| name | elke-intake-noise-triage |
| description | Use when: users provide a fully noisy unsorted evidence folder and you must inventory, type-triage, and de-duplicate files before extraction. |
ELKE Intake Noise Triage
Purpose
Transform raw unsorted input into a stable internal inventory.
System Prompt
skills/system-prompts/intake-subagent-system.md
Inputs
input_root
artifacts_root
Required Actions
- Recursively inventory all files.
- Record extension, size, and hash.
- Infer coarse document type from filename/path/content hints.
- Detect duplicate groups by hash and canonicalized filename.
- Preserve original files unchanged.
- Analyze files one-by-one (never all at once in one context payload).
- Do not emit progress/state trace files in final output.
Input Filtering Guard
Exclude known control/meta files from inventory and reconciliation counts:
_copy_manifest.json
README.md files used only as folder markers
- hidden/system files generated by OS or tooling
If excluded files are found, list them in a non-blocking note.
Required Outputs
Write to artifacts_root/02_structured/:
inventory_manifest.json
triage_mapping.json
duplicates.json
And copy raw files into:
artifacts_root/00_raw_input/
Handoff
Downstream consumer: elke-doc-extraction-normalization.
The extraction skill assumes the three structured intake artifacts exist.
Output
- inventory manifest
- duplicate groups
- triage mapping (file -> inferred type)