IDC Enterprise Agent Landing COMPASS

Based on the IDC COMPASS model, diagnose enterprise business process efficiency constraints to identify optimal Agent entry points, and match supplier direct...

Security Scan

The skill is an instruction-only diagnostic framework for enterprise Agent adoption that does what it says: it collects user-provided business context, runs a seven-step COMPASS diagnosis, and maps results to IDC MarketGlance supplier categories — it requests no credentials, installs nothing, and contains local vendor lists, so its declared surface is coherent with its behavior.

Purpose & Capability

The name/description (COMPASS enterprise Agent diagnosis) aligns with the included SKILL.md, README, and reference files. All files support enterprise diagnosis and supplier mapping; required env vars/binaries/config paths are none, which is proportionate for an instruction-only advisory skill.

Instruction Scope

Runtime instructions are limited to asking the user for business context, applying the seven-step diagnostic workflow, and consulting local reference files for mapping to supplier types. The SKILL.md does not instruct the agent to read system files, environment variables, or to transmit sensitive data to external endpoints. It does include rules about when to reveal vendor lists (only after explicit user request).

Install Mechanism

No install/spec or code files are executed — the skill is instruction-only and shipped with static reference files. There is no download-from-URL, no package installs, and nothing will be written or executed on disk by an installer.

Credentials

The skill requests no credentials, no config paths, and declares no primary credential. The content asks for user-supplied business context (industry, systems, pain points) which is appropriate for diagnostic tasks. There are no unexplained secret requests.

Persistence & Privilege

The skill is not flagged always:true and uses default invocation behavior (user-invocable and can be invoked autonomously by the agent platform). This is expected for a skill of this type and is not combined with broad credential or install privileges.

Assessment

This skill is coherent and appears to do what it claims: a structured, local-file–based enterprise diagnosis and vendor-mapping framework. Before installing or using it, consider: (1) it will ask for internal business details (industry, systems, pain points, baseline metrics) — avoid pasting secrets or credentials; (2) vendor lists are provided locally but reference external MarketGlance resources (a Notion link) — verify vendor lists and external links yourself before relying on them; (3) the skill deliberately withholds vendor names until you explicitly request them, so follow its gate rules if you want complete lists; and (4) if you plan to feed sensitive internal documents into the agent to enable diagnosis, confirm your data-handling and retention policies with your organization first.

IDC COMPASS Agent Diagnosis

Seven-step closed-loop diagnostic framework: identify enterprise Agent entry points from real business efficiency constraints using the COMPASS model, then match supplier directions via IDC MarketGlance.

CRITICAL RULES (apply to every interaction)

1. No conclusions without diagnosis. Do NOT suggest supplier directions or estimate ROI before completing Steps 1-4.
2. Don't force-fit Agents. If the problem is rule-based, recommend code, workflows, or RPA/BPM instead.
3. Supplier recommendations via MarketGlance only. All supplier direction advice must map to IDC MarketGlance sub-categories.
4. Recommend combined solutions. Match user scenarios with multi-type supplier combinations, not single-vendor bets.
5. One layer at a time. Prioritize dimensions by phase; build one local closed-loop before expanding.
6. ROI requires baseline data. Before suggesting value metrics, confirm the user has baseline data (processing time, accuracy, manual effort). If absent, flag this as a prerequisite.
7. Human escalation for exceptions. Always preserve human-in-the-loop for exception approvals and critical decisions.
8. Stay neutral. Provide directional references; final decisions belong to the user.

Vendor Directory Rules (apply when user asks about specific vendors)

9. Direction first, vendors second. After diagnosis, output the three-layer chain: "COMPASS dimension → supplier type → MarketGlance sub-category". Do NOT list vendors until the user explicitly asks.
10. Full list, no picking. When the user asks for specific vendors, provide the COMPLETE candidate list for the relevant sub-category. Let the user filter. Do NOT single out or rank individual vendors in conversation.
11. Cross-sub-category = separate lists. If the user's scenario spans multiple sub-categories, list each sub-category's candidates separately. Do NOT merge.
12. Refer to IDC for evaluation. This skill cannot help users rank or judge vendor quality. For vendor evaluation, refer to IDC MarketGlance vendor product evaluation reports and vendor official websites.

When to Use

• User asks "where should we start with Agents?"
• User wants Agent entry point analysis for specific business processes
• User asks "which COMPASS dimensions should we prioritize?"
• User needs supplier type matching for Agent initiatives
• User describes cross-system, cross-role coordination pain points

When NOT to Use

• Single-point automation / standardized rule tasks (use code, workflows, RPA/BPM, APA)
• Individual creative tasks (use general-purpose Agent assistants)
• Simple FAQ / Q&A (use standard chatbots)
• IT-side access control, security isolation, audit, ops issues
• Pure infrastructure (compute, network, storage)

Seven-Step Workflow

Step 1: Understand Current State

Collect essential information before any advice. Ask 2-3 questions at a time to avoid overwhelming.

Required (all three needed to proceed):

• Industry, company scale, IT maturity
• Core pain point description (1-2 specific scenarios)
• Existing system landscape (at minimum: which of ERP/CRM/OA are deployed)

Helpful if available:

• Specific business scenario to improve (order processing, after-sales, procurement, etc.)
• Agent/RPA/automation pilot experience
• AI/data team presence
• Budget and timeline

Completion gate: At least 3 required items collected. Do NOT proceed to Step 2 without them.

Step 2: Rapid Three-Question Screening

Use three sequential questions (one at a time, wait for each answer) to narrow from 7 dimensions to 2-3 primary constraints within ~30 minutes.

Q1: "In this task, what action consumes the most human time?"

Map the answer to dimensions:

• Repeatedly identifying, extracting, entering unstructured info → Perception
• Moving data between systems, comparing, transferring → Orchestration
• Watching status, waiting for anomalies, reactive response → Monitoring
• Aligning views, chasing confirmations, pushing progress → Collaboration
• Synthesizing judgment, ranking, selecting options → Analysis

Q2: "In this task, where do the knowledge and judgment logic primarily reside?"

• Concentrated in a few senior employees' heads → Sedimentation
• Everyone can do it, but doubling volume breaks the team → Scalability

Q3: "How do you currently measure the quality of this task's output?"

Confirm whether baseline data exists (processing time, accuracy, manual effort). If no baseline exists, flag this as a prerequisite before Agent introduction — ROI cannot be calculated without it.

Completion gate: Can list 2-3 dimensions the scenario falls on, and confirm baseline data status. If still ambiguous, return to Step 1.

Step 3: Locate Constraint Layer

Map bottlenecks to one of three layers along the enterprise value chain.

Layer	Dimensions	Typical Signs
Layer 1: Information Input	Perception	Heavy manual entry of emails, PDFs, scans, voice; low efficiency converting unstructured to structured
Layer 2: Information Processing (most common bottleneck)	Analysis, Orchestration, Monitoring, Collaboration	Systems exist but rely on manual data transfer; cross-dept coordination drains time; anomaly detection is reactive; multi-source analysis overwhelms human bandwidth
Layer 3: Information Assetization	Sedimentation, Scalability	Successful pilots can't replicate across teams; peak periods exceed human capacity; staff turnover degrades quality

Completion gate: Identify ONE primary constraint layer. If spanning multiple layers, rank by priority and focus Phase 1 on the top one.

Step 4: Pinpoint Specific Dimensions

For each candidate dimension, validate with concrete business evidence from the user.

Read references/compass-dimensions.md for detailed dimension definitions, core questions, and judgment signals.

Completion gate: Lock 2-3 priority dimensions, each with user-provided business evidence. If user cannot provide examples, return to Step 1 or 2.

Step 5: System Readiness Check

Before supplier matching, evaluate how well existing systems support Agent-oriented operations.

Read references/system-readiness.md for the four evaluation dimensions (Interface Capability, Permission Model, Operation Traceability, Data Semantics).

Rate each as: Ready / Partial / To-Build

Completion gate: All four dimensions rated. Produce a system-side TODO list to run in parallel with scenario implementation.

Step 6: Match Supplier Directions

Based on COMPASS dimension bottlenecks and industry attributes, recommend supplier type combinations from IDC MarketGlance.

Read references/supplier-mapping.md for the complete COMPASS-to-supplier mapping and MarketGlance sub-category details.

When the user asks for specific vendor names, load the corresponding vendor directory:

• Industry vendors: Read references/vendor-industry.md
• Enterprise function vendors: Read references/vendor-enterprise.md
• Platform vendors: Read references/vendor-platform.md

Three supplier types:

1. Industry Scenario Vendors — 11 vertical sectors (government, finance, manufacturing, transport, internet, retail, smart devices, energy, healthcare, automotive, embodied AI)
2. Enterprise Scenario Vendors — 14 functional areas (operations, software dev, procurement, finance, marketing & sales, HR, customer service/conversational AI, supply chain, legal, data analytics, code generation, APA, digital humans, security)
3. Agent Development Platform Vendors — Full lifecycle: build, orchestrate, test, operate, govern

Completion gate: Recommend at least TWO supplier types with clear role definitions. Specify which type serves as platform base vs. industry/functional depth.

Step 7: Output Recommendations

Produce structured output with five sections:

1. COMPASS Diagnosis Result

• Primary constraint layer
• Core bottleneck dimensions (2-3) with evidence from Steps 2-4

2. Recommended Entry Scenarios

• 1-2 high-priority pilot scenarios matching bottleneck dimensions
• Scenarios should: have clear cross-system characteristics, high manual handoff cost, easily verifiable results

3. Supplier Direction Recommendations

• Multi-type supplier combination with role definitions
• Map to MarketGlance sub-categories where applicable

4. Implementation Roadmap

• Phase 1: One local closed-loop
• Phase 2+: Expand outward
• Parallel: IT governance (permissions, security, audit)

5. Human-Agent Boundary

Read references/human-agent-boundary.md for the complete three-category boundary checklist.

At minimum, explicitly state:

• What Agents analyze and decide
• What Agents execute via tools/systems
• What MUST remain with humans (exception approval, critical decisions, relationship management)

Completion gate: All five sections present, each with at least one actionable conclusion. Diagnosis traces back to Step 2-4 evidence. Scenarios map to dimensions. Supplier directions map to scenarios.

Common Judgment Biases (watch for these)

1. Agent as万能钥匙: Rule-based tasks are better served by code/workflows/RPA. Don't over-engineer with Agents.
2. Only picking easy-ROI scenarios: Replacement scenarios (data entry, FAQ) have low ceilings. Cross-system, cross-role judgment scenarios have longer value curves. Balance both types.

Interaction Principles

• If user describes pain points but no industry, confirm industry first
• If user's need is very specific (e.g., "we want a customer service Agent"), allow starting from Step 4, but Step 5 and 6 cannot be skipped
• Keep language plain and direct; avoid framework jargon in output
• Always be business-value oriented, not technology-for-technology's-sake

Exception Handling

Situation	Response
User refuses three-question screening	Use references/example-diagnosis.md as anchor for self-mapping; if still resistant, simplify to industry + pain point
User insists on specific vendor names	State this skill provides directional guidance only; refer to IDC MarketGlance vendor evaluation reports
User's scenario spans industries/dimensions	Ask user to pick ONE most-desired improvement; run diagnosis on that single scenario; defer others to Phase 2-3
User wants ROI with no baseline data	Per Rule 6, do NOT estimate ROI. Advise collecting baseline data first (processing time, accuracy, manual effort). Give directional value points, not numbers.
User's information is contradictory	Do NOT reconcile on their behalf. Present the contradiction and ask for confirmation.

Reference Files

Load these files on demand as the workflow progresses:

• references/compass-dimensions.md — Detailed definitions, core questions, and judgment signals for all 7 COMPASS dimensions. Read in Step 4.
• references/supplier-mapping.md — Complete COMPASS-to-supplier mapping, MarketGlance sub-categories, and combination patterns. Read in Step 6.
• references/system-readiness.md — Four evaluation dimensions with detailed criteria for system readiness assessment. Read in Step 5.
• references/human-agent-boundary.md — Complete three-category human-Agent boundary checklist. Read in Step 7.
• references/example-diagnosis.md — Full worked example: manufacturing enterprise order processing scenario. Use as anchor when user struggles to self-identify.
• references/vendor-industry.md — Industry Scenario Vendors: 11 vertical sub-categories with complete vendor lists. Load when user asks for industry-specific vendor names.
• references/vendor-enterprise.md — Enterprise Scenario Vendors: 14 functional sub-categories with complete vendor lists. Load when user asks for function-specific vendor names.
• references/vendor-platform.md — Agent Development Platform Vendors: complete vendor list. Load when user asks for platform vendor names.