Pruner — Code Context Engine

Run one command to get context:

# Brief pointers (default)
pruner context /absolute/path/to/repo "<the user's ask>"

# Detailed output with execution paths and code snippets
pruner context /absolute/path/to/repo "<the user's ask>" --detail

Use an absolute repo path. Avoid cd ... && pruner context . in tool calls.

Workflow

1. Run pruner context first — default output is brief pointers (~2K tokens).
2. Work directly from the output.
3. Use --detail if pointers aren't enough — adds execution paths and code snippets (~10-15K tokens).
4. Read source files only if a snippet is truncated or you need nearby lines.
5. Do not re-explore the same query with grep/glob right after pruner.

When to escalate

• Read .pruner/context.md — if available, contains full execution paths and code snippets from the last run. Zero cost.
• Run with --detail — regenerates focused output (~10-15K tokens). Use for understanding/debugging queries.
• pruner show-symbol /path/to/repo "<name>" for callers/callees/signature.
• pruner show-file /path/to/repo "<path>" for symbols/imports in one file.

Modes

• --detail execution paths + code snippets (~10-15K tokens)
• --brief metadata only (~2K tokens)
• --full uncapped detail (~50-70K tokens on large repos)