azmym/used

Usage Tracker

Show which skills, hooks, rules, plugins, and MCP tools were activated in the current conversation.

Arguments

• No arguments (default): report usage for the last user prompt and Claude response only
• session or --session: report usage for the entire session (since last /clear or session start)

Parse the arguments from: $ARGUMENTS (accept both session and --session as the session flag)

Instructions

When this skill is invoked, scan the current conversation context and report what was used. Follow these detection rules exactly.

1. Detect Skills

Scan the conversation for:

• <command-name> tags: extract the value (e.g., superpowers:brainstorming from <command-name>/superpowers:brainstorming</command-name> or <command-name>superpowers:brainstorming</command-name>)
• Skill tool invocations in the conversation flow (where Claude called the Skill tool with a skill name)

Collect all unique skill names found.

2. Detect Hooks

Scan <system-reminder> tags for these patterns:

• Text containing "hook additional context" (e.g., "SessionStart hook additional context", "UserPromptSubmit hook additional context")
• Extract the hook name from the text before "hook additional context"

Collect all unique hook names found. Normalize to lowercase-kebab-case (e.g., "SessionStart" becomes "session-start", "UserPromptSubmit" becomes "user-prompt-submit").

3. Detect Rules

Scan the conversation for <system-reminder> blocks that contain # claudeMd sections. Within those sections, look for lines matching:

• Contents of .../.claude/rules/<name>.md - extract <name> as the rule name
• References to project-level CLAUDE.md files - report as "CLAUDE.md (project)" or similar

Collect all unique rule names found.

4. Detect Plugins

Scan the conversation for:

• <system-reminder> blocks containing "# MCP Server Instructions" followed by ## <plugin-name> headings - extract the plugin name
• <available-deferred-tools> blocks containing tool names prefixed with mcp__<plugin>__ - extract unique plugin names from the prefix
• Skill names that contain a colon prefix (e.g., superpowers:brainstorming) - the part before the colon is the plugin name
• Plugin-sourced skills listed in <system-reminder> skill availability blocks (e.g., "- superpowers:writing-plans: ...")

Collect all unique plugin names found.

Classify each plugin into two groups:

• Active plugins: plugins that had at least one skill invoked (detected in step 1). Track which skills were invoked per plugin.
• Tools-only plugins: plugins that only loaded tools (appeared in <available-deferred-tools> or MCP instructions) but had no skills invoked.

5. Detect MCP Tool Calls

Scan Claude's tool call blocks in the conversation for tool names matching the mcp__<server>__<tool> pattern. These are tools that were actually invoked, not merely available.

For each match:

• Extract the <server> portion as the server name
• Extract the <tool> portion as the tool name (strip the mcp__<server>__ prefix)

Collect unique (server, tool) pairs. Group by server name, with tools sorted alphabetically within each server. Servers sorted alphabetically.

This is distinct from step 4 (Detect Plugins), which checks tool availability. This step checks for actual invocation.

6. Scope Filtering

• Default (no --session flag): Only report items detected in the last user message and its corresponding Claude response (the most recent exchange before /used was invoked). For skills, hooks, plugins, and MCP tools, only include those that appeared in that last exchange. For rules, always include all loaded rules (they are session-scoped and always active).
• With --session flag: Report items detected across the entire conversation

7. Output Format

Output a structured summary with counts and plugin grouping. Follow this template exactly:

[Header]  [N skills | N hooks | N rules | N plugins | N mcp tools]
--------------------------------------------------------------
Skills:     skill1, skill2, skill3
Hooks:      hook1, hook2
Rules:      rule1, rule2, rule3,
            rule4, rule5, rule6
Plugins:    active-plugin (skill1, skill2)
            tools-only-plugin1*, tools-only-plugin2*
MCP Tools:  server1 (tool1, tool2)
            server2 (tool3)

* tools only (no skills invoked)

Header:

• Default mode: Last Prompt Usage
• Session mode: Session Usage

Summary counts:

• Shown in square brackets on the same line as the header
• Format: [N skills | N hooks | N rules | N plugins | N mcp tools] using actual detected counts
• Use singular form when count is 1 (e.g., 1 skill, 1 mcp tool)

Category rows:

• If a category has no detected items, show (none) as the value
• Sort items alphabetically within each category
• When a list is long, wrap to the next line with indentation aligned to the first item

Plugins row:

• List active plugins first, each on its own line, with their invoked skill names in parentheses
• List tools-only plugins after, on their own line, comma-separated, each suffixed with *
• If there are any tools-only plugins, add a blank line then the legend: * tools only (no skills invoked)
• If all plugins are tools-only, list them all with * suffix
• If all plugins are active, skip the legend

MCP Tools row:

• Group tools by server name, one server per line
• Tool names are the part after mcp__<server>__, sorted alphabetically within each server
• Servers sorted alphabetically
• If no MCP tools were invoked, show (none) as the value
• Indentation of continuation lines aligns with the first server name

General rules:

• Do not wrap the output in a code block
• Do not add any extra text beyond the format above