feat: use direct ${CLAUDE_PLUGIN_ROOT} in commands instead of skill workaround (#182)

Now that Claude Code 2.1.0 properly substitutes ${CLAUDE_PLUGIN_ROOT} in
bash execution, we can simplify commands to directly reference scripts:

- add-to-backlog.md: Direct script execution
- code-review.md: Direct script execution
- complete-task.md: Direct script execution
- prepare-completion.md: Direct script execution
- specify.md: Direct script execution

Removed:
- test-plugin-root.md (testing command no longer needed)
- Skill tool from allowed-tools (no longer needed)

This eliminates the two-step "invoke skill, then run script" workaround
that was necessary when ${CLAUDE_PLUGIN_ROOT} wasn't being substituted.

Co-authored-by: Craig Haseler <cahaseler@users.noreply.github.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: cahaseler

commands/add-to-backlog.md

@@ -2,15 +2,15 @@
 shortname: add-to-backlog
 description: Quickly add an item to the backlog without disrupting current work
 usage: /add-to-backlog "Your idea or bug description"
-allowed-tools: Bash(bun:*), Skill
+allowed-tools: Bash(bun:*)
 ---
 
 # Add to Backlog
 
-Invoke the `cc-track:cc-track-tools` skill to get the script path, then run:
+Run the backlog script directly:
 
 ```bash
-bun {base_directory}/scripts/backlog.ts "$ARGUMENTS"
+bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/backlog.ts" "$ARGUMENTS"
 ```
 
 Do not change your current focus or priorities. Continue with your work if there was a clear task you are working on.

commands/code-review.md

@@ -1,5 +1,5 @@
 ---
-allowed-tools: Read, Write, Edit, Glob, Grep, Bash(git:*), Bash(bun:*), Bash(mkdir:*), Task, Skill, AskUserQuestion, TodoWrite
+allowed-tools: Read, Write, Edit, Glob, Grep, Bash(git:*), Bash(bun:*), Bash(mkdir:*), Task, AskUserQuestion, TodoWrite
 description: Code review a pull request
 ---
 
@@ -12,18 +12,12 @@ Run validation checks and multi-agent code review without requiring a spec folde
 - Current branch: !`git branch --show-current`
 - Changes to review: !`git diff --stat`
 
-## Step 1: Get Script Path via Skill
+## Step 1: Run Validation Script
 
-First, invoke the `cc-track:cc-track-tools` skill to get the base directory for cc-track scripts.
-
-Note the base directory provided (e.g., `/path/to/skills/cc-track-tools`).
-
-## Step 2: Run Validation Script
-
-Run the code-review script using the base directory from Step 1:
+Run the code-review script:
 
 ```bash
-bun {base_directory}/scripts/code-review.ts
+bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/code-review.ts"
 ```
 
 The script will:
@@ -32,7 +26,7 @@ The script will:
 - Run test suite
 - Run dead code detection (knip)
 
-## Step 3: Interpret Validation Results
+## Step 2: Interpret Validation Results
 
 The script returns JSON output. Parse and present the results:
 
@@ -48,9 +42,9 @@ The script returns JSON output. Parse and present the results:
 
 **If validation passed:**
 - Confirm: "✅ All validation checks passed"
-- Proceed to Step 4 for multi-agent code review
+- Proceed to Step 3 for multi-agent code review
 
-## Step 4: Launch Review Agents
+## Step 3: Launch Review Agents
 
 Launch the multi-agent code review using the Task tool.
 
@@ -92,9 +86,9 @@ Use the Task tool to launch these agents simultaneously. Include the context abo
 - Launch all 5 agents in a single message without the `run_in_background` parameter
 - Orchestrator suspends until ALL agents complete
 - All results return together in the response
-- Deduplication in Step 5 has access to all findings at once
+- Deduplication in Step 4 has access to all findings at once
 
-## Step 5: Collect and Deduplicate Issues
+## Step 4: Collect and Deduplicate Issues
 
 After all agents complete:
 
@@ -111,9 +105,9 @@ After all agents complete:
    - Observation: [evidence from reviewer]
    ```
 
-**If no issues found by any reviewer:** Skip to Step 8 with "No issues found - clean review!"
+**If no issues found by any reviewer:** Skip to Step 7 with "No issues found - clean review!"
 
-## Step 6: Score Each Issue
+## Step 5: Score Each Issue
 
 For each unique issue, launch a parallel **issue-scorer** agent (Haiku) to validate it.
 
@@ -130,7 +124,7 @@ Reported by: {reviewer(s)}
 Observation: {what the reviewer observed}
 ```
 
-## Step 7: Filter and Present Results
+## Step 6: Filter and Present Results
 
 After all scorers complete:
 
@@ -169,7 +163,7 @@ After all scorers complete:
 - [N] issues were filtered as unverified or false positives
 ```
 
-## Step 8: Route Based on Issue Count
+## Step 7: Route Based on Issue Count
 
 After filtering and scoring, route based on how many validated issues were found:
 
@@ -205,13 +199,13 @@ Wait for user direction before proceeding. If they want it fixed, implement the
 📋 Found {N} validated issues. Starting structured triage...
 ```
 
-Proceed to Step 9 for inline fix-issues flow.
+Proceed to Step 8 for inline fix-issues flow.
 
-## Step 9: Inline Fix-Issues Flow
+## Step 8: Inline Fix-Issues Flow
 
 For >1 validated issues, handle triage directly (don't call `/cc-track:fix-issues` which requires a spec folder).
 
-### 9a: Create Review File
+### 8a: Create Review File
 
 1. Ensure `.cc-track/reviews/` directory exists:
    ```bash
@@ -240,7 +234,7 @@ For >1 validated issues, handle triage directly (don't call `/cc-track:fix-issue
 
 4. Add each issue to the file with Status = "pending"
 
-### 9b: Check Existing State
+### 8b: Check Existing State
 
 Read existing files to filter issues that were already triaged:
 
@@ -257,7 +251,7 @@ Read existing files to filter issues that were already triaged:
    - Remove any issue matching a Deferred entry (match by location or description)
    - Keep all other issues for triage
 
-### 9c: "Just Fix It" Protocol
+### 8c: "Just Fix It" Protocol
 
 Before presenting issues to the user, scan for trivial fixes that don't need human decision-making.
 
@@ -279,7 +273,7 @@ Before presenting issues to the user, scan for trivial fixes that don't need hum
 3. Update review file with Decision = "Just Fix It"
 4. Notify user briefly before continuing to triage
 
-### 9d: Triage Loop
+### 8d: Triage Loop
 
 For each remaining issue (sorted by score, highest first):
 
@@ -329,7 +323,7 @@ multiSelect: false
 - **Dismiss:** Update review file with Decision = "Dismiss" and rationale
 - **Discuss:** Investigate using Explore agent, then re-present with Fix/Defer/Dismiss
 
-### 9e: Action Summary
+### 8e: Action Summary
 
 After all issues are triaged:
 
@@ -352,7 +346,7 @@ After all issues are triaged:
 
 Update review file Status to "triaged".
 
-### 9f: Execute Fixes
+### 8f: Execute Fixes
 
 **If no fixes needed:**
 ```

commands/complete-task.md

@@ -1,5 +1,5 @@
 ---
-allowed-tools: Read, Edit, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(bun:*), Skill
+allowed-tools: Read, Edit, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(bun:*)
 description: Complete the current active task (Phase 2 of task completion workflow)
 ---
 
@@ -15,18 +15,12 @@ Finalize the current task by updating metadata, squashing commits, and creating
 - Git status: !`git status --short`
 - Recent commits on this branch: !`git log --oneline -5`
 
-## Step 1: Get Script Path via Skill
+## Step 1: Run Complete Task Script
 
-First, invoke the `cc-track:cc-track-tools` skill to get the base directory for cc-track scripts.
-
-Note the base directory provided (e.g., `/path/to/skills/cc-track-tools`).
-
-## Step 2: Run Complete Task Script
-
-Run the complete-task script using the base directory from Step 1:
+Run the complete-task script:
 
 ```bash
-bun {base_directory}/scripts/complete-task.ts
+bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/complete-task.ts"
 ```
 
 The script will:
@@ -38,7 +32,7 @@ The script will:
 - Switch back to main branch
 - Update CLAUDE.md to reference no_active_task.md
 
-## Step 3: Interpret Results
+## Step 2: Interpret Results
 
 The script returns JSON-like output with messages and data. Parse and present the results:
 
@@ -56,7 +50,7 @@ The script returns JSON-like output with messages and data. Parse and present th
 - Report PR status: URL if created, or existing PR if updated
 - Confirm branch switch to main
 
-## Step 4: Enhance PR Description (if PR was created)
+## Step 3: Enhance PR Description (if PR was created)
 
 If a new PR was created, enhance its description with details from the spec files:
 
@@ -89,5 +83,5 @@ The script supports these optional flags:
 
 Example with flags:
 ```bash
-bun {base_directory}/scripts/complete-task.ts --no-squash
+bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/complete-task.ts" --no-squash
 ```

commands/prepare-completion.md

@@ -1,5 +1,5 @@
 ---
-allowed-tools: Read, Edit, Glob, Grep, Bash(git:*), Bash(bun:*), Task, Skill
+allowed-tools: Read, Edit, Glob, Grep, Bash(git:*), Bash(bun:*), Task
 description: Prepare the current active task for completion (Phase 1 of task completion workflow)
 ---
 
@@ -12,18 +12,12 @@ Run validation checks and multi-agent spec-focused code review to ensure the tas
 - Current branch: !`git branch --show-current`
 - Changes to review: !`git diff --stat`
 
-## Step 1: Get Script Path via Skill
+## Step 1: Run Validation Script
 
-First, invoke the `cc-track:cc-track-tools` skill to get the base directory for cc-track scripts.
-
-Note the base directory provided (e.g., `/path/to/skills/cc-track-tools`).
-
-## Step 2: Run Validation Script
-
-Run the prepare-completion script using the base directory from Step 1:
+Run the prepare-completion script:
 
 ```bash
-bun {base_directory}/scripts/prepare-completion.ts
+bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/prepare-completion.ts"
 ```
 
 The script will:
@@ -32,9 +26,9 @@ The script will:
 - Run test suite
 - Run dead code detection (knip)
 
-**Note:** The script no longer runs code review - that's handled by the multi-agent review in Step 3.
+**Note:** The script no longer runs code review - that's handled by the multi-agent review in Step 2.
 
-## Step 3: Interpret Validation Results
+## Step 2: Interpret Validation Results
 
 The script returns JSON output. Parse and present the results:
 
@@ -50,9 +44,9 @@ The script returns JSON output. Parse and present the results:
 
 **If validation passed:**
 - Confirm: "✅ All validation checks passed"
-- Proceed to Step 4 for multi-agent code review
+- Proceed to Step 3 for multi-agent code review
 
-## Step 4: Launch Review Agents (Phase 1 - Find Issues)
+## Step 3: Launch Review Agents (Phase 1 - Find Issues)
 
 Launch the multi-agent spec-focused code review using the Task tool.
 
@@ -108,11 +102,11 @@ Use the Task tool to launch these agents simultaneously. Include the context abo
 - Launch all 8 agents in a single message without the `run_in_background` parameter
 - Orchestrator suspends until ALL agents complete
 - All results return together in the response
-- Deduplication in Step 5 has access to all findings at once
+- Deduplication in Step 4 has access to all findings at once
 
 This is different from the pipeline pattern in `/cc-track:tasks` - here we need ALL results before proceeding to deduplication, so foreground parallel is correct.
 
-## Step 5: Collect and Deduplicate Issues
+## Step 4: Collect and Deduplicate Issues
 
 After all agents complete:
 
@@ -129,9 +123,9 @@ After all agents complete:
    - Observation: [evidence from reviewer]
    ```
 
-**If no issues found by any reviewer:** Skip to Step 8 with "No issues found - clean review!"
+**If no issues found by any reviewer:** Skip to Step 7 with "No issues found - clean review!"
 
-## Step 6: Score Each Issue (Phase 2 - Validate Issues)
+## Step 5: Score Each Issue (Phase 2 - Validate Issues)
 
 For each unique issue, launch a parallel **issue-scorer** agent (Haiku) to validate it.
 
@@ -148,7 +142,7 @@ Reported by: {reviewer(s)}
 Observation: {what the reviewer observed}
 ```
 
-## Step 7: Filter and Present Results
+## Step 6: Filter and Present Results
 
 After all scorers complete:
 
@@ -187,7 +181,7 @@ After all scorers complete:
 - [N] issues were filtered as unverified or false positives
 ```
 
-## Step 8: Route Based on Issue Count
+## Step 7: Route Based on Issue Count
 
 After filtering and scoring, route based on how many validated issues were found:
 

commands/specify.md

@@ -1,6 +1,6 @@
 ---
 description: Create a feature specification through Socratic questioning and iterative refinement
-allowed-tools: Read, Edit, Write, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(bun:*), Task, AskUserQuestion, WebSearch, WebFetch, Skill
+allowed-tools: Read, Edit, Write, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(bun:*), Task, AskUserQuestion, WebSearch, WebFetch
 ---
 
 # Creating Feature Specifications
@@ -25,9 +25,11 @@ allowed-tools: Read, Edit, Write, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(bun:
 **CRITICAL**: Infrastructure must be created BEFORE any Socratic questioning or research begins. This prevents getting distracted before the branch/folder exist.
 
 1. **Get task title**: Use from user's request and proceed to create infrastructure. Only ask for clarification if the title is genuinely unclear - do not ask for approval of the title itself.
-2. **Invoke skill**: `cc-track:cc-track-tools` to get base directory
-3. **Run script**: `bun {base_directory}/scripts/specify.ts "{title}"`
-4. **Confirm results**: Branch created, spec directory ready, CLAUDE.md updated, GitHub issue created (if enabled)
+2. **Run script**:
+   ```bash
+   bun "${CLAUDE_PLUGIN_ROOT}/skills/cc-track-tools/scripts/specify.ts" "{title}"
+   ```
+3. **Confirm results**: Branch created, spec directory ready, CLAUDE.md updated, GitHub issue created (if enabled)
 
 If script fails, report error and stop.
 

commands/test-plugin-root.md

@@ -21,15 +21,6 @@
             "command": "bun run ${CLAUDE_PLUGIN_ROOT}/hooks/pre-tool-validation.ts"
           }
         ]
-      },
-      {
-        "matcher": "Bash",
-        "hooks": [
-          {
-            "type": "command",
-            "command": "bun run ${CLAUDE_PLUGIN_ROOT}/hooks/powershell-guidance.ts"
-          }
-        ]
       }
     ],
     "UserPromptSubmit": [