map-controls

Map Controls

Run an interactive mapping session for a control file. Walks through each unmapped requirement, finds candidate rules via cross-framework search and AI suggestions, lets the author select rules, and writes them back to the control file.

This skill orchestrates /inspect-control for setup and /map-requirement logic for each requirement.

Tool Strategy

This skill uses mcp__content-agent__* tools when available (preferred — deterministic, structured results). When the MCP server is not configured, fall back to filesystem-based alternatives noted as Fallback in each step. See .claude/skills/shared/mcp_fallbacks.md for detailed fallback procedures. The skill must complete successfully either way.

Without MCP server: Cross-framework similarity search is unavailable. Candidate rules will be found by keyword search only, which may miss semantically similar but differently-worded requirements.

Arguments: $ARGUMENTS — format: <control_id> [--product <product_id>] [--policy <path>]

Examples:

• /map-controls anssi --product rhel9
• /map-controls srg_gpos --product rhel10
• /map-controls hipaa
• /map-controls anssi --product rhel9 --policy security_policies/anssi_2.md

Phase 1: Setup and Validation

1. Parse arguments: Extract control_id, optional --product, and optional --policy from $ARGUMENTS.

2. Load control stats: Call mcp__content-agent__get_control_stats with control_id.

   • If not found, call mcp__content-agent__list_controls and present options via AskUserQuestion.
   • Fallback: Read controls/<control_id>.yml or products/**/controls/<control_id>.yml. Count requirements by status manually. List controls with ls controls/*.yml and ls products/*/controls/*.yml.

3. If no --product specified, discover available products and ask user via AskUserQuestion:

   • Run ls products/ (or call mcp__content-agent__list_products) to get the list of available products
   • "Which product are you mapping rules for?"
   • Options: present the most common products from the discovered list + Other

4. Check build artifacts: Call mcp__content-agent__list_built_products.

   • If the target product is NOT in the built list, ask the user via AskUserQuestion:
     • "Product '{product}' has not been built yet. Rule search works best with build artifacts (expanded Jinja templates). Build it now?"
     • Options:
       • "Yes, build now (Recommended)" — description: "Runs /build-product {product} --datastream-only before starting the mapping session"
       • "No, continue without build" — description: "Rule search will use raw source files, which may miss some rules due to Jinja template parsing errors"
   • If user chooses to build, invoke the build-product skill: Skill(skill="build-product", args="{product} --datastream-only"). Wait for the build to complete before continuing.
   • Fallback: Check if build/{product}/rules/ directory exists. If not, offer the build prompt above.

5. Report summary:

   ## Control File: {title} ({control_id})
   
   | Status    | Count |
   |-----------|-------|
   | Total     | {total} |
   | Mapped    | {mapped} |
   | Unmapped  | {unmapped} |
   

   Show the by_status breakdown, only including statuses with non-zero counts.

6. If no --policy specified, inform the user:

   Tip: You can enrich mapping with the original security policy document by adding --policy <path> (supports PDF, Markdown, HTML). This provides full policy context for each requirement, improving cross-framework matching accuracy.

7. Get work queue: Call mcp__content-agent__list_unmapped_requirements with control_id. Fallback: Read the control YAML and filter requirements where status is pending or rules: is empty/missing.

8. Ask user scope via AskUserQuestion:

   • "Which requirements do you want to work on?"
   • Options:
     • "Unmapped only ({N} requirements)" — work through the pending/unmapped queue
     • "All requirements ({total})" — review every requirement including already-mapped ones
     • "Specific requirement IDs" — let user type specific IDs

Phase 2: Per-Requirement Mapping Loop

Note: Do NOT call get_control_rule_index here — it returns a massive payload (1.6MB+) that wastes context. The per-requirement find_similar_requirements call in Step 2b already provides cross-framework discovery for each requirement individually.

For each requirement in the selected work queue, execute the mapping workflow. This follows the same logic as /map-requirement but in batch.

For each requirement:

Step 2a: Present the Requirement

Display:

---
### [{current}/{total}] Requirement: {requirement_id}
**Title**: {title}
**Description**: {description}
**Current status**: {status}
**Current rules**: {rules or "none"}
---

If --policy was provided:

1. Infer document type from the file extension (.md → markdown, .pdf → pdf, .html → html, otherwise → text)
2. Call mcp__content-agent__parse_policy_document with source, document_type, and requirement_id for the current requirement. Fallback: For markdown/text, read the file and search for the requirement ID. For PDF, inform user that PDF parsing requires the MCP server.
3. If sections returned, display the policy context:

   ### Policy Context (from {policy_path})
   **{section_title}**
   {section_content}
   

4. Store the combined section text as policy_text for use in Step 2b

Step 2b: Cross-Framework Search

Call mcp__content-agent__find_similar_requirements with:

• requirement_text: if policy_text is available, use it; otherwise use the requirement's title + " " + description
• exclude_control_id: current control_id
• max_results: 10

Fallback: Extract 3-5 key terms from the requirement text. Use Grep to search for each term across controls/*.yml and products/*/controls/*.yml. Read matched requirements to extract their rules: lists.

If results found, present grouped by framework:

Similar requirements in other frameworks:
- [{control_id}] {req_id}: "{title}" → rules: {rules}

Extract the union of all rules as cross-framework candidates.

Step 2c: Rule Search in Build Artifacts

Search for candidate rules using rendered build artifacts (Jinja-expanded, product-specific):

1. Extract 3-5 key terms from the requirement title and description.

2. For each key term, call mcp__content-agent__search_rendered_content with:

   • query: the key term
   • product: the target product from Phase 1
   • limit: 15
   • Fallback: If the product is not built, use Grep to search for key terms in rule.yml files under linux_os/guide/ and applications/. This may miss rules with Jinja-templated descriptions.

3. Deduplicate results across all term searches. Combine with cross-framework candidates from Step 2b.

4. For each candidate rule, use mcp__content-agent__get_rendered_rule (or get_rule_details) to read its title and description. Reason about which rules best match the requirement semantically.

5. Present the top candidates in a table:

   ### Rule Candidates
   
   | Rule ID | Title | Source | Reasoning |
   |---------|-------|--------|-----------|
   | {rule_id} | {title} | cross-ref / search | {why it matches} |
   

Step 2d: Product Availability Check

Since search_rendered_content only returns rules present in the target product's build, all search results are already confirmed available. For cross-framework candidates (from Step 2b) that did NOT appear in the rendered search, call mcp__content-agent__get_rule_product_availability to verify availability. Fallback: Check each rule's rule.yml for cce@<product> entries and grep for the rule ID in target product profiles/controls.

Flag rules NOT available for the target product. Assess portability:

• Templated rules are likely portable
• Check platform constraints if present

Step 2e: Author Decision

1. Build unified candidate list: combine cross-framework and AI suggestions, deduplicated, sorted by confidence.

2. Use AskUserQuestion with multiSelect: true:

   • "Select rules for '{requirement_id}: {title}'"
   • Options: top candidates (up to 4), with confidence and source info
   • If more than 4 candidates, present top 3 + "Show more candidates"

3. Use a separate AskUserQuestion:

   • "How should this requirement be marked?"
   • Options:
     • "Automated" — fully covered
     • "Partially automated" — partially covered
     • "Not applicable" — doesn't apply
     • "Create new rule" — will be collected for gap analysis
     • "Skip for now"

Step 2f: Write Selection

• If rules selected: Call mcp__content-agent__update_requirement_rules with control_id, requirement_id, rules, and appropriate status. Fallback: Find and edit the requirement's YAML file directly using Edit tool.
• If not applicable: Call mcp__content-agent__update_requirement_rules with empty rules and status="not_applicable". Fallback: Edit the requirement's YAML file to set status: not_applicable and rules: [].
• If create new rule: Add to the "new rules needed" list for Phase 3.
• If skipped: Record as skipped, move to next.

Track all changes for the final report.

Phase 3: Gap Analysis

For requirements where the author chose "Create new rule":

1. For each gap requirement, use LLM analysis to:

   • Decompose the requirement into atomic checks (one config change per rule)
   • Suggest rule IDs following naming conventions (lowercase, underscores)
   • Suggest which template might apply (call mcp__content-agent__list_templates and match; Fallback: ls shared/templates/)
   • Suggest severity based on the requirement

2. Present the decomposition:

   ### Gap: {requirement_id} — {title}
   
   This requirement could be covered by:
   1. {suggested_rule_id} (template: {template_name}) — {description}
   2. {suggested_rule_id} (template: {template_name}) — {description}
   

3. Ask user which rules to create via AskUserQuestion.

4. For each approved new rule, tell the user to run /create-rule <rule_id> with the suggested parameters after this session completes.

Phase 4: Final Report

Present a complete summary of the mapping session:

## Mapping Session Complete

**Control file**: {control_id} ({title})
**Target product**: {product}

### Changes Made
- {X} requirements mapped to existing rules
- {Y} requirements marked not applicable
- {Z} requirements skipped

### Rules Added
| Requirement | Rules | Status |
|-------------|-------|--------|
| {req_id}    | {rules} | automated |
| ...         | ...   | ... |

### New Rules Needed
| Requirement | Suggested Rules | Template |
|-------------|----------------|----------|
| {req_id}    | {rule_ids}     | {template} |

### Updated Stats

Call mcp__content-agent__get_control_stats again to show the updated coverage. Fallback: Re-read the control YAML and recount requirements by status.

| Status   | Before | After |
|----------|--------|-------|
| Mapped   | {before} | {after} |
| Unmapped | {before} | {after} |
| Coverage | {before}% | {after}% |

### Next Steps
1. Review changes: `git diff controls/`
2. Build product: `/build-product {product}`
3. Create new rules: `/create-rule <rule_id>` for each gap
4. Test rules: `/test-rule <rule_id>`
5. Draft PR: `/draft-pr`

Important Notes

• The update_requirement_rules tool replaces existing rules — if a requirement already has rules and the author wants to add more, include the existing rules in the selection.
• AI suggestions require API key — if suggest_rule_mappings fails, fall back to cross-framework search only.
• Large control files (e.g., srg_gpos with 300+ requirements) — consider processing in batches. After every 10 requirements, ask if the user wants to continue or pause.
• Track before/after stats — call get_control_stats at the start and end to show progress.
• Don't overwhelm the user — if cross-framework search and AI both return many candidates, focus on the top 5-10 with highest confidence.