A Claude Skill that turns the messy reality of pipeline data into a one-page forecast narrative for the weekly forecast call: commit, best-case, and pull-in deals; risks against committed; deals that moved this week and why; deals stalled. Reads Salesforce as ground truth and Gong as the truth-check on what reps say versus what customers said.

What you’ll need

Salesforce with a clean Stage and Close Date discipline
Gong API access for call activity per opportunity
Claude Code or Claude.ai with the Skill loaded
Your team’s stage-to-forecast-category mapping
A definition of “stalled” (e.g., no Gong call in 14 days on a commit deal)

Setup

Install the Skill. account-forecast.skill into ~/.claude/skills/. Set SFDC_TOKEN and GONG_API_KEY.
Configure rules. Edit references/forecast-rules.md with: which stages count as commit, what “stalled” means in days and signals, and what counts as a “real” close-date slip versus expected.
Run for one segment. build_forecast_narrative(segment="enterprise-amer", week_ending="2026-05-01"). The Skill outputs three sections: numbers, risk to commit, and deal-by-deal commentary on movers.
Pre-publish to Slack. Post the narrative to your forecast channel 30 minutes before the live call so leaders read it cold and arrive with sharper questions.

How it works

The Skill snapshots Salesforce at the cutoff time, then computes deltas against the prior week’s snapshot stored locally. It surfaces every deal that changed stage, close date, or amount, plus every deal that was added or removed from commit.

For each commit deal, the Skill cross-references Gong activity. If a deal is committed but has zero customer-side activity in 14 days, it’s flagged. If a deal moved to commit this week, the Skill pulls the most recent customer call and extracts a one-line “why now” from the transcript. This is the part reps love and hate equally — it makes the rep’s narrative defensible or visibly thin.

Final synthesis is a single Markdown page. RevOps reviews it, adds context, and ships.

Watch-outs

Stage discipline. If reps move deals to commit without the work being done, no Skill fixes that. Pair this with a stage-progression validator workflow.
Gong silence is not always bad. Some buyers genuinely go quiet during procurement. The Skill flags it; humans interpret.
Snapshot storage. Store snapshots somewhere version-controlled (an S3 bucket, a Snowflake table). If you only diff against last week’s live state, you lose the audit trail and the Skill’s signal degrades.

Stack

Salesforce — pipeline state, stage history, deal amounts
Gong — activity signal and customer-side voice
Claude — narrative synthesis and risk surfacing