classify

name	classify
description	Debug or inspect classifier output for specific files or paths. Shows which rule fired and why. Use when the user asks "why was this classified as X", "what kind is this file", "debug classification", or wants to understand classifier behavior.
model	haiku
effort	low
allowed-tools	["Bash","Read","Grep"]

classify

Inspect how the classifier categorizes a specific file or set of files.

Usage

Scan a single directory or repo and filter the output:

# Scan and grep for a specific file
cnbl scan <repo> 2>/dev/null | grep '"rel_path":"src/model.py"'

# Scan and show only a specific kind
cnbl scan <repo> 2>/dev/null | grep '"kind":"discard"'

# Scan with full report
cnbl scan <repo> --report 2>&1 >/dev/null

Classifier Rule Chain

The classifier checks rules top-to-bottom. First match wins:

Priority	Rule	Kind	Signal
1	is_test_harness	TestHarness	path
2	is_fixture	Discard	path
3	is_script	Script	lang
4	is_spec	Spec	lang
5	is_config	Config	lang
6	is_port	Port	content
7	is_adapter	Adapter	path
8	is_typescript	DomainLogic	lang
9	is_entrypoint	Entrypoint	path+lang
10	is_domain_logic	DomainLogic	content
11	is_glue	Glue	content
--	fallthrough	Discard	--

Debugging Tips

• If a file is unexpectedly Discard, check if it has a recognized extension (walker.rs lang_from_ext)
• If a TS file is DomainLogic when it should be Adapter, check if the path contains an adapter keyword (rule 7 runs before rule 8)
• If a Rust file with content is Discard, check if tree-sitter parsed it (zero top_level_kinds = parse failure or unsupported syntax)
• Path rules use substring matching -- "test" in any path component triggers TestHarness (rule 1)

Source

Classifier implementation: src/classifier.rs Conformance tests: tests/conformance_classifier.rs Spec: specs/classifier-hardening.md