julie: Skill for Claude Code

Impact Analysis

Analyze the impact of changing a symbol by finding all references and assessing risk. Use this BEFORE modifying widely-used symbols.

Process

Step 1: Resolve the change target

If the user gives you a symbol name, resolve the definition first so you know which file or symbol they mean:

fast_search(query="<symbol_name>")
deep_dive(symbol="<symbol_name>", context_file="<partial_file_path>", depth="overview")

Use context_file when the name is ambiguous. If the target is described conceptually rather than named exactly, try fast_search(query="<concept>", backend="semantic") or backend="hybrid" to find candidate symbols. Semantic/hybrid fast_search results are symbol-only; use explicit lexical for file paths and pure lexical comparison. Default fast_search may show labeled semantic fallback candidates only after an identifier-like unscoped lexical zero-hit. blast_radius(symbol_ids=[...]) is the tightest seed mode, but only use it when another Julie result already gave you concrete symbol IDs. If all you have is a definition file, use file_paths=[...].

Step 2: One-shot impact via blast_radius

blast_radius(file_paths=["<definition_file>"], max_depth=2, include_tests=true)

blast_radius is the primary entry point for impact analysis. One call returns ranked impacted symbols with why-reasons, likely tests, and (for revision-range seeds) deleted files. It walks the reference graph deterministically, so you don't have to chain get_context → fast_refs → deep_dive to build the same picture.

You can seed it three ways:

• file_paths=["src/foo.rs"] — default when you know the changed file but not a symbol ID
• symbol_ids=["<id>"] — tighter impact when another Julie result already gave you concrete symbol IDs
• from_revision=<number>, to_revision=<number> — advanced mode using Julie's canonical revision numbers, not Git refs or SHAs

If the impact list is large, the first page includes spillover_handle=br_xxx. Hold onto it for Step 4.