feat: initial commit — Billo Release Agent (LangGraph)

LangGraph-based release automation agent with:
- PR discovery (webhook + polling)
- AI code review via Claude Code CLI (subscription-based)
- Auto-create Jira tickets for PRs without ticket ID
- Jira ticket lifecycle management (code review -> staging -> done)
- CI/CD pipeline trigger, polling, and approval gates
- Slack interactive messages with approval buttons
- Per-repo semantic versioning
- PostgreSQL persistence (threads, staging, releases)
- FastAPI API (webhooks, approvals, status, manual triggers)
- Docker Compose deployment

1069 tests, 95%+ coverage.

src/release_agent/services/pr_dedup.py

@@ -0,0 +1,46 @@
+"""PR deduplication service.
+
+Queries the agent_threads table to find which PRs from a given list have
+not yet been processed.  This prevents the PR poller from re-triggering
+graph runs for PRs that already have an existing thread.
+"""
+
+from release_agent.models.pr import PRInfo
+
+_QUERY = """
+    SELECT pr_id, repo_name
+    FROM agent_threads
+    WHERE (pr_id, repo_name) IN (
+        SELECT unnest(%s::text[]), unnest(%s::text[])
+    )
+"""
+
+
+async def find_unprocessed_prs(pool, prs: list[PRInfo]) -> list[PRInfo]:
+    """Return the subset of prs that have no existing agent_threads record.
+
+    A PR is considered already-processed if there exists a row in agent_threads
+    with matching (pr_id, repo_name) pair. The SQL uses unnest to enforce
+    pair-wise matching (not independent ANY on each column).
+
+    Args:
+        pool: Async psycopg connection pool.
+        prs: List of PRInfo objects to check.
+
+    Returns:
+        A new list containing only PRs with no existing thread.
+        Original list is not mutated.
+    """
+    if not prs:
+        return []
+
+    pr_ids = [p.pr_id for p in prs]
+    repo_names = [p.repo_name for p in prs]
+
+    async with pool.connection() as conn:
+        async with conn.cursor() as cur:
+            await cur.execute(_QUERY, (pr_ids, repo_names))
+            rows = await cur.fetchall()
+
+    processed = {(str(row[0]), str(row[1])) for row in rows}
+    return [p for p in prs if (p.pr_id, p.repo_name) not in processed]