Build a Repo Map for Code Navigation

PROJECT CONTEXT:
My agent either reads whole files into context (expensive, mostly noise) or works blind (edits the wrong places). The repo-map pattern proven in leading agents extracts symbol signatures per file and ranks what to show, giving orientation at a fraction of the cost.

GOAL:
Build a repo map: parse files into symbol signatures, rank by relevance to the task, render within a token budget, and integrate it into the agent's context.

REQUIREMENTS:
- Symbol extraction (functions, classes, methods with signatures) via tree-sitter or language parsers
- A per-file digest: path plus its key symbols, compact enough to aggregate
- Relevance ranking so the map shows what matters for the current task
- A token-budgeted renderer that degrades gracefully (more files = less detail each)
- Map freshness: updated when files change, cached when not

CONSTRAINTS:
- The map orients; it does not replace reading - the agent still opens files to edit
- Total map cost must stay a small, fixed slice of the context budget

DEFINITION OF DONE:
- The agent locates the right file for a feature request without directory-crawling
- The map for a mid-sized repo renders within its token budget
- Task-relevant files rank visibly higher than unrelated ones
- Edits to a file are reflected in the map on next use without a full rebuild

COMMON FAILURES TO AVOID:
- Dumping the file tree and calling it a map - names without symbols barely orient
- Unranked maps where the budget is spent alphabetically instead of relevantly
- Stale maps confidently pointing at moved or deleted symbols
- Maps so large they recreate the raw-dump problem they were meant to solve