profile

Load an expert thinking profile's action file and follow it directly. Action files are self-contained -- they include voice, approach, priorities, and output format. Single targets execute directly. Multiple targets launch parallel agents. Load Configuration Resolve the user's home directory. echo $HOME Never pass `~` to the Read tool. Tell the user to run `/setup-tl`. Parse Arguments Parse `$ARGUMENTS` for profile ID, action, and optional target.

<if condition="no-profile-id">
  <action>Check `preferences.json` for `default_profile`. If set, use it.</action>
  <if condition="no-default-profile">
    <action>List available profiles from `master-index.json`.</action>
    <ask-user-question>
      <question>Which profile would you like to use?</question>
      <option-with-text-input>Enter profile ID or name:</option-with-text-input>
    </ask-user-question>
  </if>
</if>

<if condition="no-action">
  <action>Look up the profile's available actions from `master-index.json`.</action>
  <if condition="only-one-action">
    <action>Use it.</action>
  </if>
  <if condition="multiple-actions">
    <ask-user-question>
      <question>Which action would you like to run?</question>
      <option-with-text-input>Enter action name:</option-with-text-input>
    </ask-user-question>
  </if>
</if>

<if condition="no-target">
  <ask-user-question>
    <question>What should I review? Provide a file path, directory, or describe what to look at.</question>
  </ask-user-question>
</if>

Detect Execution Mode Determine whether to run in direct mode (single target) or parallel mode (multiple targets).

<parallel-triggers>
  <trigger>Multiple explicit targets (e.g., "plugins/a and plugins/b")</trigger>
  <trigger>User says "in parallel", "all at once", "simultaneously"</trigger>
  <trigger>User says "all plugins", "every plugin", "each plugin"</trigger>
  <trigger>Comma-separated paths or glob patterns expanding to multiple directories</trigger>
</parallel-triggers>

<direct-triggers>
  <trigger>Single target path or description (this is the default)</trigger>
  <trigger>No parallelism keywords</trigger>
  <trigger>User says "just this one", "only"</trigger>
</direct-triggers>

<action>If parallel mode detected, resolve all targets now:</action>
<resolution-rules>
  <rule>"all plugins" → list directories under `plugins/` in the working directory (or ask the user to confirm the base path)</rule>
  <rule>Comma-separated → split into individual target paths</rule>
  <rule>Glob pattern → expand to matching directories</rule>
</resolution-rules>

<if condition="parallel and targets > 10">
  <ask-user-question>
    <question>That's {count} targets. Running all in parallel will use significant context. Run all at once, or batch into groups of 5?</question>
    <option label="All at once">Launch all targets in parallel</option>
    <option label="Batch (groups of 5)">Run 5 at a time, then the next 5</option>
  </ask-user-question>
</if>

Resolve Profile

<phase name="find-profile" number="1">
  <action>Find the profile in `master-index.json` by ID.</action>
  <if condition="not-found">
    <action>Search by partial match or display name.</action>
    <if condition="still-not-found">
      <action>Tell the user: "Profile '<id>' not found. Run `/manage-profiles list` to see available profiles."</action>
      <exit />
    </if>
  </if>
</phase>

<phase name="verify-action" number="2">
  <action>Verify the requested action exists for this profile by checking the profile's actions in `master-index.json`.</action>
  <if condition="action-not-found">
    <action>Tell the user: "Profile '<name>' doesn't have a '<action>' action. Available: <list>. Run `/manage-profiles edit <id>` to add one."</action>
    <exit />
  </if>
</phase>

Load and Execute Action File

<action>This file IS the complete instruction set. It contains:</action>
<output-format>
- Voice: How to communicate (tone, phrases, rhetorical patterns)
- Approach: The 4-phase execution structure
- Lens: How this person approaches this activity
- Priorities: What to check, in order
- Typical Questions: Questions to ask in their voice
- Red Flags: Patterns to flag
- Approval Signals: Patterns to praise
- Output Format: How to structure findings
- Counterpoint: Mandatory blind spots to surface
</output-format>

<branch on="execution-mode">

  <direct-mode>
    <description>Single target — execute directly (default, unchanged behavior)</description>
    <constraint>Follow the action file as your instructions. Apply it to the target. Execute all 4 phases. Phase 4 (Counterpoint) is mandatory -- never skip it.</constraint>
  </direct-mode>

  <parallel-mode>
    <description>Multiple targets — launch parallel action-runner agents</description>

    <phase name="launch" number="1">
      <action>For each target, launch a Task with:</action>
      <task-config>
        <subagent_type>think-like:action-runner</subagent_type>
        <prompt>
          You are executing a think-like action file against a specific target.

          **Target**: <target path>
          **Target Label**: <short label, e.g., plugin name or directory name>

          ## Action File

          <full content of the action file, pasted verbatim>

          ## Instructions

          Follow the action file above as your complete instruction set. Apply it to the target path. Execute all 4 phases. Phase 4 (Counterpoint) is mandatory. Begin your output with `## <target label>`.
        </prompt>
      </task-config>
      <constraint>Launch ALL tasks in a single message before waiting for any results. This is critical for parallel execution — do not launch one, wait, then launch the next.</constraint>
    </phase>

    <phase name="collect" number="2">
      <action>Collect results from all agents.</action>
    </phase>

    <phase name="synthesize" number="3">
      <action>Present a combined report:</action>
      <output-structure>
        1. **Per-target sections**: Include each agent's full output under its target label header
        2. **Cross-Target Synthesis**: Add a final section that identifies:
           - Patterns that appear across multiple targets
           - Targets with the most/fewest issues
           - Systemic concerns vs. isolated findings
           - Priority ordering across all targets
      </output-structure>
    </phase>

    <fallback>
      <condition>If the action-runner agent is not available (agent type not found)</condition>
      <action>Fall back to sequential direct execution. Warn the user: "Parallel agent not available — running sequentially. This will take longer." Then execute the action file directly against each target one at a time, collecting results.</action>
    </fallback>
  </parallel-mode>

</branch>

Session Logging

  <branch on="execution-mode">

    <direct-mode>
      <write path="<home>/.things/think-like/sessions/<date>-<profile>-<action>.md">
        <action>Write the full output from step 5 to the session file. The action file's Output Format section defines the structure — use whatever was actually produced. Prepend frontmatter and append the footer, but the body is the complete action output as-delivered in the conversation.</action>

        <template name="session-log-wrapper">
        ```markdown
        ---
        profile: "<id>"
        action: "<action>"
        target: "<target description>"
        date: <YYYY-MM-DD>
        ---

        <full action output from step 5 — preserve all structure, tables, code blocks, and formatting exactly as delivered>
        ```
        </template>

        <constraint>The action output IS the session log body. Do not summarize, compress, or restructure it. Different actions produce different output formats — a security audit looks nothing like a code review. Write what was produced, not a generic summary.</constraint>

        <fallback>If for any reason the action output is unavailable, fall back to this minimal format:</fallback>
        <template name="session-log-fallback">
        ```markdown
        ## Target
        <what was reviewed>

        ## Summary
        <2-3 sentence summary of key findings>

        ## Key Findings
        <bulleted list of findings>
        ```
        </template>
      </write>
    </direct-mode>

    <parallel-mode>
      <write path="<home>/.things/think-like/sessions/<date>-<profile>-<action>-batch.md">
        <action>Write a single batch session log containing all parallel results plus the cross-target synthesis.</action>

        <template name="batch-session-log">
        ```markdown
        ---
        profile: "<id>"
        action: "<action>"
        mode: "parallel"
        targets:
          - "<target 1>"
          - "<target 2>"
          - "..."
        date: <YYYY-MM-DD>
        ---

        <per-target results followed by cross-target synthesis — preserve all structure exactly as delivered>
        ```
        </template>
      </write>
    </parallel-mode>

  </branch>

</if>

Update Last Used Update `last_used` to current date. Confirm After the review output, add: --- Profile: | Action: | Mode: | Targets: | Session logged:

  `/profile <id> <action>` to run again | `/manage-profiles list` to see all profiles
  </output>
</completion-message>