A Claude Skill that validates whether a deal genuinely meets your stage-entry criteria. For every deal that moved to a later stage in the past week, the Skill checks the required fields, the supporting Gong evidence, and the documented next steps. Flags deals that progressed on rep optimism rather than buyer commitment.

What you’ll need

Salesforce with stage-progression history (history tracking enabled)
Gong API key with deal-level call access
Claude Code with the Skill installed
Your stage definition and exit criteria as a Markdown doc
A weekly run cadence (cron or scheduler)

Setup

Document the stages. Edit references/stage-criteria.md with each stage and its hard requirements. For example: “To enter Stage 3 (Negotiation), the deal must have an identified economic buyer, a documented success criteria, and at least one decision-maker call recorded in Gong.”
Install the Skill. Drop stage-validator.skill into ~/.claude/skills/. Set SFDC_TOKEN and GONG_API_KEY.
Run weekly. validate_progressions(week_ending="2026-05-02"). The Skill pulls every deal that progressed in the window and checks each against criteria.
Output a queue. Failed validations go to a “stage-violation review” queue. The deal owner’s manager gets a weekly digest with rationale per deal.
Pair with coaching. This isn’t a punishment tool. The output is a coaching trigger.

How it works

The Skill pulls stage history and identifies progressions in the window. For each, it checks deterministic criteria first: required field values populated, required activities logged, expected stakeholders connected. These are easy.

The hard part is qualitative. Did the rep claim “buyer agreed on success criteria” when no Gong call captured that conversation? The Skill cross-references rep notes against actual call transcripts and flags claims unsupported by recorded conversation. Claude is conservative here — it flags only when the absence is clear, not on every ambiguity.

Output per deal is a one-line verdict (pass, flag, fail) plus a sentence on the gap. Managers prioritize from the top of the failed list.

Watch-outs

Rep pushback is normal. “I had the conversation off-call.” Sometimes true, often not. Pair the tool with a written-evidence policy: if it’s not in CRM or on a recording, it didn’t happen.
Don’t validate every stage. Validate the high-value transitions — typically the stage where commit forecasts start. Otherwise the noise drowns the signal.
Audit your criteria. If 80% of deals fail validation, your criteria are wrong, not your reps. Tune.

Stack

Salesforce — stage history and deal fields
Gong — recorded conversation evidence
Claude — qualitative criteria evaluation