repo-ingestor

name	repo-ingestor
description	Ingest external repositories, research papers, and benchmarks using 40-agent swarms to analyze, adapt features, and improve current workflows.

Repo Ingestor

This skill orchestrates the ingestion of external repositories, frameworks, and benchmarks to iteratively evolve and improve our own workflows and feature sets.

Core Workflows

1. Adaptive Learning Cycles (General Ingestion)

When analyzing external tools (e.g., theonlyhennygod/zeroclaw, alibaba/zvec, etc.):

• Perform a fulsome analysis of the tool.
• Adapt all features we lack, while maintaining and integrating our own existing identified features.
• Ensure our core identity remains, but consume and absorb functional improvements.
• Research Scope: Features, libraries, best use cases, security issues, permanence, runtime security, iterative improvements, and adaptation ability.
• Analysis Report: Do a comprehensive analysis of their feature set and vision versus ours. Identify which features and functions are best and need to be adapted for our improvement.

2. Benchmarks and LLM Test Repos

When incorporating testing repos (e.g., Snowflake-Labs/agent-world-model):

• Iteratively improve or aim for improvement on future repo ingestions.
• Do not overfit just to pass the tests. Philosophically understand the conditional improvements that come from future analytical perspective improvements.
• Investigate: Loop and escape paradigms, confrontational analysis schemes, and benchmark improvements over time.

3. Deep Swarm Research

• Use multiple agents to analyze multiple new repos and research papers.
• Absorb knowledge and functionality to enhance our own ecosystem.

4. Durable Ingestion Tracking

• When a repo-ingestion or remediation effort needs durable tracking, create or update a frontmatter-backed artifact from the repository-root-relative template docs/research/templates/INGESTION_TASK_TEMPLATE.md.
• Keep the artifact minimal: stable task_id, owner, status, acceptance_criteria, and evidence, plus repository-root-relative links back to the active session planning files at task_plan.md, findings.md, and progress.md.
• Do not turn the artifact into a second queue or a ticket system; the active session planning files remain the source of truth for session sequencing.