From 99a1adcecee6c19005ed82253b14d64a9b34fd22 Mon Sep 17 00:00:00 2001
From: Simon Willison <swillison@gmail.com>
Date: Thu, 27 Feb 2025 07:35:28 -0800
Subject: [PATCH] Initial llm schemas list implementation, refs #781

---
 docs/help.md | 28 ++++++++++++++++++++++
 llm/cli.py   | 68 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 llm/utils.py | 42 ++++++++++++++++++++++++++++++++
 3 files changed, 138 insertions(+)

diff --git a/docs/help.md b/docs/help.md
index f266961..aa48f78 100644
--- a/docs/help.md
+++ b/docs/help.md
@@ -77,6 +77,7 @@ Commands:
   models        Manage available models
   openai        Commands for working directly with the OpenAI API
   plugins       List installed plugins
+  schemas       Manage stored schemas
   similar       Return top N similar IDs from a collection using cosine...
   templates     Manage stored prompt templates
   uninstall     Uninstall Python packages from the LLM environment
@@ -417,6 +418,33 @@ Options:
   --help  Show this message and exit.
 ```
 
+(help-schemas)=
+### llm schemas --help
+```
+Usage: llm schemas [OPTIONS] COMMAND [ARGS]...
+
+  Manage stored schemas
+
+Options:
+  --help  Show this message and exit.
+
+Commands:
+  list*  List stored schemas
+```
+
+(help-schemas-list)=
+#### llm schemas list --help
+```
+Usage: llm schemas list [OPTIONS]
+
+  List stored schemas
+
+Options:
+  -p, --path FILE   Path to log database
+  -q, --query TEXT  Search for schemas matching this string
+  --help            Show this message and exit.
+```
+
 (help-aliases)=
 ### llm aliases --help
 ```
diff --git a/llm/cli.py b/llm/cli.py
index 09eba95..b200ee3 100644
--- a/llm/cli.py
+++ b/llm/cli.py
@@ -45,6 +45,7 @@ from .utils import (
     make_schema_id,
     output_rows_as_json,
     resolve_schema_input,
+    schema_summary,
 )
 import base64
 import httpx
@@ -1373,6 +1374,73 @@ def templates_list():
             click.echo(display_truncated(text))
 
 
+@cli.group(
+    cls=DefaultGroup,
+    default="list",
+    default_if_no_args=True,
+)
+def schemas():
+    "Manage stored schemas"
+
+
+@schemas.command(name="list")
+@click.option(
+    "-p",
+    "--path",
+    type=click.Path(readable=True, exists=True, dir_okay=False),
+    help="Path to log database",
+)
+@click.option(
+    "queries",
+    "-q",
+    "--query",
+    multiple=True,
+    help="Search for schemas matching this string",
+)
+def schemas_list(path, queries):
+    "List stored schemas"
+    path = pathlib.Path(path or logs_db_path())
+    if not path.exists():
+        raise click.ClickException("No log database found at {}".format(path))
+    db = sqlite_utils.Database(path)
+    migrate(db)
+
+    params = []
+    where_sql = ""
+    if queries:
+        where_bits = ["schemas.content like ?" for _ in queries]
+        where_sql += " where {}".format(" and ".join(where_bits))
+        params.extend("%{}%".format(q) for q in queries)
+
+    sql = """
+    select
+      schemas.id,
+      schemas.content,
+      max(responses.datetime_utc) as recently_used,
+      count(*) as times_used
+    from schemas
+    join responses
+      on responses.schema_id = schemas.id
+    {} group by responses.schema_id
+    order by recently_used
+    """.format(
+        where_sql
+    )
+    rows = db.query(sql, params)
+    for row in rows:
+        click.echo("- id: {}".format(row["id"]))
+        click.echo(
+            "  summary: |\n    {}".format(schema_summary(json.loads(row["content"])))
+        )
+        click.echo(
+            "  usage: |\n    {} time{}, most recently {}".format(
+                row["times_used"],
+                "s" if row["times_used"] != 1 else "",
+                row["recently_used"],
+            )
+        )
+
+
 @cli.group(
     cls=DefaultGroup,
     default="list",
diff --git a/llm/utils.py b/llm/utils.py
index e9496ce..a7d4a2c 100644
--- a/llm/utils.py
+++ b/llm/utils.py
@@ -256,3 +256,45 @@ def resolve_schema_input(db, schema_input):
         return json.loads(row["content"])
     except (sqlite_utils.db.NotFoundError, ValueError):
         raise click.BadParameter("Invalid schema")
+
+
+def schema_summary(schema: dict) -> str:
+    """
+    Extract property names from a JSON schema and format them in a
+    concise way that highlights the array/object structure.
+
+    Args:
+        schema (dict): A JSON schema dictionary
+
+    Returns:
+        str: A human-friendly summary of the schema structure
+    """
+    if not schema or not isinstance(schema, dict):
+        return ""
+
+    schema_type = schema.get("type", "")
+
+    if schema_type == "object":
+        props = schema.get("properties", {})
+        prop_summaries = []
+
+        for name, prop_schema in props.items():
+            prop_type = prop_schema.get("type", "")
+
+            if prop_type == "array":
+                items = prop_schema.get("items", {})
+                items_summary = schema_summary(items)
+                prop_summaries.append(f"{name}: [{items_summary}]")
+            elif prop_type == "object":
+                nested_summary = schema_summary(prop_schema)
+                prop_summaries.append(f"{name}: {nested_summary}")
+            else:
+                prop_summaries.append(name)
+
+        return "{" + ", ".join(prop_summaries) + "}"
+
+    elif schema_type == "array":
+        items = schema.get("items", {})
+        return schema_summary(items)
+
+    return ""