/prime

<prime-command> <objective> Build a lightweight understanding of codebase structure and conventions. </objective> <constraints> <rule>MINIMIZE context usage - aim for under 20k tokens total</rule> <rule>DO NOT read source code files (.py, .ts, .js, etc.) - only list them</rule> <rule>DO NOT read test files - only note their existence</rule> <rule>DO NOT read agent definitions - only list available agents</rule> <rule>DO NOT launch subagents - this is a quick overview only</rule> <rule>ONLY read: README.md, config files (pyproject.toml, package.json), and CLAUDE.md</rule> </constraints> <phase number="1" name="structure-discovery"> <step name="file-listing"> <action>Get file listing and summarize structure</action> <command>git ls-files | head -100</command> <output>List directories and count files per directory - do not enumerate every file</output> </step>

<step name="read-readme">
  <action>Read README.md only</action>
  <extract>Project purpose, tech stack, key commands</extract>
</step>

</phase> <phase number="2" name="config-detection"> <step name="find-config"> <action>Identify which config file exists (only ONE)</action> <priority>pyproject.toml > package.json > go.mod > Cargo.toml > pom.xml</priority> <read>Read ONLY the first config file found</read> </step>

<step name="check-claude-config">
  <action>Check for CLAUDE.md if it exists</action>
  <glob>**/CLAUDE.md</glob>
</step>

</phase> <phase number="3" name="inventory-only"> <step name="list-commands"> <action>List available slash commands (filenames only)</action> <glob>.claude/commands/*.md OR plugins/*/commands/*.md</glob> <output>List names only, do not read contents</output> </step>

<step name="list-agents">
  <action>List available agents (filenames only)</action>
  <glob>.claude/agents/*.md OR plugins/*/agents/*.md</glob>
  <output>List names only, do not read contents</output>
</step>

<step name="list-tests">
  <action>Note test directory existence</action>
  <glob>tests/**/*.py OR test/**/*.js OR __tests__/**/*</glob>
  <output>Report count only (e.g., "12 test files found")</output>
</step>

</phase> <phase number="4" name="summarize"> <format>Concise markdown summary with:</format> <sections> <section>Project: 1-2 sentence description</section> <section>Tech: Language, framework, package manager</section> <section>Structure: Key directories (3-5 max)</section> <section>Commands: List available /commands</section> <section>Agents: List available agents</section> <section>Tests: Framework and count</section> <section>Next: What to run for deeper analysis</section> </sections> </phase> <anti-patterns> <avoid>Reading full source files to "understand patterns"</avoid> <avoid>Reading test files to "understand testing approach"</avoid> <avoid>Reading multiple similar files</avoid> <avoid>Launching codebase-analyst agent (use /feature for deep analysis)</avoid> <avoid>Producing multi-page summaries</avoid> </anti-patterns> </prime-command>