agent-context-builder

Genera contesto operativo compatto per agenti DataCivicLab. ACB è il layer di contesto: ogni 6 ore scansiona i repo del Lab e produce artifact che dicono ad agenti e umani "cosa è successo e cosa serve attenzione".

Artifact

Artifact	Schema	Ruolo
session_bootstrap.md	—	Orientamento rapido (markdown)
workspace_triage.json	v1	Stato Lab: radar, PR, issues, discussions, registry, pipeline
topic_index.json	v5	Catalogo: 211 dataset (con columns, location), 18 analisi, explorer themes

Branch context:

https://raw.githubusercontent.com/dataciviclab/agent-context-builder/context/topic_index.json
https://raw.githubusercontent.com/dataciviclab/agent-context-builder/context/workspace_triage.json

Fonti consumate

Repo	Artifact	Cosa
tutti i repo config	registry/registry.json	Dataset (slug, columns, location, stage), signals, marts
source-observatory	data/radar/radar_summary.json	Radar 36 fonti (GREEN/YELLOW/RED)
source-observatory	data/catalog/catalog_signals.json	Drift inventariale
data-explorer	catalog/datasets.json + catalog/themes.json	Temi editoriali

Tool MCP

Esposti via agent-context-mcp (server MCP dataciviclab-context).

Tool	Quando usarlo
session_bootstrap()	Prima chiamata — orientamento rapido
workspace_triage(section=)	Stato precisi: radar, prs, issues, registry, pipeline
topic_index(resolve=)	Deep-dive su dataset/analisi per slug o fonte
search(query)	Ricerca cross-cutting: issues, PR, dataset, analisi
refresh_context()	Trigger rebuild CI

Esempi

# Orientamento
session_bootstrap()
# → markdown con radar, PR, issues, discussions

# Stato radar
workspace_triage(section="radar")
# → {"green": 35, "yellow": 0, "red": 1, ...}

# Deep-dive dataset
topic_index(resolve="ispra_ru_base")
# → {slug, name, source, period, stage, gcs_path, analyses: [...]

# Ricerca
search("rifiuti")
# → {issues: [...], datasets: [{slug, name, stage}], analyses: [...]}

Configurazione

{
  "mcpServers": {
    "dataciviclab-context": {
      "command": "agent-context-mcp",
      "env": { "GITHUB_TOKEN": "<opzionale>" }
    }
  }
}

Utilizzo locale

pip install -e ".[dev]"
agent-context build --config dataciviclab.config.yml --out generated/

Sviluppo

pip install -e ".[dev]"
pytest          # 141 test
ruff check src/ tests/
mypy src/ tests/

Licenza

MIT