2976 2026 02 20 Duckdb Lock Scope Flag 18db1838

DuckDB Lock Fix, --scope Flag, .gitignore, and Consolidation Todo

For Claude: REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.

Goal: Fix the DuckDB write-lock conflict, add --scope flag to sentiment-score.py for lesson output targeting, create a .gitignore for local-scope output, and add a BACKLOG todo for MCP consolidation analysis.

Architecture: Three independent changes to the agentskill-kaizen plugin: (1) one-line plugin.json patch to make the motherduck MCP server read-only, freeing the write lock for the script; (2) a new --scope Typer option that routes lesson output to one of three destinations; (3) a .gitignore in the plugin root and a BACKLOG todo entry. No new dependencies. No schema changes.

Tech Stack: Python 3.11+, Typer, DuckDB, uv/PEP 723, JSON (plugin.json)

Task 1: Fix DuckDB lock — make MCP server read-only

Files:

• Modify: plugins/agentskill-kaizen/.claude-plugin/plugin.json line 32

Context: mcp-server-motherduck is launched with --read-write, which acquires DuckDB's exclusive write lock. When sentiment-score.py runs while the MCP server is active, it fails with IOException: Could not set lock. DuckDB allows multiple readers but only one writer. Making the MCP server read-only releases the write lock so the script can acquire it.

Step 1: Apply the one-line patch

In plugins/agentskill-kaizen/.claude-plugin/plugin.json, change:

"args": ["mcp-server-motherduck", "--db-path", "${HOME}/.claude/kaizen/kaizen.duckdb", "--read-write"]

to:

"args": ["mcp-server-motherduck", "--db-path", "${HOME}/.claude/kaizen/kaizen.duckdb", "--read-only"]

Step 2: Verify the JSON is still valid

python3 -c "import json, sys; json.load(open('plugins/agentskill-kaizen/.claude-plugin/plugin.json')); print('OK')"

Expected: OK

Step 3: Commit

git add plugins/agentskill-kaizen/.claude-plugin/plugin.json

Then /commit-staged

Task 2: Add --scope flag to sentiment-score.py

Files:

• Modify: plugins/agentskill-kaizen/scripts/sentiment-score.py

Context: The --scope flag controls where lesson output (future feature) is written. Three destinations:

For now, the flag is plumbing only — it validates, resolves, and prints the resolved destination path to stderr. The actual lesson extraction logic is a separate future task. Adding the flag now means callers can pass --scope today and it will be wired up when lesson extraction lands.

Step 1: Add ScopeTarget enum after the _DEFAULT_DB constant (around line 392)

import enum  # add to stdlib imports at top of file

class ScopeTarget(str, enum.Enum):
    """Output scope for lesson files."""
    user = "user"
    project = "project"
    local = "local"

Note: str, enum.Enum is required for Typer to accept enum values as CLI arguments.

Step 2: Add _resolve_scope_path helper after ScopeTarget

def _resolve_scope_path(scope: ScopeTarget) -> Path:
    """Return the resolved output path for *scope*.

    Args:
        scope: The target scope for lesson output.

    Returns:
        Absolute path to the lesson file or directory.
    """
    if scope == ScopeTarget.user:
        return Path("~/.claude/kaizen/lessons/").expanduser()
    if scope == ScopeTarget.project:
        return Path.cwd() / ".claude" / "kaizen" / "lessons.md"
    # local
    return Path.cwd() / ".claude" / "kaizen" / "lessons.local.md"

Step 3: Add scope parameter to the score() CLI function

In the score() function signature, add after the db parameter:

scope: Annotated[
    ScopeTarget | None,
    typer.Option(
        "--scope",
        help=(
            "Lesson output scope: 'user' → ~/.claude/kaizen/lessons/, "
            "'project' → {cwd}/.claude/kaizen/lessons.md (git-tracked), "
            "'local' → {cwd}/.claude/kaizen/lessons.local.md (gitignored)."
        ),
        rich_help_panel="Output Options",
    ),
] = None,

Step 4: Add scope path resolution and reporting in the score() function body

After the _write_duckdb call (around line 617), add:

if scope is not None:
    scope_path = _resolve_scope_path(scope)
    stderr.print(
        f"[dim]Lesson scope:[/dim] [cyan]{scope.value}[/cyan] → {scope_path}"
    )

Step 5: Verify the CLI help renders correctly

uv run --script plugins/agentskill-kaizen/scripts/sentiment-score.py score --help

Expected: --scope appears under "Output Options" with the three scope descriptions.

Step 6: Verify the flag is accepted without error

uv run --script plugins/agentskill-kaizen/scripts/sentiment-score.py score --scope local --help

Expected: exits 0, no error about unknown option.

Step 7: Commit

git add plugins/agentskill-kaizen/scripts/sentiment-score.py

Then /commit-staged

Task 3: Add .gitignore for local scope output

Files:

• Create: plugins/agentskill-kaizen/.gitignore

Context: The local scope writes {cwd}/.claude/kaizen/lessons.local.md — a per-project file that should never be committed. A .gitignore in the plugin root handles this for the plugin's own working directory.

Step 1: Create the .gitignore

Create plugins/agentskill-kaizen/.gitignore with content:

# Local-scope lesson output (gitignored by design — use --scope project for tracked lessons)
.claude/kaizen/lessons.local.md

Step 2: Verify git sees it as ignored

touch plugins/agentskill-kaizen/.claude/kaizen/lessons.local.md 2>/dev/null || true
git check-ignore -v plugins/agentskill-kaizen/.claude/kaizen/lessons.local.md

Expected: output references .gitignore and the file pattern.

Step 3: Clean up the test file

rm -f plugins/agentskill-kaizen/.claude/kaizen/lessons.local.md

Step 4: Commit

git add plugins/agentskill-kaizen/.gitignore

Then /commit-staged

Task 4: Add BACKLOG todo for MCP consolidation analysis

Files:

• Modify: .claude/BACKLOG.md (repo root of claude_skills)

Context: The plugin runs two MCP servers: kaizen-duckdb (motherduck, SQL queries) and kaizen-analysis (custom server.py, PM4Py tools + dashboard). The question is whether sentiment-score.py as a standalone script should instead be an MCP tool in server.py, and whether the two servers can be consolidated. This needs investigation before any architectural change.

Step 1: Add a P2 BACKLOG entry

In .claude/BACKLOG.md, add the following entry to the ## P2 - Nice to Have section:

### kaizen: MCP consolidation analysis

**Source**: Design session 2026-02-20
**Added**: 2026-02-20
**Description**: The plugin currently runs two MCP servers (`kaizen-duckdb` via mcp-server-motherduck, `kaizen-analysis` via server.py) plus a standalone CLI script (`sentiment-score.py`). Investigate: (1) What does each MCP server provide that the other cannot? Can they be merged into a single server? (2) Why is `sentiment-score.py` a standalone script rather than an MCP tool inside `server.py`? What would be gained or lost by moving scoring into the MCP server (always-on scoring, no manual invocation, lock ownership)? (3) Is there a clean boundary between "batch processing" (script) and "query/serve" (MCP) that should be preserved?
**Decision needed**: Consolidate vs. keep separate, with rationale.
**Suggested location**: `plugins/agentskill-kaizen/`

Step 2: Update the frontmatter counts

In .claude/BACKLOG.md frontmatter, increment p2-count by 1.

Step 3: Commit

git add .claude/BACKLOG.md

Then /commit-staged

Done

After all four tasks, verify:

# MCP config is valid JSON
python3 -c "import json; json.load(open('plugins/agentskill-kaizen/.claude-plugin/plugin.json')); print('plugin.json OK')"

# Script help shows --scope
uv run --script plugins/agentskill-kaizen/scripts/sentiment-score.py score --help | grep scope

# .gitignore exists
cat plugins/agentskill-kaizen/.gitignore

# BACKLOG has the new entry
grep "MCP consolidation" .claude/BACKLOG.md