Core Presales Intelligence

Core Presales Intelligence — Shared Conventions

This skill defines the shared rules for the presales skill family: nine sibling skills under skills/presales/ that pull context from Gong, Salesforce, Google Drive, and Workiva's internal GRC/Documents environment to help Solution Consultants (SCs) and Value Consultants (VCs) prepare for calls, write briefs, and respond to RFPs.

It is not meant to be triggered by a user directly (disable-model-invocation: true). Every sibling skill's own SKILL.md must restate the non-negotiable rules below (read-only constraint, evidence tagging, synthetic data) in its own words, and link back here (../core-presales-intelligence/SKILL.md) and to ../shared/evidence-schema.md for the handoff payload shape.

1. Evidence classification taxonomy

Every factual claim in any presales-skill output must be tagged with exactly one of the following classifications. Use the literal string values shown — downstream adapters key off of them.

Every evidence[] entry in the handoff payload (see §4 and ../shared/evidence-schema.md) must use one of these six values in its classification field. Do not invent new classification values.

2. Fallback strings for unverifiable claims

When a claim cannot be classified into one of the six categories above because the underlying source could not be checked, do not guess or silently omit it. Use one of these two fixed strings as the claim text (or as the value of an otherwise-required field):

• "Unknown — requires validation." Use when no data source (Gong, Salesforce, Drive, internal GRC/Documents) returned anything on the topic, and the skill has no basis for even a working hypothesis. This says "we don't know" plainly.

• "Requires Workiva-source validation." Use specifically when a claim is about Workiva's own product, pricing, security posture, roadmap, or internal capability, and the skill cannot confirm it against a verified Workiva-internal source. This flags "don't let a customer-facing rep repeat this until Workiva confirms it" — a narrower and higher-stakes case than generic unknowns.

Both strings must be used verbatim (exact casing and punctuation) so downstream adapters and reviewers can pattern-match on them.

3. Read-only hard constraint (non-negotiable)

Every presales skill is READ-ONLY. None of the nine sibling skills — nor this shared foundation — may ever write, update, create, or delete data in Salesforce, Gong, Google Workspace, or any Workiva internal system, even when the underlying connector technically supports write operations (e.g. Salesforce via Workato is a read+write tier connector, but presales skills only ever call its read/query tools).

Concretely:

• Only call query/search/read/get-style tools (see §4 for the approved list).
• Never call a tool that creates, updates, or deletes a Salesforce record, a Google Drive file, a Gong entity, or a Workiva internal document — with the narrow exception of add_gtmasst_log, which each sibling skill may use only to append its own audit/activity log entry, never to alter opportunity, account, or contact data.
• All output is a draft or recommendation for a human to review and manually apply. No skill in this family autonomously executes an action on behalf of the SC/VC.
• Every sibling skill's own SKILL.md must restate this constraint explicitly, in its own words, near the top of the file. Do not assume the reader has read this shared document.

4. Available connector tools

These are the actual MCP tools presales skills may reference or call in this environment. Sibling skills should only call tools from this list, and only in their read capacity per §3.

Gong MCP (read-only)

• Ask Account
• Ask Deal
• Generate Brief

Salesforce (via Workato — connector is read+write tier; presales skills only ever READ)

• search_accounts
• search_opportunities
• search_contacts
• search_leads
• search_products
• search_tasks
• execute_soql_query
• get_current_user_id
• get_inside_sales_opp_audit_data
• retrieve_semantic_model
• view_report
• add_gtmasst_log (write-of-own-audit-log only — see §3 exception)

Google Drive (read-only usage only)

• search files
• read file content
• download file content
• get file metadata

Workiva internal GRC/Documents MCP (internal-only demo environment, read-only)

• query/search/read tools exposed by that server for internal demo content, product documentation, and GRC reference material

5. Handoff payload convention

Every presales skill ends its output with a single machine-readable JSON block, so two downstream adapter skills can consume it without re-parsing prose:

• workiva-mcp-adapter — maps the payload into a Workiva internal GRC/Documents query format.
• demo-framework-adapter — maps the payload into a Workiva demo-environment setup payload.

The full schema lives in ../shared/evidence-schema.md. The minimal common shape is:

{
  "skill": "gong-call-brief",
  "account_id": "acme-test-corp-001",
  "opportunity_id": "0061-acme-esg-renewal",
  "summary": "One paragraph, human-readable summary of the skill's output.",
  "evidence": [
    {
      "claim": "Acme Test Corp's close process takes 11 business days.",
      "classification": "confirmed_customer_statement",
      "source": "Gong call 2026-08-14, Jordan Rivera"
    }
  ],
  "open_questions": [
    "Does Acme Test Corp's close process span multiple ERPs?"
  ],
  "handoff_target": ["workiva-mcp-adapter", "demo-framework-adapter"]
}

Fields:

• skill — the sibling skill's kebab-case name (e.g. gong-call-brief, rfp-gap-analysis).
• account_id / opportunity_id — Salesforce IDs (or synthetic placeholders in examples), null if not applicable to that skill's output.
• summary — short human-readable synopsis.
• evidence — array of {claim, classification, source} objects. Every classification value must be one of the six taxonomy values from §1. Unverifiable claims use one of the §2 fallback strings as the claim text. source should be traceable (transcript date + speaker, Salesforce object
  • tool name, document title, etc.) — never blank.
• open_questions — array of plain-text strings, one per unresolved question requiring human/customer validation.
• handoff_target — array containing one or both of "workiva-mcp-adapter", "demo-framework-adapter", indicating which downstream adapter(s) should consume this payload.

See ../shared/evidence-schema.md for the full JSON Schema definition, field types, and validation rules.

6. Synthetic data rule

All examples in every presales skill's SKILL.md — including this one — must use obviously fake, synthetic account and person names. Never use real customer, prospect, or Workiva-internal-only names or data in example text.

Approved-style examples: Acme Test Corp, Globex Test Industries, Northwind Test Holdings, Jordan Rivera, Sam Okafor, Priya Chandran.

Do not reuse real Salesforce account IDs, opportunity IDs, or Gong call IDs in example text — use clearly synthetic placeholders (e.g. acme-test-corp-001, 0061-acme-esg-renewal).

7. Checklist for sibling skills

Each of the nine sibling SKILL.md files should:

1. Restate the read-only constraint (§3) explicitly, in its own words.
2. Tag every claim in its output with one of the six evidence classifications (§1), using the exact fallback strings (§2) where evidence is missing.
3. Only call connector tools from the approved list (§4), and only in read capacity.
4. End its output with the handoff JSON block conforming to ../shared/evidence-schema.md (§5).
5. Use only synthetic account/person names in its own examples (§6).
6. Link back to ../core-presales-intelligence/SKILL.md and ../shared/evidence-schema.md rather than duplicating this document in full.