Personal Claude Code plugin marketplace. Marketplace name: svd-agent-skills.
# Add the marketplace (once)
/plugin marketplace add svd/agent-skills
# Install individual plugins
/plugin install manage-claude-projects@svd-agent-skills
/plugin install session-analyzer@svd-agent-skills
/plugin install mcp-client-kit@svd-agent-skills
/plugin install session-handoff@svd-agent-skills
/plugin install mgraphctl@svd-agent-skillsInventory, deep-stat, and safely clean up the projects Claude Code tracks under ~/.claude.
Skills: /manage-claude-projects:manage-claude-projects
Trigger phrases: "list my claude projects", "how much have I spent on project X", "token usage for this project", "clean up old claude projects", "delete this project from claude".
Warning: The
removesubcommand is destructive. It deletes session transcripts, rewrites~/.claude.json, and removes todos/file-history entries. Always use--backup(the skill enforces this). Avoid while other Claude Code sessions are running —configremoval rewrites~/.claude.jsonand can race live session writes.
Parse a Claude Code session JSONL transcript into a structured Markdown report covering tool calls, skill vs. LLM attribution, errors and recovery, token usage, and optimization recommendations.
Skills: /session-analyzer:session-analyzer
Trigger phrases: "analyze this session", "session report", "how many tokens did it use", "what errors occurred", "show me the tool calls", or any session UUID / path.
Generate typed Python wrappers for any MCP server — one async def per tool with
real return types, so you call tools from code instead of pushing JSON schemas
through the model's context window. Ships a CLI (mcpgen) plus skills that drive it.
Skills: /mcp-client-kit:generate-mcp-wrappers, /mcp-client-kit:generate-mcp-runner
Trigger phrases: "generate MCP wrappers", "typed wrappers for an MCP server", "mcpgen", "smoke-test run.py for MCP wrappers".
Write and resume structured context handoffs between Claude sessions — new chat, new
machine, next day — without losing state, decisions, or mistakes already found. Handoffs
use trust tags ([V] verified / [?] unverified memory / [S] stale-risk), track a
parent/chain across handoffs for the same project, and a stdlib-only helper script
(handoff_tool.py) creates/validates/inspects them and flags staleness against the repo's
git state.
Skills: /session-handoff:session-handoff
Trigger phrases: "handoff", "hand off this session", "continue this in a new chat", "context is getting long", "park this project", "resume where we left off".
Work with Microsoft 365 through the Microsoft Graph API — Outlook mail and calendar, Teams
chats, channels and presence, meetings with transcripts and Copilot insights, SharePoint,
OneDrive, OneNote, Planner, To Do, people and org chart. Backed by a Python CLI (mgraphctl)
run with uv; every write verb supports --dry-run, and auth uses MSAL with a token cache in
the OS keychain.
Skills: /mgraphctl:mgraphctl
Trigger phrases: "check my inbox", "what meetings do I have", "send a Teams message", "find that SharePoint doc", "summarize this meeting transcript", "my tasks".
Both plugins use the same pricing table (USD per 1M tokens), matched by substring on the model id. Cache write = 1.25× input, cache read = 0.1× input (0.025× on Fable/Mythos 5.1):
| Model | Input | Output | Cache write | Cache read |
|---|---|---|---|---|
| fable-5-1 / mythos-5-1 | 10.00 | 50.00 | 12.50 | 0.25 |
| fable / mythos | 10.00 | 50.00 | 12.50 | 1.00 |
| opus | 5.00 | 25.00 | 6.25 | 0.50 |
| sonnet-5 | 2.00 | 10.00 | 2.50 | 0.20 |
| sonnet | 3.00 | 15.00 | 3.75 | 0.30 |
| haiku | 1.00 | 5.00 | 1.25 | 0.10 |
Covers Fable 5/5.1, Mythos 5/5.1, Opus 4.5–4.8, Sonnet 5, Sonnet 4.5/4.6, Haiku 4.5 (opus =
current Opus 4.5+ rate). Row order matters — the first key that is a substring of the model id
wins, so sonnet-5 must precede sonnet and fable-5-1 must precede fable. Cost figures are estimates from token counts, not actual
billing amounts.
Versions use commit-SHA tracking — every push to main ships the latest code to
users who run /plugin marketplace update svd-agent-skills.
Both scripts are pure Python 3 stdlib — no pip install required.