Validate Lead

/validate-lead — ICP Lead Validation

Validate whether a company is an ICP fit based on research and your ICP scoring criteria.

Usage

• /validate-lead {company} [job_url] — Validate a single company
• /validate-lead --batch — Validate unreviewed companies from your outreach sheet
• /validate-lead --improve "learning" — Add a learning to improve future validations

Prerequisites

• config/config.yaml with ICP section filled (scoring criteria, no-go rules, tiers)
• Recommended: my-context/icp-definition.md filled (via /setup --icp)

---

Default Mode Flow

Step 1: Load Context (parallel reads)

Read these files in parallel:

• ~/.claude/skills/validate-lead/learnings.md
• my-context/icp-definition.md (if exists)
• config/config.yaml -> icp section

Step 2: Research the Company (parallel where possible)

• If job URL provided: WebFetch the listing -> extract role title, tech stack, requirements, company description
• WebSearch "{company} what they do product" -> understand what they build
• WebSearch "{company} employees team size" -> estimate headcount
• If LinkedIn URL provided: WebFetch it

Step 3: Apply No-Go Rules (check FIRST before scoring)

Read config.icp.no_go_rules. Auto-reject if ANY rule matches.

If No-Go triggered -> output verdict immediately with reason. Skip scoring.

Step 4: Score on ICP Scale

Use config.icp.scoring_criteria for criteria. Each criterion scored 0 or 1. Sum scores and apply tier thresholds from config.icp.tiers:

• Score >= config.icp.tiers.core = Core ICP
• Score >= config.icp.tiers.broader = Broader ICP
• Below broader threshold = Not ICP

Step 5: Output

Present results in this format:

## {Company Name} — Lead Validation

### Verdict
| Field | Value |
|-------|-------|
| ICP Fit? | Yes / No |
| Action | Go ahead / No go |
| Summary | {one-line summary} |

### Research Summary
- **What they do:** ...
- **Headcount:** ...
- **Geography:** ...
- **Product type:** ...
- **Job listing signals:** ... (if applicable)

### ICP Scorecard ({X}/{max})
| # | Criterion | Score | Reasoning |
|---|-----------|-------|-----------|
| 1 | {criteria from config} | 0/1 | ... |
| 2 | ... | 0/1 | ... |

### Caveats
- ...

If config.integrations.outreach_sheet.enabled is true:

• Map verdict to the configured columns: "Copy these values to the sheet: {column}={value}, ..."
• Include the sheet link: https://docs.google.com/spreadsheets/d/{config.integrations.outreach_sheet.spreadsheet_id}/edit

Step 6: Ask for Feedback

After output, ask: "Any feedback on this verdict? I'll save it for future validations."

If user provides feedback -> append to ~/.claude/skills/validate-lead/learnings.md with date and company context.

---

Batch Mode Flow

Step 1: Read the Sheet

Requires config.integrations.outreach_sheet.enabled: true.

sheets_getText(spreadsheetId: "{config.integrations.outreach_sheet.spreadsheet_id}")

Step 2: Find Unreviewed Rows

Filter for rows where the verdict column and action column are blank/empty (using column letters from config).

Step 3: Present List

Show the unreviewed companies with their signal/URL columns. Ask user which ones to validate (all, or specific ones).

Step 4: Validate Sequentially

Run the default validation flow for each selected company. Pause briefly between web searches to be rate-limit friendly.

Step 5: Summary Table

After all validations, output a summary:

| Company | Score | Tier | Verdict | Action | Comment |
|---------|-------|------|---------|--------|---------|
| ... | X/{max} | Core/Broader/No | Yes/No | Go ahead/No go | ... |

---

Improve Mode Flow

When user runs /validate-lead --improve "learning text":

1. Read ~/.claude/skills/validate-lead/learnings.md
2. Append the learning under ## User Feedback & Learnings with today's date
3. Confirm: "Saved learning. This will be applied to future validations."
4. Do NOT run any validation

---

Important Notes

• Google Sheets MCP is read-only — always output values for manual copy-paste
• When uncertain about a company, default to "Go ahead with caveats" — better to review than miss
• Always check learnings.md first — past feedback may override default scoring
• The skill improves over time as feedback accumulates in learnings.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: himanshusaleria
• Source: himanshusaleria/gtm-engine
• License: MIT

Install and usage instructions live in the source repository linked above.