| name | extraction-proposer |
| description | Scan ICE-Crawler extraction logs, pick promising algorithms/tools, and emit skill creation proposals (name, scope, source files, next steps). |
Extraction Proposer
Use this skill after ICE-Crawler runs to convert harvested fossils into concrete skill proposals. It expects the extraction registry under ../ice-crawler-harvester/extractions (relative to this skill folder) and writes proposal specs under proposals/.
Reference: references/registry-workflow.md
Prerequisites
- ICE-Crawler run metadata appended to
extractions/index.jsonl.
- Per-repo notes in
extractions/<repo-slug>/SUMMARY.md (preferred) with code pointers.
- jq or Python available for filtering JSONL (optional but useful).
Workflow
-
Review registry
- Open
extractions/index.jsonl to find recent entries. Use jq or Python to filter by tags, repo, or notable files.
- Inspect corresponding
SUMMARY.md files for algorithm descriptions and code paths.
-
Select a candidate
- Criteria examples: unique algorithm, reusable CLI, monitoring utility, scaffolding snippet, etc.
- Note the run folder (
state/runs/<run_id>) and manifest path for provenance.
-
Extract details
- List the files from
artifact_manifest.json (or the trimmed subset copied into extractions/<slug>/).
- Summarize what the algorithm/tool does, triggers, dependencies, and why it deserves a skill.
-
Create a proposal
- Use the template below to write
proposals/<candidate>.json (create proposals/ if missing):
{
"ts": "2026-02-24T16:40:00Z",
"skill_name": "triadic-selector",
"description": "Deterministic triadic-balanced file selector for repository harvesting pipelines.",
"source_repo": "https://github.com/...",
"run_dir": "state/runs/run_20260224_113500",
"manifest": "state/runs/run_20260224_113500/artifact_manifest.json",
"notable_files": ["engine/glacier_selector.py", "docs/triadic_strategy.md"],
"summary_path": "extractions/triadic-selector/SUMMARY.md",
"proposed_skill_structure": {
"SKILL.md": ["workflow", "parameters", "safety"],
"references/triadic.md": ["derivation", "examples"],
"scripts/selector_demo.py": "optional CLI"
},
"next_actions": [
"Copy selector code into scripts/",
"Write SKILL.md instructions",
"Add references"
]
}
-
Hand off
- Once a proposal JSON is ready, use
skill-creator (or manual process) to implement the actual skill described.
- Update
extractions/<repo-slug>/SUMMARY.md with the proposal link so the registry stays synchronized.
Tips
- Keep proposals small and focused; one algorithm/tool per spec.
- Always cite the original run folder and manifest for traceability.
- If multiple skills can emerge from a single repo, create separate proposals referencing the same run.
- When a skill is built, link back to the proposal JSON for provenance.
This skill ensures every ICE-Crawler extraction can graduate into a reusable capability with clean provenance.