question is ambiguous (e.g., "case law on good faith") and the jurisdiction materially affects which results are relevant, ask one clarifying question before searching. Do not block on optional inputs.

Tool chain

This skill runs through the LQ.AI governed gateway. All CourtListener tool calls are tier-gated and audited; the "Sources consulted" panel (visible to the user after the run) shows the provenance trail.

Step 1 — search_case_law

Call search_case_law with a query derived from the user's question. Tips:

• Use specific legal terms from the question rather than paraphrasing in lay language.
• If a jurisdiction was provided, include it as a filter.
• Prefer narrower queries first; broaden if the initial results are sparse or off-topic.
• Run multiple searches if the question has distinct sub-issues (e.g., a threshold test and a remedy standard).

search_case_law returns a list of candidate clusters (cases). Each cluster includes a docket identifier and case metadata. Do not assert anything about the case yet — the search result is a pointer, not a full read.

Step 2 — get_cluster

For each promising candidate from Step 1, call get_cluster with the cluster identifier to retrieve the case metadata and the list of opinion documents within it. This confirms the court, date, parties, and which opinion(s) are available to read.

Select the opinion(s) most likely to contain the relevant passage (majority opinion first; concurrences or dissents only if the question specifically implicates them).

Step 3 — read_opinion or find_in_case

To read the full opinion text, call read_opinion with the opinion identifier from the cluster. For long opinions or targeted lookups, call find_in_case with search terms to locate the specific passage.

Ground every statement in retrieved text. Before citing a case for a proposition, locate the supporting passage. Quote or closely paraphrase the relevant excerpt in the report. Do not summarize from memory or from the search-result snippet alone.

If read_opinion or find_in_case returns no content or an error, record the gap explicitly in the report.