How the Main Router in SKILL.md Directs Patent-Disclosure Skill Requests

The main router in SKILL.md acts as a policy gatekeeper that inspects user commands, validates explicit sub-skill targeting, enforces package isolation, and dispatches requests to the appropriate sub-skill's own SKILL.md file.

The SKILL.md file at the root of the handsomestWei/patent-disclosure-skill repository serves as the central routing engine for the entire patent-disclosure skill suite. Understanding how this main router in SKILL.md works is essential for extending the system or integrating it with client applications. The router implements a strict six-step verification flow that guarantees predictable behavior and prevents cross-skill contamination.

Intent Recognition and Capability Matching

The router begins by inspecting the user's intent against a capability table defined in the root SKILL.md (lines 16-22). Each row in this table maps a specific trigger phrase to a target sub-skill entry file.

When a request arrives with the Read intent, the router matches the command string against trigger phrases such as /patent-disclosure, /patent-application, /patent-search, /patent-docket, /patent-oa, or /patent-exam-policy. Upon finding a match, the router identifies the corresponding entry SKILL.md path (e.g., skills/patent-disclosure/SKILL.md) that should handle the request.

Explicit Naming Requirements for High-Level Workflows

For complex workflows including application file generation, docket orchestration, policy brief creation, and Office Action (OA) responses, the router enforces a strict explicit naming rule (lines 27-33 in SKILL.md).

If the user invokes a high-level command without explicitly naming the target sub-skill, the router aborts the flow immediately. Instead of proceeding with ambiguous context, it returns a prompt requiring the user to specify which sub-skill they intend to invoke. This prevents accidental execution of resource-intensive workflows like full patent application generation.

Read-Only Shortcuts for Patent Documents

The router implements a priority shortcut for read-only flows (lines 28-30). When the input consists of a patent number (e.g., CN1234567A) or a PDF file reference, and the intent is classified as "read", the router bypasses the standard disclosure workflow entirely.

In this scenario, the request is immediately forwarded to skills/patent-reader/SKILL.md without traversing the disclosure steps. This optimization ensures that document summarization and analysis tasks execute with minimal latency and without triggering unnecessary disclosure protocols.

Cross-Package Isolation Guards

To maintain clean architectural boundaries, the router enforces a cross-package call prohibition (lines 26-27). The system bans any import of tools or utilities from sibling sub-skill packages. Each sub-skill must rely exclusively on its own local copies of prompts, tools, and helper functions.

This isolation guard prevents side effects where one sub-skill's configuration might inadvertently alter another's behavior. When the router dispatches to skills/patent-application/SKILL.md, that sub-skill operates in a sealed environment using only its bundled resources.

Dispatch and Post-Routing Validation

Once the router validates the command and confirms explicit naming (where required), it performs a Read operation on the selected sub-skill's SKILL.md file. Control then transfers completely to that sub-skill, which manages its own detailed workflow—whether that involves the eight-step disclosure process, four-document application bundle generation, or docket orchestration.

After the sub-skill completes execution, the router performs post-routing checks (lines 48-53). It verifies that all required outputs are written to the designated outputs/ subdirectory (e.g., outputs/patent-application/, outputs/docket/) and confirms that no prohibited side-effects have occurred outside the skill's sandbox.

Practical Routing Examples

The following Python examples demonstrate how client applications interact with the router's decision flow:

# Triggering the full Disclosure workflow

request = {
    "intent": "Read",
    "command": "/patent-disclosure",
    "args": {"project_path": "/my/patent_project"}
}
router.handle(request)  # Dispatches to skills/patent-disclosure/SKILL.md

# Direct read of a patent PDF bypasses disclosure steps

request = {
    "intent": "Read",
    "command": "CN1234567A",
    "args": {}
}
router.handle(request)  # Shortcuts directly to skills/patent-reader/SKILL.md

# Blocked request due to missing explicit naming

request = {
    "intent": "Read",
    "command": "/policy-brief"
}
router.handle(request)  # Router aborts and prompts for explicit skill name

Sub-Skill Architecture and File Structure

The router manages dispatch to seven specialized sub-skills, each contained in its own directory under skills/:

• skills/patent-disclosure/SKILL.md – Handles the end-to-end disclosure process (steps 1-8)
• skills/patent-application/SKILL.md – Generates the four-document application bundle (claims, specification, abstract, drawings)
• skills/patent-docket/SKILL.md – Orchestrates full docket workflows (disclosure through application) in a single pass
• skills/patent-search/SKILL.md – Performs bibliographic searches and constructs search queries
• skills/patent-reader/SKILL.md – Reads patent PDFs and produces plain-language summaries
• skills/patent-oa/SKILL.md – Supports Office Action response drafting
• skills/patent-exam-policy/SKILL.md – Generates policy briefs and examination policy updates

Each sub-skill maintains its own isolated SKILL.md that defines granular workflow steps, while the root SKILL.md maintains the high-level routing matrix that binds trigger commands to these files.

Summary

• The main router in SKILL.md implements a six-step verification flow: intent matching, explicit naming enforcement, read-only shortcut evaluation, cross-package isolation checks, sub-skill dispatch, and post-execution validation.
• High-level workflows require explicit sub-skill naming; ambiguous commands are rejected with clarification prompts.
• Patent numbers and PDF inputs trigger immediate routing to the Reader skill, bypassing disclosure workflows entirely.
• Strict package isolation prevents sub-skills from importing tools from sibling directories, ensuring clean architectural boundaries.
• All sub-skills route through dedicated SKILL.md files located in skills/patent-{name}/ directories, with outputs strictly confined to outputs/patent-{name}/ subdirectories.

Frequently Asked Questions

What happens if I invoke a patent command without specifying the sub-skill name?

The router blocks the execution and returns a prompt asking you to explicitly name the target sub-skill. According to the routing logic in SKILL.md (lines 27-33), high-level workflows such as application generation or policy brief creation require unambiguous targeting to prevent accidental resource consumption.

Can sub-skills share utility functions or tools across packages?

No. The router explicitly bans cross-package imports (lines 26-27 in SKILL.md). Each sub-skill must maintain its own local copy of utilities, prompts, and tools. This design prevents side effects and ensures that modifications to one sub-skill cannot inadvertently break others.

How does the router handle patent numbers versus workflow commands?

When the router detects a patent number (e.g., CN1234567A) or PDF reference with a Read intent, it immediately shortcuts to skills/patent-reader/SKILL.md without invoking the disclosure workflow. This read-only optimization (lines 28-30) ensures document analysis occurs with minimal overhead compared to full disclosure processing.

Where does the router store outputs after a sub-skill completes execution?

The router enforces a strict output convention requiring all generated files to reside in outputs/patent-{subskill}/ directories (lines 48-53). For example, application documents must appear under outputs/patent-application/, while docket outputs belong in outputs/docket/. The router validates these locations during post-routing checks.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}