Archive Grounding

Archive Grounding

This skill handles ZIP archives as container inputs in the any-input -> grounding -> downstream research / summary / report pipeline.

Its job is not to directly pretend that the whole archive has already been deeply grounded after unpacking. Its job is to:

1. unpack the archive,
2. build a stable archive bundle,
3. identify supported child files,
4. run the corresponding child skill for each supported child file,
5. collect those child bundles under the current archive bundle's child_outputs/ directory,
6. and only then write a real archive-level grounded.md.

Position in the pipeline

This skill is for archives that package together multiple materials, such as:

• project material bundles
• meeting material bundles
• report + slides + tables + audio/video attachments
• mixed research input packages

This skill is not a replacement for the child grounding skills themselves.

• document-grounding still handles document content
• table-grounding still handles spreadsheets / CSV tables
• pptx-grounding still handles PowerPoint decks
• meeting-audio-grounding handles meeting audio inputs
• meeting-video-grounding handles meeting videos
• audio_structuring remains the atomic audio transcription backend reused by the higher-level audio/video meeting entry skills

This skill acts as an orchestrator over packaged files.

Supported child file routing

The first version should route supported child files using a simple extension-based mapping:

• .pdf, .docx, .md, .txt -> document-grounding
• .xlsx, .csv -> table-grounding
• .pptx -> pptx-grounding
• .mp3, .wav, .m4a -> meeting-audio-grounding
• .mp4, .mov, .mkv -> meeting-video-grounding

Unsupported files may be recorded and skipped, but they must not be silently treated as grounded.

Required Workflow

When using this skill, you must follow this workflow:

1. First run the existing script entrypoint:

   bash .cursor/skills/archive-grounding/scripts/run.sh <input_zip> <output_root>

2. The script generates an archive bundle skeleton and inventory files such as:

   • extracted.md
   • extracted_meta.json
   • manifest.json
   • routed_items.json
   • unpacked/
   • child_outputs/

3. After the archive bundle is generated, read:

   • extracted.md
   • extracted_meta.json
   • manifest.json
   • routed_items.json

4. Then enumerate the supported child files in the archive.

5. For each supported child file, you must actually run the corresponding child skill. Do not stop after inventory / route planning.

6. For child files inside the archive, all downstream child grounding outputs must be written inside the current archive bundle's child_outputs/ directory, not into the global grounding root.

7. Use the recommended child output path recorded in routed_items.json when present.

8. A child file is only considered completed if one of the following is true:

   • its corresponding child bundle was actually generated under child_outputs/, or
   • a concrete failure reason is explicitly recorded.

9. Do not pretend that a child file was fully grounded merely because it was detected, inventoried, or assigned a route.

10. Only after the supported child files have been processed (or explicit failures have been recorded) may you write the archive-level grounded.md.

11. Do not create a placeholder grounded.md.

12. The task is not complete if:

• only the archive bundle skeleton exists,
• manifest.json / routed_items.json exist but no child skills were actually run,
• child bundles were written to the global grounding root instead of this archive bundle's child_outputs/, or
• archive-level grounded.md was written before the child processing step was completed.

Input

bash .cursor/skills/archive-grounding/scripts/run.sh <input_zip> <output_root>

Arguments

• input_zip: path to a .zip archive
• output_root: parent directory under which the archive bundle should be created

Example:

bash .cursor/skills/archive-grounding/scripts/run.sh \
  /path/to/materials.zip \
  data/grounded_notes

This will create something like:

/data/grounded_notes/archive-materials/

Output bundle contract

The script should create:

<archive_bundle>/
├─ extracted.md
├─ extracted_meta.json
├─ manifest.json
├─ routed_items.json
├─ unpacked/
├─ child_outputs/
└─ grounded.md   # written later by the agent, not by the script

File roles

extracted.md

Human-readable archive overview for the agent.

It should include:

• archive overview
• detected file inventory summary
• supported vs skipped items
• routed child skills
• the rule that child outputs must live under child_outputs/

extracted_meta.json

Global archive metadata, such as:

• source archive path
• archive id
• total file count
• supported file count
• skipped file count
• unpack directory
• generation timestamp