| name | ice-crawler-harvester |
| description | Run ICE-Crawler’s Frost→Glacier→Crystal pipeline to ingest repositories safely, emit bounded artifact bundles, and hand off sealed fossils for downstream agents. |
ICE-Crawler Harvester
Use this skill to run the ICE-Crawler pipeline wherever you have the project cloned. Set an environment variable such as ICE_CRAWLER_ROOT that points to your local clone of the Ice-Crawler repo and run the commands from there. The instructions below assume PowerShell, but any shell works.
Reference: references/ice-crawler-workflow.md
Prerequisites
- Python 3.10+ with Tkinter (for the UI) and Git on PATH.
- ICE-Crawler repository checked out locally (path referenced via
ICE_CRAWLER_ROOT).
- Optional:
agentics/ hooks if you need partitioned follow-up tasks.
Workflow 1 — Full UI Run (interactive)
cd $env:ICE_CRAWLER_ROOT- Launch the UI:
python icecrawler.py
- Paste any cloneable Git URL (browse/blob URLs are normalized automatically) and press the glowing PRESS TO SUBMIT TO ICE CRAWLER button.
- Watch the phase ladder (Frost → Glacier → Crystal → Residue) and log panels update in real time.
- When the run completes, open
state/runs/run_<timestamp>/ to inspect the fossilized artifact bundle.
UI Controls
- Ctrl+B toggle left ladder; Ctrl+Shift+B toggle right logs; Ctrl+J toggle terminal.
- Drag PanedWindow sashes to resize panels.
- UI never touches git; it mirrors
ui_events.jsonl written by the orchestrator.
Workflow 2 — Headless CLI Run
cd $env:ICE_CRAWLER_ROOT
$run = "state/runs/run_$(Get-Date -Format 'yyyyMMdd_HHmmss')"
$temp = "state/_temp_repo"
New-Item -ItemType Directory -Force -Path $run | Out-Null
python -m engine.orchestrator "https://github.com/openclaw/openclaw.git" $run 80 256 $temp
Arguments: <repo_url> <state_run_dir> <max_files> <max_kb> <temp_dir>
max_files controls the Glacier selection cap.max_kb enforces a per-file size ceiling when copying into artifact/.temp_dir is purged automatically; failure to delete triggers a residue violation.
Outputs & Follow-up
state/runs/<run>/artifact/ — crystallized file tree (repo-relative paths preserved).artifact_manifest.json + artifact_hashes.json — integrity anchors for downstream tools.ai_handoff/manifest_compact.json + root_seal.txt — sealed bundle for agent prompts.ui_events.jsonl, run_cmds.jsonl — truth logs for UI or automation.residue_truth.json — teardown attestation; treat violations as failures.- Extraction registry — append a row to
skills/ice-crawler-harvester/extractions/index.jsonl and drop notes under extractions/<repo-slug>/ so future skills can mine algorithms/tools (see extractions/README.md).
Extending / Integrating
- Call the orchestrator from scripts or scheduled jobs to keep repo fossils fresh.
- Parse
artifact_manifest.json to feed other skills (e.g., code summarizers, diff analyzers).
- Hook
agentics/ when you need automatic partitioning of Frost metadata or Crystal artifacts into bounded tasks.
- Adjust
max_files / max_kb per run to dial ingest size; keep limits conservative for safety.
Troubleshooting
- Missing Tkinter on Linux →
sudo apt install python3-tk (or distro equivalent).
- Git credential prompts bubble through the orchestrator; ensure SSH keys or tokens are configured.
- Residue violation (
state/_temp_repo not deleted) aborts the run; rerun after manual cleanup if needed.
Follow this skill to get deterministic repo fossils with ICE-Crawler’s provenance guarantees.
