bundle-analyser

Part of the analyser family — analyses collections of files in folders or zip archives.

Accepts a folder path or a zip file, walks all files, dispatches each to the appropriate analyser family member (via auto-analyser), and returns per-file signals plus structural signals about the collection as a whole.

Status

Coming soon. Name reserved on PyPI.

Analyser Family

Tool	Input	Role
auto-analyser	single file or zip	routes to the right specialist
bundle-analyser	folder or zip	walks a collection, calls auto-analyser per file
git-analyser	local git repo or remote URL	commit history + per-file signals
code-analyser	source file	Python, JS, TS, HTML, CSS, SQL signals
document-analyser	document file	PDF, DOCX, Markdown signals

Port

8008 (reserved)

Local (CLI) routing

bundle-analyser dispatches every file through auto-analyser, whose built-in defaults expect HTTP services on localhost:800x. For a fully local setup — specialists installed in the same venv, invoked as CLIs — point auto-analyser at them via ./auto-analyser.yaml (beside where you run from) or ~/.config/auto-analyser/config.yaml:

analysers:
  document-analyser: { type: cli, command: document-analyser }
  code-analyser: { type: cli, command: code-analyser }
  speech-analyser: { type: cli, command: speech-analyser }
  video-analyser: { type: cli, command: video-analyser }
  image-analyser: { type: cli, command: image-analyser }
  records-analyser: { type: cli, command: records-analyser }
  diagram-analyser: { type: cli, command: diagram-analyser }

Each listed member must expose the family contract: a manifest subcommand and <command> <file> --json → JSON on stdout. Unlisted members keep the HTTP defaults and error per-file when the service isn't running.

Cascade results (0.6+)

When auto-analyser's cascade passes fire (e.g. document → provenance / conversation / reflection), each cascade block is promoted to a first-class result row with the same shape as a direct route — so consumers resolve signal paths by member name unchanged. Promoted rows carry via: "cascade:<triggered_by>", marking the signal set as auto-detected: heuristics can misfire, and the human reader decides how much weight to give them. Cascade rows don't inflate analysed_files or the extension distribution.