bp-intent-create

name	bp-intent-create
description	This skill should be used when the user asks to "create intent mapping", "add intent detection", "generate intent-mapping.yaml", "define keyword routing", "set up 3VL matching", "create routing rules", or needs to create intent-mapping.yaml for a gateway. Triggers on "intent create", "create intent", "intent mapping", "keyword routing", "3VL setup", "routing rules", "intent detection".
allowed-tools	Read, Write, Glob, Grep, AskUserQuestion

Create Intent Mapping

Generate a complete intent-mapping.yaml for a gateway command by discovering skills, extracting keywords from descriptions, detecting overlap, designing flag categories, generating 3VL rules, and writing the final routing configuration.

Keyword Extraction Algorithm: patterns/keyword-extraction-algorithm.md Flag Generation Rules: patterns/flag-generation-rules.md Intent Mapping Template: ${CLAUDE_PLUGIN_ROOT}/templates/intent-mapping.yaml.template

---

Overview

This skill produces a complete intent-mapping.yaml file that enables a gateway command to route user input to the correct skill using three-valued logic (3VL) keyword matching. The process:

1. Discover all skills in the target plugin
2. Extract trigger keywords from each skill's description
3. Detect keyword overlap between skills
4. Design flag categories with user review
5. Generate 3VL rules mapping flag combinations to skills
6. Write the final intent-mapping.yaml file

Output contract: A valid intent-mapping.yaml at commands/{plugin-name}/intent-mapping.yaml containing intent_flags, intent_rules (with 3VL conditions), actions, and a fallback.

---

Phase 1: Discover Skills

Step 1.1: Determine Plugin Scope

Present the user with an option to select which plugin to generate intent mapping for:

{
  "questions": [{
    "question": "Which plugin should I create intent mapping for?",
    "header": "Plugin",
    "multiSelect": false,
    "options": [
      {
        "label": "Current plugin",
        "description": "Use the plugin at ${CLAUDE_PLUGIN_ROOT}"
      },
      {
        "label": "Specify path",
        "description": "I'll provide the path to a plugin directory"
      }
    ]
  }]
}

Response handling:

HANDLE_SCOPE(response):
  SWITCH response:
    CASE "Current plugin":
      computed.plugin_root = CLAUDE_PLUGIN_ROOT
    CASE "Specify path":
      # Ask user for the path via follow-up prompt
      computed.plugin_root = user_provided_path
      # Validate the path exists and contains .claude-plugin/plugin.json
      if not file_exists(computed.plugin_root + "/.claude-plugin/plugin.json"):
        WARN "No .claude-plugin/plugin.json found at this path. Are you sure this is a plugin?"

Extract the plugin name from computed.plugin_root:

# Read .claude-plugin/plugin.json to get the plugin name
plugin_manifest = Read(computed.plugin_root + "/.claude-plugin/plugin.json")
computed.plugin_name = plugin_manifest.name

Step 1.2: Glob for All SKILL.md Files

Search the plugin for SKILL.md files in both skills/ and skills/ directories:

FIND_SKILLS():
  computed.skill_files = []

  # Search skills/ directory (workflow-based skills)
  files = Glob(computed.plugin_root + "/skills/*/SKILL.md")
  computed.skill_files += files

  # Search skills/ directory (prose-based skills)
  files = Glob(computed.plugin_root + "/skills/*/SKILL.md")
  computed.skill_files += files

  # Deduplicate by absolute path
  computed.skill_files = deduplicate(computed.skill_files)

If computed.skill_files is empty:

No SKILL.md files found in this plugin. Verify the plugin has skills in skills/ or skills/ directories.

Then offer to retry with a different path or exit.

Step 1.3: Extract Frontmatter from Each Skill

For each discovered SKILL.md, read the file and extract the YAML frontmatter:

EXTRACT_METADATA():
  computed.skills = []

  FOR file IN computed.skill_files:
    content = Read(file.path)
    frontmatter = parse_yaml_frontmatter(content)

    skill = {
      name:        frontmatter.name,
      description: frontmatter.description,
      path:        file.path,
      directory:   parent_directory(file.path),
      tools:       split(frontmatter["allowed-tools"], ", ")
    }

    computed.skills.append(skill)

Display the discovered skills:

## Discovered Skills: {computed.plugin_name}

Found {len(computed.skills)} skills:

| # | Skill Name | Path |
|---|------------|------|
{for i, skill in enumerate(computed.skills)}
| {i+1} | {skill.name} | {skill.path} |
{/for}

Store in computed.skills[] with name, description, path, directory, and tools fields.

---

Phase 2: Extract Keywords

Step 2.1: Parse Trigger Phrases from Descriptions

For each skill in computed.skills, parse the description field to extract candidate keywords using the algorithm documented in patterns/keyword-extraction-algorithm.md.

function extract_keywords(description):
  keywords = []

  # 1. Extract quoted phrases: "create issue", "close PR"
  quoted = regex_find_all(description, /"([^"]+)"/)
  keywords.extend(quoted)

  # 2. Extract "Triggers on" section
  triggers_match = regex_find(description, /Triggers on\s+(.+?)\.?\s*$/)
  if triggers_match:
    trigger_items = split_and_trim(triggers_match.group(1), ",")
    # Strip surrounding quotes from each item
    trigger_items = [strip_quotes(item) for item in trigger_items]
    keywords.extend(trigger_items)

  # 3. Extract action verbs paired with their objects
  verb_phrases = regex_find_all(description,
    /\b(create|delete|update|find|show|list|analyze|convert|validate|generate|set up|add|remove|check|verify|scan|discover|build|rebuild)\s+\w+(\s+\w+)?/i)
  keywords.extend(verb_phrases)

  # 4. Extract standalone action verbs
  verbs = regex_find_all(description,
    /\b(create|delete|update|find|show|list|analyze|convert|validate|generate|setup|init|discover|scan|verify|check|build|rebuild|refresh|sync)\b/i)
  keywords.extend(verbs)

  # 5. Deduplicate preserving order, normalize to lowercase
  seen = set()
  unique = []
  for kw in keywords:
    normalized = kw.strip().lower()
    if normalized not in seen and len(normalized) > 1:
      seen.add(normalized)
      unique.append(normalized)

  return unique

Detail: See patterns/keyword-extraction-algorithm.md for the complete regex catalog, edge case handling, and priority ordering rules.

Apply extraction to each skill:

EXTRACT_ALL_KEYWORDS():
  FOR skill IN computed.skills:
    skill.raw_keywords = extract_keywords(skill.description)

Step 2.2: Present Extracted Keywords for Review

Display the extracted keywords grouped by skill so the user can verify:

## Extracted Keywords

{for skill in computed.skills}
### {skill.name}
Keywords: {", ".join(skill.raw_keywords)}

{/for}

Ask the user to confirm or adjust:

{
  "questions": [{
    "question": "Are the extracted keywords correct? You can adjust them in the next step.",
    "header": "Keywords",
    "multiSelect": false,
    "options": [
      {
        "label": "Keywords look good",
        "description": "Proceed to overlap detection"
      },
      {
        "label": "Edit keywords for a skill",
        "description": "I want to add, remove, or change keywords for one or more skills"
      }
    ]
  }]
}

Response handling:

HANDLE_KEYWORD_REVIEW(response):
  SWITCH response:
    CASE "Keywords look good":
      GOTO Phase 3
    CASE "Edit keywords for a skill":
      # Present follow-up listing each skill
      # For each skill the user wants to edit, ask for the new keyword list
      # Update computed.skills[].raw_keywords with the edited values
      # Redisplay the updated table and re-ask for confirmation

---

Phase 3: Detect Overlap

Step 3.1: Find Keywords Appearing in Multiple Skills

Build a keyword-to-skills index to identify keywords that appear across multiple skills:

function detect_overlap(skills):
  keyword_index = {}  # keyword -> [skill_names]

  FOR skill IN skills:
    FOR kw IN skill.raw_keywords:
      keyword_index.setdefault(kw, []).append(skill.name)

  # Find overlapping keywords (appear in 2+ skills)
  computed.overlap = {}
  FOR kw, skill_list IN keyword_index:
    IF len(skill_list) > 1:
      computed.overlap[kw] = skill_list

  return computed.overlap

Step 3.2: Assess Overlap Severity

For each skill, calculate the percentage of its keywords that overlap with other skills:

function assess_overlap_severity(skills, overlap):
  FOR skill IN skills:
    overlapping_count = count(kw for kw in skill.raw_keywords if kw in overlap)
    total_count = len(skill.raw_keywords)
    skill.overlap_pct = (overlapping_count / total_count * 100) if total_count > 0 else 0

    if skill.overlap_pct > 30:
      skill.needs_negative_keywords = true
    else:
      skill.needs_negative_keywords = false

If significant overlap exists (any skill has >30% shared keywords), recommend 3VL negative keywords for disambiguation:

## Keyword Overlap Analysis

{if len(computed.overlap) > 0}
**Overlapping keywords found:**

| Keyword | Skills |
|---------|--------|
{for kw, skill_list in computed.overlap.items()}
| "{kw}" | {", ".join(skill_list)} |
{/for}

**Per-skill overlap:**

| Skill | Overlap % | Needs Negative Keywords |
|-------|-----------|------------------------|
{for skill in computed.skills}
| {skill.name} | {skill.overlap_pct:.0f}% | {skill.needs_negative_keywords} |
{/for}

> **Recommendation:** Skills with >30% overlap should use negative keywords (3VL `F` conditions)
> to disambiguate. The flag generation phase will create appropriate negative_keywords entries.
{else}
No significant keyword overlap detected. Each skill has unique trigger phrases.
{/if}

Step 3.3: Display Overlap Matrix

If overlap exists, display a matrix showing which skills share keywords:

function display_overlap_matrix(skills, overlap):
  # Build skill-pair overlap counts
  pairs = {}
  FOR kw, skill_list IN overlap:
    FOR i, s1 IN enumerate(skill_list):
      FOR s2 IN skill_list[i+1:]:
        pair_key = (s1, s2)
        pairs.setdefault(pair_key, 0)
        pairs[pair_key] += 1

  # Display as table
  DISPLAY "### Overlap Matrix (shared keyword count)"
  # Render NxN matrix with skill names as row/column headers

---

Phase 4: Design Flag Categories

Step 4.1: Group Keywords into has_X Flags

Start with the standard flag set that every gateway should include, then add skill-specific flags.

Standard flags (always included):

Flag	Keywords	Purpose
has_help	"help", "how do i", "how to", "?", "explain", "guide", "what can", "show commands"	Natural language help
has_help_flag	"--help", "-h", "-?"	Explicit help flag
has_flags_help	"flags", "options", "arguments", "--verbose", "--quiet", "runtime flags"	Flag documentation
has_init	"create", "new", "initialize", "init", "start", "setup"	Create new things
has_modify	"add", "update", "edit", "change", "modify", "extend"	Modify existing things
has_query	"find", "search", "show", "list", "what", "where", "check", "status"	Query/inspect things
has_delete	"delete", "remove", "clear", "reset"	Remove things

Detail: See patterns/flag-generation-rules.md for the complete standard flag definitions, naming conventions, and when to include optional flags like has_logging_help, has_display_help, and has_prompts_help.

Skill-specific flag generation:

function generate_skill_flags(skills, overlap):
  computed.skill_flags = []

  FOR skill IN skills:
    # Derive a flag name from the skill name
    # Strip common prefixes (plugin name) to get the short ID
    skill_id = derive_short_id(skill.name)
    flag_name = "has_" + skill_id

    # Collect unique keywords (not already covered by standard flags)
    unique_kws = [kw for kw in skill.raw_keywords
                  if kw not in STANDARD_FLAG_KEYWORDS]

    # Generate negative keywords from overlapping skills
    negative_kws = []
    if skill.needs_negative_keywords:
      # For each overlapping keyword, find the OTHER skills that share it
      for kw in skill.raw_keywords:
        if kw in overlap:
          other_skills = [s for s in overlap[kw] if s != skill.name]
          # Add keywords unique to those other skills as negatives
          for other in other_skills:
            other_unique = get_unique_keywords(other, skills)
            negative_kws.extend(other_unique[:3])  # Top 3 distinguishing keywords

    negative_kws = deduplicate(negative_kws)

    computed.skill_flags.append({
      flag_name:         flag_name,
      skill_name:        skill.name,
      skill_id:          skill_id,
      keywords:          unique_kws,
      negative_keywords: negative_kws,
      description:       "Specific to " + skill.name
    })

Step 4.2: Generate Complete Flag Catalog

Combine standard flags with skill-specific flags into computed.all_flags:

ASSEMBLE_FLAGS():
  computed.all_flags = STANDARD_FLAGS + computed.skill_flags

Display the proposed flag catalog:

## Proposed Flag Categories

### Standard Flags
| Flag | Keywords | Description |
|------|----------|-------------|
{for flag in STANDARD_FLAGS}
| {flag.flag_name} | {", ".join(flag.keywords[:5])}... | {flag.description} |
{/for}

### Skill-Specific Flags
| Flag | Keywords | Negative Keywords | Description |
|------|----------|-------------------|-------------|
{for flag in computed.skill_flags}
| {flag.flag_name} | {", ".join(flag.keywords[:5])}... | {", ".join(flag.negative_keywords[:3]) or "none"} | {flag.description} |
{/for}

Step 4.3: Review Flag Design

Present the proposed flag design for user review and approval:

{
  "questions": [{
    "question": "Review the proposed flag categories. Any adjustments needed?",
    "header": "Flags",
    "multiSelect": false,
    "options": [
      {
        "label": "Looks good",
        "description": "Proceed with these flag categories"
      },
      {
        "label": "Add flag",
        "description": "I want to add a custom flag category"
      },
      {
        "label": "Remove flag",
        "description": "Some flags are unnecessary"
      },
      {
        "label": "Merge flags"
         
      
    
  

Response handling:

HANDLE_FLAG_REVIEW(response):
  SWITCH response:
    CASE "Looks good":
      GOTO Phase 5
    CASE "Add flag":
      # Ask for: flag name, keywords, description
      # Validate: name starts with "has_", uses snake_case
      # Append to computed.all_flags
      # Redisplay catalog and re-ask
    CASE "Remove flag":
      # Present list of non-standard flags as options
      # Remove selected flag from computed.all_flags
      # Warn if removing a skill-specific flag (skill will not be routable)
      # Redisplay catalog and re-ask
    CASE "Merge flags":
      # Present pairs of flags that share keywords
      # User selects which to merge
      # Combine keywords lists, use first flag's name
      # Remove the merged-away flag
      # Redisplay catalog and re-ask

---

Phase 5: Generate Rules

Step 5.1: Map Flag Combinations to Skills

Generate 3VL rules that map flag combinations to skill actions. Rules follow these patterns:

Pure single intents -- one skill flag true, competing flags false:

function generate_pure_rules(skill_flags):
  rules = []
  FOR flag IN skill_flags:
    conditions = {flag.flag_name: "T"}
    # Set competing skill flags to F for disambiguation
    FOR other_flag IN skill_flags:
      if other_flag.flag_name != flag.flag_name:
        # Only add F condition if there is keyword overlap
        if has_overlap(flag, other_flag):
          conditions[other_flag.flag_name] = "F"

    rules.append({
      name:        flag.skill_id + "_only",
      conditions:  conditions,
      action:      "delegate_" + flag.skill_id,
      description: "Route to " + flag.skill_name
    })

  return rules

Compound intents -- help combined with a skill-specific flag:

function generate_help_rules(skill_flags):
  rules = []
  FOR flag IN skill_flags:
    rules.append({
      name:        "help_with_" + flag.skill_id,
      conditions:  {"has_help": "T", flag.flag_name: "T"},
      action:      "show_skill_help_" + flag.skill_id,
      description: "Help with " + flag.skill_name
    })
  return rules

Action type compound rules -- action type flag combined with skill flag:

function generate_action_compound_rules(skill_flags):
  rules = []
  # Only generate these if skills map to action types
  # E.g., has_init + has_setup -> delegate_setup
  # These are generated when skill keywords overlap with action type keywords
  FOR flag IN skill_flags:
    for action_flag in ["has_init", "has_modify", "has_query", "has_delete"]:
      if shares_keywords(flag, action_flag):
        rules.append({
          name:        action_flag.replace("has_", "") + "_" + flag.skill_id,
          conditions:  {action_flag: "T", flag.flag_name: "T"},
          action:      "delegate_" + flag.skill_id,
          description: action_flag.replace("has_", "").title() + " via " + flag.skill_name
        })
  return rules

Step 5.2: Set Priorities

Assign priorities to rules following the standard ordering:

function assign_priorities(all_rules):
  # Priority tiers (lower number = higher priority, evaluated first)
  PRIORITY_TIERS = {
    "explicit_help_flag":  100,  # --help/-h always wins
    "help_with_flags":      90,  # Runtime flag docs
    "help_with_*":          80,  # Skill-specific help
    "general_help":         70,  # Catch-all help
    "*_only":               10,  # Pure single-intent skill routing
    "action_compound":      15,  # Action type + skill compound
    "show_menu":             0   # Fallback (always last)
  }

  FOR rule IN all_rules:
    # Match rule name against priority patterns
    if rule.name == "explicit_help_flag":
      rule.priority = 100
    elif rule.name == "help_with_flags":
      rule.priority = 90
    elif rule.name.startswith("help_with_"):
      rule.priority = 80
    elif rule.name == "general_help":
      rule.priority = 70
    elif rule.name.endswith("_only"):
      rule.priority = 10
    else:
      rule.priority = 15

  # Sort by priority descending (highest priority first in rule list)
  all_rules.sort(key=lambda r: r.priority, reverse=True)

Step 5.3: Configure Fallback

Every intent mapping must have a fallback rule that activates when no other rule matches:

function generate_fallback(plugin_name, skills):
  computed.fallback_rule = {
    name:        "show_menu",
    conditions:  {},  # Empty conditions = always matches
    action:      "show_main_menu",
    description: "No clear intent, show interactive menu"
  }

  computed.fallback_action = {
    type: "user_prompt",
    prompt: {
      question: "What would you like to do with " + plugin_name + "?",
      header: "Menu",
      options: [
        {
          id:          skill_flag.skill_id,
          label:       skill_flag.skill_name,
          description: truncate(skill_flag.description, 80)
        }
        for skill_flag in computed.skill_flags
      ]
    }
  }

Store the assembled rule set in computed.all_rules (including help rules, skill rules, and the fallback).

---

Phase 6: Generate File

Step 6.1: Load Template

Load the intent-mapping.yaml template from the plugin's template directory:

LOAD_TEMPLATE():
  template_path = CLAUDE_PLUGIN_ROOT + "/templates/intent-mapping.yaml.template"
  computed.template = Read(template_path)

  if computed.template is empty or null:
    WARN "Template not found at " + template_path
    # Fall back to generating from scratch without template
    computed.use_template = false
  else:
    computed.use_template = true

Step 6.2: Populate Template

Replace template placeholders with computed values:

POPULATE_TEMPLATE():
  if computed.use_template:
    output = computed.template

    # Replace simple placeholders
    output = replace(output, "{{plugin_name}}", computed.plugin_name)
    output = replace(output, "{{PLUGIN_TITLE}}", computed.plugin_name.upper().replace("-", " "))
    output = replace(output, "{{plugin_description}}", computed.plugin_description or "")

    # Replace iterable sections (Mustache-style)
    # {{#skill_flags}} ... {{/skill_flags}}
    output = expand_section(output, "skill_flags", computed.skill_flags)
    output = expand_section(output, "skill_help_rules", computed.skill_flags)
    output = expand_section(output, "skill_rules", computed.all_rules)
    output = expand_section(output, "skills", computed.skills)
    output = expand_section(output, "skill_help_actions", computed.skill_flags)
    output = expand_section(output, "examples", computed.examples or [])

  else:
    # Generate YAML directly from computed state
    output = generate_yaml_from_state()

If not using the template, generate the YAML structure directly:

function generate_yaml_from_state():
  yaml = "# Intent mapping for " + computed.plugin_name + " gateway\n"
  yaml += "# Generated by bp-intent-create\n\n"

  # Emit intent_flags
  yaml += "intent_flags:\n"
  FOR flag IN computed.all_flags:
    yaml += "  " + flag.flag_name + ":\n"
    yaml += "    keywords:\n"
    FOR kw IN flag.keywords:
      yaml += '      - "' + kw + '"\n'
    IF flag.negative_keywords:
      yaml += "    negative_keywords:\n"
      FOR nkw IN flag.negative_keywords:
        yaml += '      - "' + nkw + '"\n'
    yaml += '    description: "' + flag.description + '"\n\n'

  # Emit intent_rules
  yaml += "intent_rules:\n"
  FOR rule IN computed.all_rules:
    yaml += "  - name: " + '"' + rule.name + '"\n'
    yaml += "    conditions:\n"
    FOR flag_name, value IN rule.conditions:
      yaml += "      " + flag_name + ": " + value + "\n"
    yaml += "    action: " + rule.action + "\n"
    yaml += '    description: "' + rule.description + '"\n\n'

  # Emit fallback
  yaml += "  # Default fallback\n"
  yaml += '  - name: "show_menu"\n'
  yaml += "    conditions: {}\n"
  yaml += "    action: show_main_menu\n"
  yaml += '    description: "No clear intent, show interactive menu"\n\n'

  # Emit actions
  yaml += "actions:\n"
  FOR skill_flag IN computed.skill_flags:
    yaml += "  delegate_" + skill_flag.skill_id + ":\n"
    yaml += "    type: invoke_skill\n"
    yaml += '    skill: "' + skill_flag.skill_name + '"\n'
    yaml += "    pass_arguments: true\n\n"

  # Emit show_main_menu action
  yaml += "  show_main_menu:\n"
  yaml += "    type: user_prompt\n"
  yaml += "    prompt:\n"
  yaml += '      question: "What would you like to do with ' + computed.plugin_name + '?"\n'
  yaml += '      header: "Menu"\n'
  yaml += "      options:\n"
  FOR skill_flag IN computed.skill_flags:
    yaml += "        - id: " + skill_flag.skill_id + "\n"
    yaml += '          label: "' + skill_flag.skill_name + '"\n'
    yaml += '          description: "' + truncate(skill_flag.description, 60) + '"\n'

  return yaml

Step 6.3: Write to Output Path

Determine the output path and write the generated file:

WRITE_OUTPUT():
  output_path = computed.plugin_root + "/commands/" + computed.plugin_name + "/intent-mapping.yaml"
  output_dir = parent_directory(output_path)

  # Check if output directory exists
  if not directory_exists(output_dir):
    WARN "Directory " + output_dir + " does not exist. Creating it."
    # The Write tool will create parent directories

  # Check if file already exists
  if file_exists(output_path):
    # Present overwrite options
    ASK_USER:
      question: "intent-mapping.yaml already exists at this path. What should I do?"
      options:
        - "Overwrite": Replace the existing file
        - "Backup and replace": Rename existing to intent-mapping.yaml.bak, then write new
        - "Write to alternate path": I'll specify a different output path
        - "Cancel": Do not write the file

    HANDLE response:
      "Overwrite":          # Proceed with Write
      "Backup and replace": # Read existing, Write to .bak, then Write new
      "Write to alternate": # Ask for path, use that instead
      "Cancel":             # Display the generated YAML in terminal and exit

  Write(output_path, computed.generated_yaml)

Step 6.4: Display Summary and Offer Validation

After writing, display a summary of what was generated:

## Intent Mapping Generated

**File:** {output_path}
**Plugin:** {computed.plugin_name}

### Statistics
- **Flags:** {len(computed.all_flags)} total ({len(STANDARD_FLAGS)} standard + {len(computed.skill_flags)} skill-specific)
- **Rules:** {len(computed.all_rules)} total ({count_help_rules} help + {count_skill_rules} skill + 1 fallback)
- **Skills routed:** {len(computed.skill_flags)}

### Flag Summary
| Flag | Keywords | Negatives |
|------|----------|-----------|
{for flag in computed.all_flags}
| {flag.flag_name} | {len(flag.keywords)} | {len(flag.negative_keywords)} |
{/for}

### Rule Summary
| Rule | Conditions | Action |
|------|------------|--------|
{for rule in computed.all_rules}
| {rule.name} | {format_conditions(rule.conditions)} | {rule.action} |
{/for}

Offer validation:

{
  "questions": [{
    "question": "Would you like to validate the generated intent mapping?",
    "header": "Validate",
    "multiSelect": false,
    "options": [
      {
        "label": "Validate now",
        "description": "Check for rule conflicts, unreachable skills, and missing coverage"
      },
      {
        "label": "Done",
        "description": "Intent mapping is complete"
      }
    ]
  }]
}

Response handling:

HANDLE_VALIDATION(response):
  SWITCH response:
    CASE "Validate now":
      GOTO validation subroutine (inline)

      # Check 1: Every skill has at least one rule that routes to it
      routed_skills = set(rule.action.replace("delegate_", "") for rule in computed.all_rules)
      unrouted = [sf.skill_id for sf in computed.skill_flags if sf.skill_id not in routed_skills]
      if unrouted:
        WARN "Skills with no routing rule: " + ", ".join(unrouted)

      # Check 2: No two rules have identical conditions
      condition_sets = [frozenset(rule.conditions.items()) for rule in computed.all_rules]
      duplicates = find_duplicates(condition_sets)
      if duplicates:
        WARN "Duplicate rule conditions found: " + str(duplicates)

      # Check 3: Fallback exists
      has_fallback = any(rule.conditions == {} for rule in computed.all_rules)
      if not has_fallback:
        WARN "No fallback rule with empty conditions"

      # Check 4: Help rules exist for each skill
      help_rules = [r for r in computed.all_rules if r.name.startswith("help_with_")]
      if len(help_rules) < len(computed.skill_flags):
        WARN "Not all skills have help rules"

      if no warnings:
        DISPLAY "Validation passed. All skills are routed, no conflicts detected."

    CASE "Done":
      DISPLAY "Intent mapping generation complete."
      EXIT

---

State Flow

Phase 1                    Phase 2                  Phase 3
────────────────────────────────────────────────────────────────
computed.plugin_root    -> computed.skills[]      -> computed.overlap
computed.plugin_name       .raw_keywords             .overlap_pct
computed.skill_files                                  .needs_negative_keywords

Phase 4                    Phase 5                  Phase 6
────────────────────────────────────────────────────────────────
computed.all_flags      -> computed.all_rules     -> computed.generated_yaml
computed.skill_flags       .priority                 output_path
                           computed.fallback_rule    (written file)
                           computed.fallback_action

---

Reference Documentation

• Keyword Extraction Algorithm: patterns/keyword-extraction-algorithm.md (local to this skill)
• Flag Generation Rules: patterns/flag-generation-rules.md (local to this skill)
• Intent Mapping Template: ${CLAUDE_PLUGIN_ROOT}/templates/intent-mapping.yaml.template
• Workflow Generation Pattern: ${CLAUDE_PLUGIN_ROOT}/patterns/authoring-guide.md
• Node Mapping Pattern: ${CLAUDE_PLUGIN_ROOT}/lib/patterns/node-mapping.md

---

Related Skills

• Gateway command generation: ${CLAUDE_PLUGIN_ROOT}/skills/bp-gateway-create/SKILL.md
• Plugin discovery and classification: ${CLAUDE_PLUGIN_ROOT}/skills/bp-plugin-discover/SKILL.md
• Workflow validation: ${CLAUDE_PLUGIN_ROOT}/skills/bp-skill-validate/SKILL.md
• Prose skill analysis: ${CLAUDE_PLUGIN_ROOT}/skills/bp-skill-analyze/SKILL.md