bp-enhance

name	bp-enhance
description	This skill should be used when the user asks to "enhance a skill", "improve skill structure", "add pseudocode", "add state management", "tighten prose", "add guards", "strengthen skill", "add checkpoints", "improve without workflow". Triggers on "enhance", "improve", "add structure", "pseudocode", "state management", "strengthen", "tighten", "guards", "checkpoints".
allowed-tools	Read, Write, Edit, Glob, Grep, AskUserQuestion
inputs	[{"name":"skill_path","type":"string","required":false,"description":"Path to the skill directory or SKILL.md (prompted if not provided)"}]
outputs	[{"name":"enhancements_applied","type":"array","description":"List of enhancement types applied to the skill"},{"name":"updated_skill","type":"boolean","description":"Whether SKILL.md was modified"}]

Enhance Skill Structure

Add structure to prose-based skills — pseudocode blocks, state management, precondition guards, phase boundary contracts, checkpoints, error handling — WITHOUT converting to workflow YAML. This is the "prose territory" improvement path: tighter prose, not more formalization.

Phase Detection Algorithm: ${CLAUDE_PLUGIN_ROOT}/patterns/phase-detection-algorithm.md Complexity Scoring Algorithm: ${CLAUDE_PLUGIN_ROOT}/patterns/complexity-scoring-algorithm.md Quality Indicators: ${CLAUDE_PLUGIN_ROOT}/patterns/quality-indicators.md

---

Overview

This skill operates on a single SKILL.md file and improves its internal structure while preserving the prose orchestration approach. It does NOT create workflow YAML files.

What It Does	What It Does NOT Do
Adds pseudocode blocks for complex logic	Convert phases to workflow YAML
Introduces computed.* state management	Change the skill's phase count or scope
Adds precondition guards at phase boundaries	Remove existing prose instructions
Clarifies phase input/output contracts	Alter the skill's external interface
Adds checkpoint/restore patterns	Replace natural-language with rigid schemas
Adds error handling and recovery paths	Second-guess the author's design choices
Completes frontmatter input/output declarations

Enhancement types:

Type	What It Adds	When Recommended
state_management	computed.* namespace, flag definitions	No state refs found, >2 phases
pseudocode_blocks	Structured logic for complex phases	Phase has >3 conditionals, no pseudocode
precondition_guards	Entry conditions for phases	Phases depend on prior phase output, no guards
phase_boundaries	Clearer phase separation with inputs/outputs	Phases lack clear contracts
checkpoint_patterns	Save/restore points	>4 phases, long-running operations
error_handling	Failure modes and recovery paths	No error handling, external tool calls
io_contracts	Explicit frontmatter declarations	Missing inputs/outputs in frontmatter

---

Phase 1: Mode Detection

Parse invocation arguments to determine enhancement behavior.

Step 1.1: Parse Arguments

PARSE_MODE(args):
  computed.mode = "interactive"
  computed.skill_path = null
  computed.analyze_only = false
  computed.requested_enhancement = null

  IF args contains "--analyze-only":
    computed.analyze_only = true

  IF args contains "--enhancement <type>":
    type = extract_value(args, "--enhancement")
    IF type IN ENHANCEMENT_CATALOG:
      computed.requested_enhancement = type
      computed.mode = "targeted"
    ELSE:
      DISPLAY "Unknown enhancement type: " + type
      DISPLAY "Valid types: state_management, pseudocode_blocks, precondition_guards,"
      DISPLAY "  phase_boundaries, checkpoint_patterns, error_handling, io_contracts"
      EXIT

  IF args contains a path (bare argument or --skill <path>):
    computed.skill_path = extract_path(args)

  IF computed.analyze_only AND computed.requested_enhancement:
    DISPLAY "Warning: --analyze-only and --enhancement are mutually exclusive."
    DISPLAY "Running in analyze-only mode (--analyze-only takes precedence)."
    computed.requested_enhancement = null
    computed.mode = "interactive"

Step 1.2: Resolve Skill Path

If computed.skill_path is not set, prompt the user:

{
  "questions": [{
    "question": "Which skill would you like to enhance?",
    "header": "Target Skill",
    "multiSelect": false,
    "options": [
      {"label": "Provide path", "description": "I'll give you the skill directory or SKILL.md path"},
      {"label": "Search current plugin", "description": "Look for skills in this plugin's skills/ directory"}
    ]
  }]
}

Response handling:

HANDLE_TARGET(response):
  SWITCH response:
    CASE "Provide path":
      # Ask for path via follow-up prompt
      computed.skill_path = resolve_skill_path(user_provided_path)

    CASE "Search current plugin":
      candidates = Glob("${CLAUDE_PLUGIN_ROOT}/skills/*/SKILL.md")
      # Present candidates as AskUserQuestion with each skill name as an option
      computed.skill_path = selected_candidate_path

Step 1.3: Load Skill Content

LOAD_SKILL():
  IF is_directory(computed.skill_path):
    computed.skill_path = computed.skill_path + "/SKILL.md"

  computed.skill_content = Read(computed.skill_path)
  computed.skill_dir = parent_directory(computed.skill_path)
  computed.skill_name = basename(computed.skill_dir)

  # Parse frontmatter
  frontmatter = content between first "---" and second "---"
  parsed = parse_yaml(frontmatter)

  computed.frontmatter = {
    name: parsed.name,
    description: parsed.description,
    allowed_tools: split(parsed["allowed-tools"], ", "),
    inputs_defined: "inputs" IN parsed AND len(parsed.inputs) > 0,
    outputs_defined: "outputs" IN parsed AND len(parsed.outputs) > 0,
    inputs: parsed.inputs OR [],
    outputs: parsed.outputs OR []
  }

---

Phase 2: Analyze

Examine the skill's current structure to identify enhancement opportunities.

Step 2.1: Identify Phases

Use the phase detection algorithm from ${CLAUDE_PLUGIN_ROOT}/patterns/phase-detection-algorithm.md:

IDENTIFY_PHASES():
  body = content_after_frontmatter(computed.skill_content)
  computed.phases = []

  FOR line_num, line IN enumerate(body_lines):
    confidence = detect_phase_marker(line)
    IF confidence:
      phase = {
        id: slugify(extracted_title),
        title: extracted_title,
        line_start: line_num,
        line_end: null,  # filled in below
        confidence: confidence
      }
      computed.phases.append(phase)

  # Fill in line_end for each phase (start of next phase or end of body)
  FOR i IN range(len(computed.phases)):
    IF i + 1 < len(computed.phases):
      computed.phases[i].line_end = computed.phases[i + 1].line_start - 1
    ELSE:
      computed.phases[i].line_end = len(body_lines)

  IF len(computed.phases) == 0:
    computed.phases = [{
      id: "main",
      title: "Main",
      line_start: 1,
      line_end: len(body_lines),
      confidence: "low"
    }]

Step 2.2: Analyze Current Structure

For each phase, measure existing structural elements:

ANALYZE_STRUCTURE():
  computed.current_structure = {
    phase_count: len(computed.phases),
    pseudocode_blocks: [],
    state_refs: [],
    precondition_guards: [],
    phase_contracts: [],
    checkpoints: [],
    error_handlers: [],
    io_contracts: {
      inputs_declared: computed.frontmatter.inputs_defined,
      outputs_declared: computed.frontmatter.outputs_defined
    }
  }

  FOR phase IN computed.phases:
    content = get_lines(body, phase.line_start, phase.line_end)

    # Count pseudocode blocks
    pseudocode_blocks = find_fenced_blocks(content, "pseudocode")
    FOR block IN pseudocode_blocks:
      computed.current_structure.pseudocode_blocks.append({
        phase: phase.id,
        line: block.start_line,
        length: block.line_count
      })

    # Count state management references
    state_patterns = find_all_patterns(content, [
      /computed\.\w+/,
      /flags\.\w+/,
      /state\.\w+/
    ])
    FOR ref IN state_patterns:
      computed.current_structure.state_refs.append({
        phase: phase.id,
        reference: ref.matched_text,
        line: ref.line_num
      })

    # Check for precondition guards
    guard_patterns = find_all_patterns(content, [
      /GUARD:|PRECONDITION:|REQUIRES:/i,
      /IF\s+computed\.\w+\s+(is|==|!=)\s+null.*EXIT/i,
      /Assert|Verify.*before\s+proceed/i
    ])
    FOR guard IN guard_patterns:
      computed.current_structure.precondition_guards.append({
        phase: phase.id,
        guard_text: guard.matched_text,
        line: guard.line_num
      })

    # Check phase boundary clarity
    has_input_declaration = matches_any(content, [
      /inputs?:|receives?:|expects?:/i,
      /from\s+phase\s+\d/i
    ])
    has_output_declaration = matches_any(content, [
      /outputs?:|produces?:|sets?:/i,
      /stored?\s+in\s+computed\./i
    ])
    computed.current_structure.phase_contracts.append({
      phase: phase.id,
      has_input_spec: has_input_declaration,
      has_output_spec: has_output_declaration
    })

    # Check checkpoint usage
    checkpoint_patterns = find_all_patterns(content, [
      /CHECKPOINT:|SAVE_STATE:|RESTORE:/i,
      /checkpoint|save.*progress|resume/i
    ])
    FOR cp IN checkpoint_patterns:
      computed.current_structure.checkpoints.append({
        phase: phase.id,
        text: cp.matched_text,
        line: cp.line_num
      })

    # Check error handling
    error_patterns = find_all_patterns(content, [
      /ERROR:|FAILURE:|FALLBACK:/i,
      /IF.*fail|IF.*error/i,
      /on_error|on_failure|recovery/i,
      /DISPLAY.*error|warn.*user/i
    ])
    FOR err IN error_patterns:
      computed.current_structure.error_handlers.append({
        phase: phase.id,
        text: err.matched_text,
        line: err.line_num
      })

    # Per-phase complexity metrics (for recommendation targeting)
    phase.conditionals = count_patterns(content, [
      /If\s+.*then/i, /When\s+/i, /Based on/i,
      /Depending on/i, /Unless/i, /Either.*or/i
    ])
    phase.tool_calls = count_tool_references(content)
    phase.has_pseudocode = len(pseudocode_blocks) > 0
    phase.has_state_refs = len(state_patterns) > 0
    phase.has_guards = len(guard_patterns) > 0
    phase.has_error_handling = len(error_patterns) > 0
    phase.prose_lines = count_non_empty_lines(content)

Step 2.3: Compute Structure Summary

SUMMARIZE_STRUCTURE():
  s = computed.current_structure

  computed.structure_summary = {
    total_pseudocode_blocks: len(s.pseudocode_blocks),
    total_state_refs: len(s.state_refs),
    unique_state_vars: count_unique(ref.reference for ref in s.state_refs),
    total_guards: len(s.precondition_guards),
    phases_with_input_spec: count(c for c in s.phase_contracts if c.has_input_spec),
    phases_with_output_spec: count(c for c in s.phase_contracts if c.has_output_spec),
    total_checkpoints: len(s.checkpoints),
    total_error_handlers: len(s.error_handlers),
    phases_with_pseudocode: count_unique(b.phase for b in s.pseudocode_blocks),
    phases_without_pseudocode: s.phase_count - count_unique(b.phase for b in s.pseudocode_blocks)
  }

---

Phase 3: Recommend

Generate contextual enhancement recommendations based on the Phase 2 analysis.

Step 3.1: Evaluate Each Enhancement Type

GENERATE_RECOMMENDATIONS():
  computed.recommendations = []

  # --- state_management ---
  IF computed.structure_summary.total_state_refs == 0
     AND computed.current_structure.phase_count > 2:
    computed.recommendations.append({
      type: "state_management",
      rationale: "Skill has " + str(computed.current_structure.phase_count) +
                 " phases but no computed.* state references. Adding a state " +
                 "namespace makes data flow between phases explicit.",
      impact: "high",
      example: generate_state_management_example()
    })

  # --- pseudocode_blocks ---
  FOR phase IN computed.phases:
    IF phase.conditionals > 3 AND NOT phase.has_pseudocode:
      computed.recommendations.append({
        type: "pseudocode_blocks",
        target_phase: phase.id,
        rationale: "Phase '" + phase.title + "' has " + str(phase.conditionals) +
                   " conditionals but no pseudocode block. Pseudocode makes " +
                   "branching logic unambiguous.",
        impact: "high",
        example: generate_pseudocode_example(phase)
      })

  # --- precondition_guards ---
  FOR i, phase IN enumerate(computed.phases):
    IF i > 0 AND NOT phase.has_guards:
      prior = computed.phases[i - 1]
      prior_contract = find(computed.current_structure.phase_contracts,
                            c => c.phase == prior.id)
      IF prior_contract AND prior_contract.has_output_spec:
        computed.recommendations.append({
          type: "precondition_guards",
          target_phase: phase.id,
          rationale: "Phase '" + phase.title + "' depends on output from '" +
                     prior.title + "' but has no entry guard. A precondition " +
                     "guard catches missing prerequisites early.",
          impact: "medium",
          example: generate_guard_example(phase, prior)
        })

  # --- phase_boundaries ---
  phases_missing_contracts = [
    c for c in computed.current_structure.phase_contracts
    if NOT c.has_input_spec OR NOT c.has_output_spec
  ]
  IF len(phases_missing_contracts) > computed.current_structure.phase_count / 2:
    computed.recommendations.append({
      type: "phase_boundaries",
      rationale: str(len(phases_missing_contracts)) + " of " +
                 str(computed.current_structure.phase_count) +
                 " phases lack clear input/output contracts. Adding contracts " +
                 "makes each phase self-documenting.",
      impact: "medium",
      example: generate_boundary_example()
    })

  # --- checkpoint_patterns ---
  IF computed.current_structure.phase_count > 4
     AND computed.structure_summary.total_checkpoints == 0:
    has_long_running = any(p.tool_calls > 3 OR p.prose_lines > 60
                          for p in computed.phases)
    IF has_long_running:
      computed.recommendations.append({
        type: "checkpoint_patterns",
        rationale: "Skill has " + str(computed.current_structure.phase_count) +
                   " phases with long-running operations but no checkpoints. " +
                   "Checkpoints allow resuming from a known-good state on failure.",
        impact: "medium",
        example: generate_checkpoint_example()
      })

  # --- error_handling ---
  phases_without_errors = [p for p in computed.phases
                           if NOT p.has_error_handling AND p.tool_calls > 0]
  IF len(phases_without_errors) > 0:
    computed.recommendations.append({
      type: "error_handling",
      target_phases: [p.id for p in phases_without_errors],
      rationale: str(len(phases_without_errors)) + " phase(s) invoke external " +
                 "tools but have no error handling. Adding failure modes and " +
                 "recovery paths prevents silent failures.",
      impact: "high",
      example: generate_error_handling_example(phases_without_errors[0])
    })

  # --- io_contracts ---
  IF NOT computed.frontmatter.inputs_defined OR NOT computed.frontmatter.outputs_defined:
    missing = []
    IF NOT computed.frontmatter.inputs_defined: missing.append("inputs")
    IF NOT computed.frontmatter.outputs_defined: missing.append("outputs")
    computed.recommendations.append({
      type: "io_contracts",
      rationale: "Frontmatter is missing " + join(missing, " and ") +
                 " declarations. Explicit I/O contracts let callers " +
                 "understand the skill's interface without reading the body.",
      impact: "medium",
      example: generate_io_contract_example()
    })

Step 3.2: Generate Concrete Examples

Each recommendation includes a concrete pseudocode example showing what the enhancement would look like when applied to the target skill.

function generate_state_management_example():
  RETURN """
  ## State Flow

Phase 1 Phase 2 Phase 3 ───────────────────────────────────────────────── computed.target -> computed.analysis -> computed.result computed.mode computed.metrics computed.report

At Phase 1, initialize the state namespace:

```pseudocode
INIT_STATE():
  computed.target = null
  computed.mode = null
  computed.analysis = null

"""

function generate_pseudocode_example(phase): RETURN """ Add to Phase '{phase.title}':

PROCESS_{upper(phase.id)}():
  IF condition_a:
    computed.result = handle_case_a()
  ELIF condition_b AND condition_c:
    computed.result = handle_case_b()
  ELSE:
    computed.result = default_handler()
    DISPLAY "Falling back to default: " + computed.result

"""

function generate_guard_example(phase, prior_phase): RETURN """ Add at the start of Phase '{phase.title}':

GUARD_{upper(phase.id)}():
  IF computed.{prior_phase.id}_output IS null:
    DISPLAY "Cannot proceed: Phase '{prior_phase.title}' has not completed."
    DISPLAY "Required state: computed.{prior_phase.id}_output"
    EXIT

"""

function generate_boundary_example(): RETURN """ Add to each phase heading:

Phase N: Title

Inputs: computed.prior_result (from Phase N-1) Outputs: computed.this_result (used by Phase N+1)

This pattern makes data flow visible at a glance without reading the phase body. """

function generate_checkpoint_example(): RETURN """ Add after critical operations:

CHECKPOINT_AFTER_ANALYSIS():
  computed.checkpoint = {
    phase: "analysis",
    timestamp: current_iso_timestamp(),
    state: {
      target: computed.target,
      metrics: computed.metrics
    }
  }
  # On re-invocation, check for checkpoint:
  IF computed.checkpoint AND computed.checkpoint.phase == "analysis":
    DISPLAY "Resuming from analysis checkpoint."
    RESTORE computed.target, computed.metrics FROM computed.checkpoint.state
    SKIP to Phase 3

"""

function generate_error_handling_example(phase): RETURN """ Add to Phase '{phase.title}' around tool calls:

HANDLE_{upper(phase.id)}_ERRORS():
  result = Read(computed.target_path)
  IF result IS error:
    DISPLAY "Failed to read " + computed.target_path + ": " + result.message
    IF result.type == "file_not_found":
      DISPLAY "Verify the path exists and try again."
      EXIT
    ELIF result.type == "permission_denied":
      DISPLAY "Check file permissions on " + computed.target_path
      EXIT
    ELSE:
      DISPLAY "Unexpected error — aborting."
      EXIT

"""

function generate_io_contract_example(): RETURN """ Add to the YAML frontmatter:

inputs:
  - name: target_path
    type: string
    required: false
    description: Path to the target (prompted if not provided)
outputs:
  - name: result
    type: object
    description: The operation result

"""

### Step 3.3: Display Recommendations

Present the recommendations with their examples:

Enhancement Recommendations: {computed.skill_name}

Current structure:

• Phases: {computed.current_structure.phase_count}
• Pseudocode blocks: {computed.structure_summary.total_pseudocode_blocks}
• State references: {computed.structure_summary.unique_state_vars} unique variables
• Precondition guards: {computed.structure_summary.total_guards}
• Checkpoints: {computed.structure_summary.total_checkpoints}
• Error handlers: {computed.structure_summary.total_error_handlers}
• I/O contracts: inputs={yes/no}, outputs={yes/no}

Recommendations ({count})

{for i, rec in enumerate(computed.recommendations)}

{i+1}. {rec.type} [{rec.impact} impact]

{rec.rationale}

{rec.example}

---

{/for}

IF `computed.analyze_only == true`: display recommendations and STOP. Do not proceed to
Phase 4.

---

## Phase 4: Select

Determine which enhancements to apply.

### Step 4.1: Select Enhancements

If `computed.mode == "targeted"` (the `--enhancement` flag was set):

```pseudocode
HANDLE_TARGETED():
  matching = [r for r in computed.recommendations
              if r.type == computed.requested_enhancement]
  IF len(matching) > 0:
    computed.selected_enhancements = matching
  ELSE:
    DISPLAY "Enhancement type '" + computed.requested_enhancement +
            "' was requested but no recommendation was generated for it."
    DISPLAY "This means the skill already has adequate coverage for this type,"
    DISPLAY "or the prerequisites for this enhancement are not met."
    EXIT

If computed.mode == "interactive", present recommendations as a multi-select:

{
  "questions": [{
    "question": "Which enhancements would you like to apply?",
    "header": "Select Enhancements",
    "multiSelect": true,
    "options": [
      {
        "label": "state_management",
        "description": "Add computed.* namespace and state flow diagram [high impact]"
      },
      {
        "label": "pseudocode_blocks",
        "description": "Add pseudocode for complex conditional logic [high impact]"
      },
      {
        "label": "precondition_guards",
        "description": "Add entry guards at phase boundaries [medium impact]"
      },
      {
        "label" 
         
      
      
         
         
      
      
         
         
      
      
         
         
      
    
  

NOTE: Only present options that have matching recommendations from Phase 3. Filter the options list to include only types present in computed.recommendations.

HANDLE_SELECTION(response):
  selected_types = parse_multi_select(response)
  computed.selected_enhancements = [
    r for r in computed.recommendations
    if r.type IN selected_types
  ]

  IF len(computed.selected_enhancements) == 0:
    DISPLAY "No enhancements selected. Exiting."
    EXIT

---

Phase 5: Apply

Apply each selected enhancement to the SKILL.md file.

Step 5.1: Plan Application Order

Enhancements are applied in a specific order to avoid conflicts:

APPLICATION_ORDER = [
  "io_contracts",           # Frontmatter changes first (top of file)
  "state_management",       # State flow section (typically after overview)
  "phase_boundaries",       # Phase heading annotations
  "precondition_guards",    # Added at phase starts
  "pseudocode_blocks",      # Added within phase bodies
  "error_handling",         # Added within phase bodies
  "checkpoint_patterns"     # Added at phase ends
]

PLAN_APPLICATION():
  computed.application_plan = sorted(
    computed.selected_enhancements,
    key=lambda e: APPLICATION_ORDER.index(e.type)
  )
  computed.enhancements_applied = []

Step 5.2: Apply Each Enhancement

For each enhancement in the application plan:

APPLY_ENHANCEMENTS():
  FOR enhancement IN computed.application_plan:
    # Generate the content to insert or modify
    content = generate_enhancement_content(enhancement)

    # Determine insertion point
    location = determine_insertion_point(enhancement, computed.skill_content)

    # Show diff-style preview
    DISPLAY "--- Enhancement: " + enhancement.type + " ---"
    DISPLAY ""
    DISPLAY "Location: " + location.description
    DISPLAY ""
    DISPLAY "Content to add:"
    DISPLAY "```"
    DISPLAY content
    DISPLAY "```"
    DISPLAY ""

    # Apply via Edit tool
    SWITCH enhancement.type:
      CASE "io_contracts":
        # Edit the frontmatter section to add inputs/outputs
        Edit(computed.skill_path,
          old_string=existing_frontmatter_closing,
          new_string=io_declarations + frontmatter_closing)

      CASE "state_management":
        # Insert State Flow section after Overview (or after frontmatter)
        Edit(computed.skill_path,
          old_string=first_phase_heading,
          new_string=state_flow_section + "\n\n" + first_phase_heading)

      CASE "phase_boundaries":
        # Annotate each phase heading with Inputs/Outputs
        FOR phase IN computed.phases:
          phase_heading = get_phase_heading_line(phase)
          Edit(computed.skill_path,
            old_string=phase_heading,
            new_string=phase_heading + "\n\n" + boundary_annotation(phase))

      CASE "precondition_guards":
        # Insert guard block after the phase heading
        target = enhancement.target_phase
        phase_heading = get_phase_heading_line(target)
        Edit(computed.skill_path,
          old_string=first_content_after(phase_heading),
          new_string=guard_block + "\n\n" + first_content_after(phase_heading))

      CASE "pseudocode_blocks":
        # Insert pseudocode block into the target phase
        target = enhancement.target_phase
        insertion_point = find_best_insertion_point(target)
        Edit(computed.skill_path,
          old_string=insertion_point,
          new_string=insertion_point + "\n\n" + pseudocode_block)

      CASE "error_handling":
        # Insert error handling around tool calls in target phases
        FOR target IN enhancement.target_phases:
          tool_call_section = find_tool_call_section(target)
          Edit(computed.skill_path,
            old_string=tool_call_section,
            new_string=wrapped_with_error_handling(tool_call_section))

      CASE "checkpoint_patterns":
        # Insert checkpoint block at end of long-running phases
        FOR phase IN computed.phases:
          IF phase.tool_calls > 3 OR phase.prose_lines > 60:
            phase_end = get_phase_end_marker(phase)
            Edit(computed.skill_path,
              old_string=phase_end,
              new_string=checkpoint_block(phase) + "\n\n" + phase_end)

    # Track applied enhancement
    computed.enhancements_applied.append(enhancement.type)

    # Re-read the file after each edit to keep content in sync
    computed.skill_content = Read(computed.skill_path)

Step 5.3: Handle Apply Errors

HANDLE_APPLY_ERROR(enhancement, error):
  DISPLAY "Failed to apply enhancement '" + enhancement.type + "': " + error.message
  DISPLAY ""
  DISPLAY "This can happen when:"
  DISPLAY "  - The insertion point text was not found (file structure changed)"
  DISPLAY "  - The Edit tool could not find a unique match for old_string"
  DISPLAY ""
  DISPLAY "The enhancement content has been displayed above. You can apply it manually."
  # Continue with remaining enhancements — do not abort the batch

---

Phase 6: Verify

Re-analyze the modified skill and compare before/after metrics.

Step 6.1: Re-Analyze

Re-run the analysis from Phase 2 on the modified SKILL.md:

VERIFY():
  # Store pre-enhancement metrics
  computed.before_metrics = copy(computed.structure_summary)

  # Re-read and re-analyze
  computed.skill_content = Read(computed.skill_path)
  IDENTIFY_PHASES()       # Phase 2, Step 2.1
  ANALYZE_STRUCTURE()     # Phase 2, Step 2.2
  SUMMARIZE_STRUCTURE()   # Phase 2, Step 2.3

  computed.after_metrics = copy(computed.structure_summary)

Step 6.2: Compare and Report

COMPARE_METRICS():
  computed.improvement = {}

  metrics_to_compare = [
    "total_pseudocode_blocks",
    "unique_state_vars",
    "total_guards",
    "phases_with_input_spec",
    "phases_with_output_spec",
    "total_checkpoints",
    "total_error_handlers"
  ]

  FOR metric IN metrics_to_compare:
    before = computed.before_metrics[metric]
    after = computed.after_metrics[metric]
    computed.improvement[metric] = {
      before: before,
      after: after,
      delta: after - before
    }

Display the improvement summary:

## Enhancement Results: {computed.skill_name}

**Enhancements applied:** {join(computed.enhancements_applied, ", ")}
**Skill path:** {computed.skill_path}

### Before / After

| Metric | Before | After | Change |
|--------|--------|-------|--------|
| Pseudocode blocks | {before} | {after} | +{delta} |
| State variables | {before} | {after} | +{delta} |
| Precondition guards | {before} | {after} | +{delta} |
| Phases with input spec | {before}/{total} | {after}/{total} | +{delta} |
| Phases with output spec | {before}/{total} | {after}/{total} | +{delta} |
| Checkpoints | {before} | {after} | +{delta} |
| Error handlers | {before} | {after} | +{delta} |

{if all deltas == 0}
No measurable structural changes detected. Review the skill manually to
verify enhancements were applied correctly.
{else}
Structural improvements applied successfully.
{/if}

Step 6.3: Set Outputs

SET_OUTPUTS():
  outputs.enhancements_applied = computed.enhancements_applied
  outputs.updated_skill = len(computed.enhancements_applied) > 0

---

State Flow

Phase 1            Phase 2                  Phase 3                Phase 4              Phase 5            Phase 6
──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
computed.mode   -> computed.phases        -> computed              -> computed           -> computed         -> computed
computed          computed.current           .recommendations        .selected             .enhancements      .before_metrics
.skill_path       _structure               (with examples)           _enhancements         _applied          computed
computed          computed.structure                                computed                                  .after_metrics
.analyze_only     _summary                                          .application                             computed
computed          computed.frontmatter                               _plan                                    .improvement
.requested
_enhancement

---

Reference Documentation

• Phase Detection Algorithm: ${CLAUDE_PLUGIN_ROOT}/patterns/phase-detection-algorithm.md
• Complexity Scoring Algorithm: ${CLAUDE_PLUGIN_ROOT}/patterns/complexity-scoring-algorithm.md
• Quality Indicators: ${CLAUDE_PLUGIN_ROOT}/patterns/quality-indicators.md

---

Related Skills

• Assess skill coverage and fit: ${CLAUDE_PLUGIN_ROOT}/skills/bp-assess/SKILL.md
• Extract to workflows: ${CLAUDE_PLUGIN_ROOT}/skills/bp-extract/SKILL.md
• Build new skills: ${CLAUDE_PLUGIN_ROOT}/skills/bp-build/SKILL.md