Discovery & Search Tools¶
Discovery Tools¶
create_file_manifest¶
Creates a complete inventory of the repository using ripgrep. This is typically the first tool called during analysis.
- Produces
files.all.txtwith all tracked files - Respects
.gitignorepatterns - Provides file counts by extension for language detection
repomix_orientation¶
Generates a token distribution tree showing project structure with token counts per directory and file. Helps the agent understand the codebase's shape before diving into specific areas.
repomix_bundle¶
Bundles selected files into a single markdown document using repomix with Tree-sitter compression. The output preserves file boundaries and includes metadata.
repomix_bundle_with_context¶
Bundles repository files with git context (diffs and commit logs) in a single call. Unlike repomix_bundle which reads from a file list, this tool operates directly on a repo path with optional glob patterns.
- Always includes git context (working tree diffs and/or recent commit history)
- Supports
include_patternsfor scoping (e.g.,"src/**/*.py,tests/**/*.py") - Configurable
include_logs_countfor number of recent commits - Optional Tree-sitter compression and base64 truncation
repomix_compressed_signatures¶
Extracts signatures and type information only, stripping function bodies via Tree-sitter. This provides the API surface of a codebase at a fraction of the token cost (Tenet 3).
repomix_split_bundle¶
Splits large bundles into multiple chunks that fit within token budgets. Used for monorepos or large codebases that exceed single-bundle limits.
repomix_json_export¶
Exports bundle data as structured JSON for programmatic consumption by downstream tools.
write_file_list¶
Writes a list of file paths to a file for subsequent repomix_bundle calls. Used to create the curated file list (e.g., files.business.txt) before bundling.
- Deduplicates and sorts paths
- Creates parent directories automatically
- Returns the output path and file count
Search Tools¶
rg_search¶
Text search using ripgrep with support for regex patterns, file type filtering, glob filtering, and context lines. Supports a count_only mode that returns per-file match counts without the actual content.
Used for finding specific patterns, imports, configuration values, and string literals across the codebase.
read_file_bounded¶
Reads a specific file with line range bounds. Used sparingly -- only for files that have earned deep reading through high scores across multiple signals.
write_file¶
Writes content to a file within the .code-context/ output directory. Used by the agent to persist CONTEXT.md, FILE_INDEX.md, and other analysis artifacts.
- Path must be within a
.code-context/directory (security enforced) - Creates parent directories automatically
- Returns the written path and byte count