Dependency Mapping

You answer three mandatory impact questions before any non-trivial code edit: what depends on this symbol, what breaks if it changes, and which tests cover it. Output is an evidence-backed impact report — not guesses from file names.

Hard Rules

Never edit code — map only. Editing routes to safe-change. Answer all three impact questions in references/IMPACT-QUERIES.md with cited paths before any edit elsewhere. Prefer LSP or tree-sitter when available; fall back to ripgrep and import tracing — document which ladder rung you used. Tag every claim [EXTRACTED] (read in source) or [INFERRED] (structural guess). If scope is whole-repo architecture, call codebase-understanding first, then return here for symbol-level blast radius.

Workflow

Step 1 — Scope the target

Identify: symbol name, file path, change type (signature, behavior, rename, delete). Ask ONE question if ambiguous: "Which symbol or file should I map?"

Step 2 — Run the capability ladder

1. LSP — go-to-references, find-references when the host exposes them.
2. tree-sitter / AST — import/call edges when a parser is available.
3. Text search — ripgrep for symbol name, qualified imports, string literals (routes, config keys).
4. Test map — search test/, __tests__/, *_test.*, *.spec.* for imports or describe blocks naming the symbol.

Record ladder rung in the report.

Step 3 — Answer the three-question gate

Read references/IMPACT-QUERIES.md. Fill every row with evidence. If any row is unknown, mark [AMBIGUOUS] and list what to verify — do not proceed to edits until resolved or user accepts risk.

Step 4 — Emit impact report

Use the output format below. State risk: low | medium | high with a plain-language reason.

Step 5 — Hand off

If the user will edit next, recommend safe-change with this report attached.

Gotchas

• Re-exports and barrel files (index.ts) hide real callers — trace through export chains.
• Dynamic dispatch (getattr, plugin registries, DI containers) may miss static references — flag as [INFERRED] gap.
• Generated code and lockfiles are not callers — exclude them.
• A symbol with zero static callers may still break via reflection, config, or API contracts.

Output Format

## Impact report — [symbol] in [file]

Capability ladder: [LSP | tree-sitter | text-search]
Risk: [low | medium | high] — [one sentence]

| Question | Answer | Evidence |
|----------|--------|----------|
| What depends on this? | [callers/importers] | [paths] |
| What breaks if I change it? | [breaking surfaces] | [paths] |
| Which tests cover it? | [test files] | [paths or NONE] |

Affected files: [list]
Recommended verify: [typecheck command] + [test command or "none — behaviorVerified: false"]

Examples

Teaser: User asks "what breaks if I rename validateSkill?" → report lists 4 importers in eval-pipeline + 2 tests; risk medium.

Full pairs: references/examples.md

Common Rationalizations

Verification

• All three impact questions answered with paths
• Capability ladder rung stated
• Risk enum assigned with reason
• No code edits performed

Red Flags

• Impact report with zero cited paths
• HIGH risk assigned without listing breaking surfaces
• Edit started before three-question gate complete

Prune Log

Last pruned: 2026-07-05

• Initial release from high-leverage skill spec (Skill 2 family)

Impact Report

Dependency mapped: [symbol] | Risk: [level] | Callers: N | Tests: N | Ladder: [rung]
Handoff: [safe-change recommended | map-only complete]