infospace-bench/src/infospace_bench/cli.py

380 lines
13 KiB
Python
Raw Normal View History

2026-05-14 11:32:25 +02:00
from __future__ import annotations
import argparse
2026-05-14 17:16:39 +02:00
from collections import Counter
2026-05-14 11:32:25 +02:00
import json
import sys
from pathlib import Path
2026-05-14 15:35:04 +02:00
from .checks import run_collection_checks
2026-05-14 16:26:42 +02:00
from .engine import engine_capability_contract, plan_asset_sync, sync_assets
2026-05-14 11:32:25 +02:00
from .errors import InfospaceError
from .evaluation_io import read_entity_evaluations
2026-05-14 17:16:39 +02:00
from .history import (
build_viability_report,
find_snapshot,
get_history,
metric_trend,
read_metrics_file,
record_check_results,
)
from .inspection import export_mermaid, relationship_summary
2026-05-14 11:32:25 +02:00
from .lifecycle import add_artifact, create_infospace, load_infospace
2026-05-14 14:53:16 +02:00
from .markdown_adapter import validate_infospace_artifacts
2026-05-14 15:06:17 +02:00
from .semantics import list_entities, list_relations
from .workflow import (
FixtureAssistedGenerationAdapter,
load_workflows,
plan_workflow,
run_workflow,
)
2026-05-14 11:32:25 +02:00
def build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(prog="infospace-bench")
sub = parser.add_subparsers(dest="command", required=True)
create = sub.add_parser("create", help="Create an infospace")
create.add_argument("workspace")
create.add_argument("slug")
create.add_argument("--name", required=True)
create.add_argument("--topic-domain", default="")
inspect = sub.add_parser("inspect", help="Inspect an infospace")
inspect.add_argument("root")
2026-05-14 17:16:39 +02:00
status = sub.add_parser("status", help="Summarize replacement readiness state")
status.add_argument("root")
2026-05-14 11:32:25 +02:00
add = sub.add_parser("add-artifact", help="Add an artifact to an infospace")
add.add_argument("root")
add.add_argument("source")
add.add_argument("--kind", required=True)
add.add_argument("--title", default="")
export = sub.add_parser("export", help="Print the infospace representation")
export.add_argument("root")
2026-05-14 14:53:16 +02:00
validate = sub.add_parser("validate", help="Validate infospace artifacts")
validate.add_argument("root")
2026-05-14 15:06:17 +02:00
entities = sub.add_parser("entities", help="List parsed entity artifacts")
entities.add_argument("root")
relations = sub.add_parser("relations", help="List parsed relation artifacts")
relations.add_argument("root")
2026-05-14 15:35:04 +02:00
history = sub.add_parser("history", help="List evaluation snapshot history")
history.add_argument("root")
history.add_argument("--metric", default="")
history_diff = sub.add_parser(
"history-diff",
help="Diff two evaluation snapshots by snapshot ID or date",
)
history_diff.add_argument("root")
history_diff.add_argument("before")
history_diff.add_argument("after")
metrics = sub.add_parser(
"metrics",
help="Run collection checks and persist metrics/history",
)
metrics.add_argument("root")
2026-05-14 17:16:39 +02:00
check = sub.add_parser(
"check",
help="Run collection checks and persist metrics/history",
)
check.add_argument("root")
viability = sub.add_parser("viability", help="Evaluate configured thresholds")
viability.add_argument("root")
graph = sub.add_parser("graph", help="Export the artifact relationship graph")
graph.add_argument("root")
graph.add_argument("--format", choices=["json", "mermaid"], default="json")
workflow = sub.add_parser("workflow", help="Inspect, plan, and run workflows")
workflow_sub = workflow.add_subparsers(dest="workflow_command", required=True)
workflow_inspect = workflow_sub.add_parser(
"inspect",
help="Inspect workflow declarations",
)
workflow_inspect.add_argument("root")
workflow_plan = workflow_sub.add_parser(
"plan",
help="Plan a workflow without writing outputs",
)
workflow_plan.add_argument("root")
workflow_plan.add_argument("workflow_id")
workflow_run = workflow_sub.add_parser(
"run",
help="Run a deterministic workflow",
)
workflow_run.add_argument("root")
workflow_run.add_argument("workflow_id")
workflow_run.add_argument(
"--fixture-responses",
default="",
help="Run assisted stages with deterministic fixture responses",
)
2026-05-14 16:26:42 +02:00
engine = sub.add_parser("engine", help="Inspect and sync engine boundary state")
engine_sub = engine.add_subparsers(dest="engine_command", required=True)
engine_inspect = engine_sub.add_parser(
"inspect",
help="Inspect the optional engine capability contract",
)
engine_inspect.add_argument("root")
engine_plan = engine_sub.add_parser(
"plan-sync",
help="Plan artifact-to-asset sync without mutation",
)
engine_plan.add_argument("root")
engine_sync = engine_sub.add_parser(
"sync",
help="Dry-run artifact-to-asset sync unless --apply is passed",
)
engine_sync.add_argument("root")
engine_sync.add_argument("--apply", action="store_true")
2026-05-14 11:32:25 +02:00
return parser
def main(argv: list[str] | None = None) -> int:
parser = build_parser()
args = parser.parse_args(argv)
try:
if args.command == "create":
infospace = create_infospace(
Path(args.workspace),
args.slug,
name=args.name,
topic_domain=args.topic_domain,
)
_write_json({"slug": infospace.config.slug, "root": str(infospace.root)})
elif args.command == "inspect":
_write_json(load_infospace(Path(args.root)).to_dict())
2026-05-14 17:16:39 +02:00
elif args.command == "status":
_write_json(_status_payload(Path(args.root)))
2026-05-14 11:32:25 +02:00
elif args.command == "add-artifact":
artifact = add_artifact(
Path(args.root),
Path(args.source),
kind=args.kind,
title=args.title,
)
_write_json({"artifact": artifact.to_dict()})
elif args.command == "export":
_write_json(load_infospace(Path(args.root)).to_dict())
2026-05-14 14:53:16 +02:00
elif args.command == "validate":
results = validate_infospace_artifacts(Path(args.root))
valid = all(result.valid for result in results)
_write_json(
{
"valid": valid,
"results": [result.to_dict() for result in results],
}
)
return 0 if valid else 1
2026-05-14 15:06:17 +02:00
elif args.command == "entities":
_write_json(
{
"entities": [
entity.to_dict() for entity in list_entities(Path(args.root))
]
}
)
elif args.command == "relations":
_write_json(
{
"relations": [
relation.to_dict()
for relation in list_relations(Path(args.root))
]
}
)
2026-05-14 15:35:04 +02:00
elif args.command == "history":
history = get_history(Path(args.root))
if args.metric:
_write_json(
{
"metric": args.metric,
"trend": metric_trend(history, args.metric),
}
)
else:
_write_json({"history": [item.to_dict() for item in history]})
elif args.command == "history-diff":
history = get_history(Path(args.root))
before = find_snapshot(history, args.before)
after = find_snapshot(history, args.after)
if before is None or after is None:
missing = []
if before is None:
missing.append(args.before)
if after is None:
missing.append(args.after)
raise InfospaceError(
"missing_snapshot",
"Could not resolve requested snapshot reference",
{"missing_refs": missing},
)
_write_json({"diff": before.diff(after).to_dict()})
elif args.command == "metrics":
2026-05-14 17:16:39 +02:00
_write_json(_record_checks(Path(args.root)).to_dict())
elif args.command == "check":
2026-05-14 15:35:04 +02:00
infospace = load_infospace(Path(args.root))
2026-05-14 17:16:39 +02:00
report = run_collection_checks(infospace.artifacts)
result = record_check_results(
infospace.root,
report,
artifact_evaluations=_read_output_evaluations(infospace.root),
)
2026-05-14 17:16:39 +02:00
_write_json(
{
**result.to_dict(),
"details": report.details,
}
2026-05-14 15:35:04 +02:00
)
2026-05-14 17:16:39 +02:00
elif args.command == "viability":
_write_json(build_viability_report(Path(args.root)))
elif args.command == "graph":
infospace = load_infospace(Path(args.root))
summary = relationship_summary(infospace.artifacts)
if args.format == "mermaid":
print(export_mermaid(summary), end="")
else:
_write_json(_relationship_summary_payload(summary))
elif args.command == "workflow":
if args.workflow_command == "inspect":
_write_json(
{
"workflows": [
workflow.to_dict()
for workflow in load_workflows(Path(args.root))
]
}
)
elif args.workflow_command == "plan":
_write_json(
plan_workflow(Path(args.root), args.workflow_id).to_dict()
)
elif args.workflow_command == "run":
adapter = (
FixtureAssistedGenerationAdapter.from_file(
Path(args.fixture_responses)
)
if args.fixture_responses
else None
)
_write_json(
run_workflow(
Path(args.root),
args.workflow_id,
assisted_adapter=adapter,
).to_dict()
)
else:
parser.error(f"Unhandled workflow command: {args.workflow_command}")
2026-05-14 16:26:42 +02:00
elif args.command == "engine":
if args.engine_command == "inspect":
_write_json(
{
"root": str(Path(args.root)),
"contract": engine_capability_contract().to_dict(),
}
)
elif args.engine_command == "plan-sync":
_write_json(plan_asset_sync(Path(args.root)).to_dict())
elif args.engine_command == "sync":
_write_json(
sync_assets(Path(args.root), dry_run=not args.apply).to_dict()
)
else:
parser.error(f"Unhandled engine command: {args.engine_command}")
2026-05-14 11:32:25 +02:00
else:
parser.error(f"Unhandled command: {args.command}")
except InfospaceError as exc:
print(json.dumps(exc.to_dict(), indent=2), file=sys.stderr)
return 2
return 0
2026-05-14 17:16:39 +02:00
def _status_payload(root: Path) -> dict:
infospace = load_infospace(root)
validation = validate_infospace_artifacts(infospace.root)
entities = list_entities(infospace.root)
relations = list_relations(infospace.root)
history = get_history(infospace.root)
metrics = read_metrics_file(
infospace.root / "output" / "metrics" / "metrics.yaml"
)
viability = (
build_viability_report(infospace.root) if infospace.config.viability else None
)
graph = relationship_summary(infospace.artifacts)
return {
"slug": infospace.config.slug,
"name": infospace.config.name,
"topic": infospace.config.topic.to_dict(),
"artifact_count": len(infospace.artifacts),
"artifact_kinds": dict(
sorted(Counter(item.kind for item in infospace.artifacts).items())
),
"entity_count": len(entities),
"relation_count": len(relations),
"validation": {
"checked_artifact_count": len(validation),
"valid": all(result.valid for result in validation),
},
"history": {
"snapshot_count": len(history),
"latest_snapshot_id": history[-1].snapshot_id if history else "",
},
"metrics": metrics,
"viability": viability,
"graph": {
"node_count": graph.node_count,
"edge_count": graph.edge_count,
"relationship_types": graph.relationship_types,
},
}
def _record_checks(root: Path):
infospace = load_infospace(root)
return record_check_results(
infospace.root,
run_collection_checks(infospace.artifacts),
artifact_evaluations=_read_output_evaluations(infospace.root),
2026-05-14 17:16:39 +02:00
)
def _read_output_evaluations(root: Path):
return read_entity_evaluations(root / "output" / "evaluations")
2026-05-14 17:16:39 +02:00
def _relationship_summary_payload(summary) -> dict:
return {
"node_count": summary.node_count,
"edge_count": summary.edge_count,
"nodes": summary.nodes,
"edges": [
{"source": edge.source, "target": edge.target, "type": edge.type}
for edge in summary.edges
],
"relationship_types": summary.relationship_types,
}
2026-05-14 11:32:25 +02:00
def _write_json(payload: dict) -> None:
print(json.dumps(payload, indent=2))