Prior Art

Find the existing evidence first, then summarize what it means. This skill is narrower than lev-research: it is for provenance, lineage, decision history, and "what already exists?" questions.

If the repo itself is unfamiliar, pair this with codebase-archaeology first so the prior-art sweep starts from real project structure instead of guesses.

1. Frame the search

Action: Identify the project scope, the target question, and whether the user wants one project or many.

Instruction: Treat each git repo or workspace as a separate project. Use the current working tree as the default project when no path is given. If .lev does not exist, do not fail; widen the crawl to the repo's other docs and markdown surfaces.

Validation: You can name the project(s), the evidence question, and the output shape before searching.

On failure: Ask for the missing repo path or clarify whether this is a project-specific or global hunt.

2. Use agent tools first, CLI second

Principle: Agent-native tools (Grep, Glob, Read, SemanticSearch) are instant. CLI tools (lev find, cass) have startup and backend costs. Always prefer agent tools over CLI for prior-art searches unless the question is genuinely cross-domain and local fused retrieval would add signal.