athrael-soju/query

Schema:

• sessions (id TEXT PK, started_at TEXT, ended_at TEXT, duration_min REAL, title TEXT, agent_summary TEXT, exchange_count INT, summary_at TEXT, summary_msg_count INT, tags TEXT, summary_offered INT DEFAULT 0)
• messages (id INT PK AUTO, session_id TEXT FK, timestamp TEXT, role TEXT, content TEXT, metadata TEXT)
• topics (id INTEGER PK AUTO, title TEXT NOT NULL, domain TEXT NOT NULL, tags TEXT NOT NULL, created TEXT, updated TEXT)
• statements (id INTEGER PK AUTO, topic_id INTEGER FK→topics(id), claim TEXT NOT NULL, created TEXT, updated TEXT)
• tasks (id INTEGER PK AUTO, title TEXT NOT NULL, domain TEXT NOT NULL, status TEXT DEFAULT 'open', priority TEXT DEFAULT 'medium', horizon TEXT DEFAULT 'later', metadata TEXT, created TEXT, updated TEXT)
• updates (id INTEGER PK AUTO, task_id INTEGER FK→tasks(id), content TEXT NOT NULL, timestamp TEXT)

JSON metadata columns (query with json_extract(metadata, '$.field')):

• tasks.metadata — optional; {"source_session_id": "<sid>"} from add_task when a session id is known, else NULL
• messages.metadata — user: {"cwd", "permission_mode", "usage": {...}}; assistant: {"tool_calls": {<tool>: <count>}, "usage": {...}}; system: analysis telemetry

query.py runs your SQL as-is — no template, no required shape. Let the question scope the query: "today/now" → horizon='now', "overview/how many" → COUNT/GROUP BY, detail → narrow columns (substr(claim,1,160) for long text), sized to what's asked — add a LIMIT only when the result would otherwise be excessive, not by reflex. Table mode does not truncate — a result over ~16KB is refused with an error asking you to re-scope; pass --full only when you genuinely want every row.

Execute the SQL directly. Append --json for JSON output.

$PY "${CLAUDE_PLUGIN_ROOT}/scripts/query.py" "$ARGUMENTS"