hermes - 💡(How to fix) Fix [Bug]: kanban boards list leaks sqlite connections across boards in long-lived processes [2 pull requests]

Bug Description

I found a real sqlite fd leak in the kanban boards listing path.

hermes kanban boards list calls _board_task_counts() for each board, and that helper currently uses with kb.connect(board=slug) as conn:.

For sqlite3.Connection, that context-manager pattern does not close the connection. It only manages transaction scope. As a result, listing boards in a long-lived process can leave per-board sqlite handles open.

This gets worse as the number of boards grows, because the command touches each board DB to collect task counts.

What I expected instead was for the board-count helper to close each sqlite connection explicitly after reading the counts.

Steps to Reproduce

1. Start from current main
2. Use a single long-lived Python process with a temporary HERMES_HOME
3. Create a number of kanban boards
4. Call hermes_cli.kanban.run_slash("boards list")
5. Observe the process fd count before and after

Minimal repro I used:

import os, tempfile
os.environ["HERMES_HOME"] = tempfile.mkdtemp(prefix="hermes_boards_leak_")

from hermes_cli import kanban_db as kb
from hermes_cli.kanban import run_slash

conn = kb.connect()
conn.close()

for i in range(20):
    kb.create_board(f"b{i}")

base = len(os.listdir("/proc/self/fd"))
run_slash("boards list")
after = len(os.listdir("/proc/self/fd"))

print({"base": base, "after": after, "delta": after - base})


### Expected Behavior

Running `boards list` should not leave per-board sqlite handles open or grow the process fd count.

### Actual Behavior

A single `boards list` call can leave many sqlite file descriptors open.

In my repro on current main with 21 boards total, one `run_slash("boards list")` call changed the fd count from:

`{'base': 5, 'after': 68, 'delta': 63}`

The leaked descriptors were the per-board `kanban.db`, `kanban.db-wal`, and `kanban.db-shm` handles.

### Affected Component

CLI (interactive chat)

### Messaging Platform (if gateway-related)

N/A (CLI only)

### Debug Report

```shell
Report     https://paste.rs/dw6Bj
agent.log  https://paste.rs/DZQZ4

Operating System

Ubuntu on WSL2 (Windows 11)

Python Version

Python 3.12.3

Hermes Version

Hermes Agent v0.14.0 (2026.5.16)

Additional Logs / Traceback (optional)

This is a steady-state resource leak rather than an immediate traceback.

The impact grows with the number of boards because `boards list` opens each board DB to collect task counts. In a long-lived process, repeated use can push the process toward fd exhaustion.

Root Cause Analysis (optional)

The leak is in hermes_cli/kanban.py, inside _board_task_counts().

That helper uses with kb.connect(board=slug) as conn:. For sqlite connections, with conn: does not close the connection when the block exits; it only commits or rolls back the transaction.

Since boards list calls that helper once per board, sqlite handles accumulate across boards in long-lived processes.

Proposed Fix (optional)

Wrap the board-count read path in contextlib.closing(kb.connect(board=slug)) so each sqlite handle is closed explicitly after reading counts.

I also prepared regression coverage to make sure _board_task_counts() closes sqlite-like connections whose context manager does not call close().

Are you willing to submit a PR for this?

• I'd like to fix this myself and submit a PR