feat: Add postgres-mcp-usage skill and plugin manifest

[Dev-iL]

Adds a Claude Code skill teaching optimal MCP tool sequencing (schema discovery → analysis → optimization) with per-tool prerequisites and parameters. Includes .claude-plugin/plugin.json for marketplace distribution.

Depends on #158

[Dev-iL]

@jssmith this one too

[jssmith]

Review — concise

Verdict: Approve with minor comments. Well-structured skill documentation and plugin manifest.

Strengths

• Skill doc (postgres-mcp-usage.md) is comprehensive: covers all 9 tools, access modes, recommended workflows, extension prerequisites.
• Plugin manifest (plugin.json) is correctly structured with appropriate metadata.
• Clear separation: skill doc covers tool usage, AGENTS.md covers contribution.

Minor (non-blocking)

1. Hardcoded version in skill doc
"Last verified": "2026-03-19 against postgres-mcp v0.3.0" — this will go stale. Consider referencing __version__ or noting "verify against server.py if tool count differs". The doc already says this, so it's fine as-is.

2. method="llm" documented as experimental
The skill correctly notes LLM optimization is experimental and excluded. Good scoping.

3. Plugin version 0.3.0 hardcoded
Should track the package version from pyproject.toml. Consider syncing or noting they should match.

This review was created by an AI agent (OpenHands) on behalf of @jssmith.

[jssmith]

Review — concise

Verdict: Approve with one fixable issue. The skill doc is accurate against the current codebase — I verified all 9 tools, parameters, access modes, timeouts, and extension prerequisites against server.py. The version-sync test (test_plugin_metadata.py) is a nice addition that resolves the hardcoded-version concern from the prior review.

One issue

Broken file references: The skill says "see AGENTS.md and CLAUDE.md" but neither file exists in the repository. This will confuse users who follow the pointer. Either add those files or remove the reference.

Everything else checks out:

• Tool count (9), names, and parameter defaults match server.py
• get_top_queries limit correctly documented as applying only to mean_time/total_time
• explain_query analyze + hypothetical_indexes mutual exclusion confirmed
• 30-second timeout in RESTRICTED mode confirmed
• analyze_query_indexes max 10 queries confirmed (MAX_NUM_INDEX_TUNING_QUERIES)
• Health check types match HealthType enum
• Extension prerequisite table verified against code
• plugin.json version matches pyproject.toml and is enforced by test

This review was created by an AI agent (OpenHands) on behalf of @jssmith.

[Dev-iL]

@openhands-agent AGENTS.md and CLAUDE.md are added in #158.