1.0.0

[PalNilsson]

[1.0.0] — 2026-04-08

First stable release. All four agents (ingestion, cric, document-monitor,
github-doc-sync) are production-ready. This release focuses on correctness
under concurrent read/write access, operational observability, and release
tooling.

Fixed

Concurrency — DuckDB torn-read protection

All database write operations that involve multiple SQL statements are now
wrapped in explicit BEGIN / COMMIT / ROLLBACK transactions. Before this
fix, a query arriving from AskPanDA (via the Bamboo MCP tool) during a write
cycle could observe a missing table, an empty table, or a partially-inserted
result.

• cric_fetcher._load() — the full DROP TABLE → CREATE TABLE → INSERT
  sequence for queuedata is now a single atomic transaction. Concurrent
  readers always see either the previous complete snapshot or the new one.
• DuckDBStore.write_table(overwrite=True) — same fix applied to the
  generic overwrite path used by the ingestion agent's source history tables.
• BigPandaJobsFetcher._fetch_and_persist() — the three-table write
  (jobs upsert + selectionsummary replace + errors_by_count replace) for
  each queue is now a single transaction. A reader can never observe a state
  where some tables have been updated for a cycle and others have not.

Concurrency — ChromaDB staging swap

DocumentMonitorAgent._ingest_file() previously deleted old chunks from the
live ChromaDB collection and then inserted new ones, leaving a window where
the document was invisible to concurrent queries. The update path now uses an
atomic staging swap:

1. Write all new chunks into a temporary <name>__staging collection.
2. Delete the old chunks from the live collection.
3. Insert from staging into the live collection.
4. Drop the staging collection (in a finally block).

If the staging write fails the live collection is never touched. If the swap
step fails the old chunks remain visible.

Added

scripts/bump_version.py — release versioning script

A new script for bumping the version string across all relevant files in one
command:

python scripts/bump_version.py 0.1.0 1.0.0

Validates both version strings against PEP 440, reports each file updated, and
exits non-zero on any failure — safe to run in CI. After a successful bump the
script prints a reminder to reinstall the package so agents pick up the new
version at runtime:

IMPORTANT: reinstall the package so agents report the new version at runtime:
    pip install -e .

common/cli.py — shared startup banner

A new log_startup_banner(logger, prog) helper in
bamboo_mcp_services.common.cli emits a consistent startup line on every
agent launch:

bamboo-cric  version=1.0.0  python=3.12.3

The version is resolved at runtime from the installed package metadata
(importlib.metadata) so it always reflects the version in pyproject.toml
without requiring a hardcoded constant.

Previously three of the four CLIs each contained an inline 9-line copy of this
logic, and bamboo-document-monitor had no version logging at all. The helper
replaces all four with a single call.

Per-queue progress logging in BigPandaJobsFetcher

The ingestion agent now logs a progress line before fetching each queue,
showing the current position in the cycle:

BigPandaJobsFetcher: processing queue 'BNL' (10/230)

This makes it straightforward to monitor long cycles (e.g. 230 queues × 60 s
inter-queue delay ≈ 4 hours) and to identify which queue a failure or slowdown
is associated with.

New test coverage — 16 tests across 3 files

File	What is tested
tests/agents/cric_agent/test_cric_agent.py	Transaction safety: successful load gives complete table; failed load ROLLBACK preserves previous snapshot; table never absent after a write
tests/agents/ingestion_agent/test_bigpanda_jobs_fetcher.py	Transaction safety: all three tables updated atomically; failed mid-write ROLLBACK leaves baseline intact
tests/test_duckdb_store.py (new)	write_table append and overwrite modes; ROLLBACK on insert failure preserves previous data; table existence after a failed overwrite; record_snapshot round-trips

Changed

• bamboo-document-monitor CLI now uses log_startup_banner and therefore
  logs version and Python information on startup, matching the other three
  agents.
• DuckDBStore.write_table(overwrite=False) is unchanged — the append path
  involves no DROP and was already safe.

Notes for operators

AskPanDA / Bamboo MCP read connections — the transaction fixes above protect
against torn reads caused by the write process, but for full safety the MCP
query tool should open DuckDB files with read_only=True:

conn = duckdb.connect(database="cric.db", read_only=True)
conn = duckdb.connect(database="jobs.duckdb", read_only=True)

DuckDB enforces a single-writer policy at the file level. A second connection
opened with read_only=False (the default) while the agent holds the write
connection will either block or raise IOException: Database is already open.
read_only=True connections are explicitly allowed to coexist with one writer.