Agent-Skills.md

Agent Skills: Company Intelligence Skill

Unified company intelligence pipeline that auto-detects stage (prospect/active/dual) from SFDC, runs stage-appropriate prompts, and outputs to Obsidian vault. Use when user says 'company intelligence for [domain]', 'deal brief for [prospect]', 'analyze [company]', 'what do we know about [company]'. Merges Deal Intelligence Brief (prospects) and Client Intelligence (active customers) into one skill.

UncategorizedID: tekliner/improvado-agentic-frameworks-and-skills/company-intelligence
41

Install this agent skill to your local

pnpm dlx add-skill https://github.com/tekliner/improvado-agentic-frameworks-and-skills/tree/HEAD/skills/company-intelligence

Skill Files

Browse the full folder contents for company-intelligence.

Download Skill

Loading file tree…

skills/company-intelligence/SKILL.md

Skill Metadata

Name
company-intelligence
Description
"Unified company intelligence pipeline that auto-detects stage (prospect/active/dual) from SFDC, runs stage-appropriate prompts, and outputs to Obsidian vault. Use when user says 'company intelligence for [domain]', 'deal brief for [prospect]', 'analyze [company]', 'what do we know about [company]'. Merges Deal Intelligence Brief (prospects) and Client Intelligence (active customers) into one skill."

Company Intelligence Skill

Core Purpose: Auto-detect company stage from SFDC (prospect / active customer / both), load all available context (Gong, Jira, Gmail, SFDC, client files), run stage-appropriate Gemini prompts, and write intelligence report to Obsidian vault.

Quick Reference

Use this skill when:

  • "Company intelligence for 898marketing.com"
  • "What do we know about [company]?"
  • "Deal brief for [prospect domain]"
  • "Analyze [company] — prospect or customer?"
  • "Generate intelligence for [company]"
  • "Update report for [company]"

Workflow:

graph LR
    Input[Domain or Company Name] --> Stage[Detect Stage from SFDC]
    Stage --> Context[Load Context - Gong Jira Gmail SFDC]
    Context --> Route{Prospect or Customer?}
    Route -->|Prospect| Brief[Deal Intelligence Brief - 14 sections]
    Route -->|Active| Pipeline[3-Phase Project Intelligence]
    Route -->|Both| Both[Run Both Pipelines]
    Brief --> Obsidian[Write to Obsidian Vault]
    Pipeline --> Obsidian
    Both --> Obsidian

Prerequisites

Pipeline follows Miras Data Pipeline Philosophy — 2 files, 0 magic:

  • algorithms/A8_G&A_div/company_intelligence/pipeline.py — Algorithm (WHAT we do, ~50 lines)
  • algorithms/A8_G&A_div/company_intelligence/impl.py — Implementation (HOW we do it)
  • algorithms/A8_G&A_div/company_intelligence/templates/ — Stage-specific prompt templates

Environment:

  • GOOGLE_API_KEY or GEMINI_API_KEY in .env
  • ClickHouse access for internal analytics

Step-by-Step Execution

Step 1: Run the Unified Pipeline

One-command execution:

claude_venv/bin/python "algorithms/A8_G&A_div/company_intelligence/pipeline.py" \
  DOMAIN_OR_NAME

With flags:

# Force prospect-only (even if active customer)
claude_venv/bin/python "algorithms/A8_G&A_div/company_intelligence/pipeline.py" \
  898marketing.com --prospect-only

# Force customer-only
claude_venv/bin/python "algorithms/A8_G&A_div/company_intelligence/pipeline.py" \
  898marketing.com --customer-only

# Full refresh (ignore existing reports)
claude_venv/bin/python "algorithms/A8_G&A_div/company_intelligence/pipeline.py" \
  898marketing.com --force-refresh

# With client folder for extra file context
claude_venv/bin/python "algorithms/A8_G&A_div/company_intelligence/pipeline.py" \
  t-mobile.com --client-folder client_cases/im_XXXX___TMobile

Step 2: Review Pipeline Output

The pipeline will print:

  1. Stage detection — prospect, active_customer, or dual_status
  2. Context stats — how many tokens loaded, from which sources
  3. Mode — first_run or update
  4. Generation progress — which phases/sections are being generated
  5. Output path — Obsidian vault location

Step 3: Verify Obsidian Output

Output locations:

  • vault/Improvado/01-Revenue/11-Sales/{Company Name}/ — all stages

File names by stage:

  • Prospect: {Company} - Deal Intelligence Brief.md
  • Active Customer: {Company} - Client Intelligence Report.md
  • Dual Status: Both files generated

Frontmatter includes: session_id, stage, company, domain, generated_at, mode, agency_id, tags, related_to

Step 4: Open in Obsidian

# Open the generated report (example for 898 Marketing)
open "obsidian://open?vault=vault&file=Improvado%2F01-Revenue%2F11-Sales%2F898%20Marketing%2F898%20Marketing%20-%20Deal%20Intelligence%20Brief"

Stage Detection Logic

Two SFDC views queried:

| View | What It Finds | Key Fields | |------|---------------|------------| | biz_active_customers | Active customers | dts_agency_id, sfdc_primary_csm_name, sfdc_arr | | biz_active_opportunities | Open deals | opportunity_stage, opportunity_amount, opportunity_close_date |

Stage outcomes:

| Result | Meaning | Pipeline | |--------|---------|----------| | prospect | Open opportunity, not a customer | Deal Intelligence Brief (14 sections) | | active_customer | Active customer, no open opportunities | Client Intelligence (3-phase) | | dual_status | BOTH — customer with open upsell/expansion | Both pipelines run | | unknown | Not found in SFDC | Error — check domain spelling |

Test stage detection in Python REPL:

# Debug mode — run pipeline step by step
from impl import *
stage = detect_stage("898marketing.com")
context = build_context("898marketing.com", stage)
print(f"Gong: {len(context['gong_calls'])}, Jira: {len(context['jira_tickets'])}")

Pipeline Outputs by Stage

Prospect: Deal Intelligence Brief (14 Sections)

Based on Peter Cloutier's proven workflow. Single Gemini call produces:

| # | Section | Content | |---|---------|---------| | 1 | Executive Summary | Deal snapshot, health, key blockers | | 2 | Opportunity Details | SFDC stage, amount, close date, competitors | | 3 | Stakeholder Map | Contacts, roles, engagement level | | 4 | Business Case | Pain points, desired outcomes (verbatim quotes) | | 5 | Technical Use Case | Data sources, requirements, POC status | | 6 | Critical Open Issues | Per-issue status, root cause, resolution | | 7 | Commercial Blockers | Contract terms, pricing concerns | | 8 | Evaluation Criteria | Customer's checklist with status | | 9 | Competitive Intelligence | Named competitors, comparison | | 10 | Engagement Timeline | Chronological interaction table | | 11 | JIRA Tickets | Existing + recommended to create | | 12 | Next Steps | SE priorities + AE priorities | | 13 | Risk Assessment | Probability x impact matrix | | 14 | Key Quotes | Verbatim customer quotes |

Active Customer: Client Intelligence (3 Phases)

| Phase | Purpose | Output | |-------|---------|--------| | Phase 1 | Identify 3-9 MECE projects | JSON project list | | Phase 2 | Generate per-project report (12 sections each) | Markdown reports | | Phase 3 | Executive portfolio summary | Combined document |

Update Mode

When re-running on a company that already has a report:

  • First-run mode: Full pipeline — all phases from scratch
  • Update mode: Reuses existing project list, only regenerates reports
  • Force refresh: --force-refresh flag ignores existing reports

Key Database Tables

| Table | Shard | Content | Lookback | |-------|-------|---------|----------| | biz_active_customers | Palantir | Customer + SFDC data | Current | | biz_active_opportunities | Palantir | Open deals + SFDC | Current | | dim_gong_calls | Palantir | Call metadata + transcripts | 365 days | | mrt_jira_tickets | Palantir | Tickets with comments | 365 days | | stg_gmail_cs_messages | Palantir | CS emails | 180 days |

Citation Format

All reports enforce source attribution:

  • [SFDC: field_name] — Salesforce data
  • [CALL: YYYY-MM-DD | Call Title] — Gong call
  • [EMAIL: YYYY-MM-DD | Subject] — Email thread
  • [JIRA: TICKET-ID] — Jira ticket
  • [FILE: filename] — Client file

Troubleshooting

¶1 "Company not found in SFDC"

  • Check domain spelling (use .com suffix)
  • Try company name fragment instead of domain
  • Verify company exists: ch internal "SELECT * FROM biz_active_customers WHERE lower(company_domain_id) LIKE '%name%'"

¶2 Rate limits (429 from Gemini)

  • Pipeline uses gemini-2.5-flash (fast, cheap)
  • If rate-limited, wait 60 seconds and retry
  • Reduce context with lower --max-tokens

¶3 JSON parse errors in Phase 1

  • Pipeline has built-in JSON recovery for truncated responses
  • Check /tmp/gemini_response_*.txt for raw output

¶4 Missing Gong/Jira/Email data

  • Domain pattern matching uses first part of domain (e.g., "898marketing" from "898marketing.com")
  • Try alternative search terms if no data found

Success Criteria

  • [ ] Stage detected correctly (prospect/customer/dual)
  • [ ] Context loaded from all available sources
  • [ ] Report generated with proper structure for stage
  • [ ] All claims have source citations
  • [ ] Obsidian frontmatter includes session_id, stage, company
  • [ ] Report written to correct vault folder
  • [ ] Report opens correctly in Obsidian

Related Skills

| Skill | Relationship | |-------|-------------| | /client-intelligence | Predecessor — active customer only, no stage detection | | /gong-call-analyzer | Shared dependency — Gong call extraction | | /gmail-operations | Shared dependency — email search | | /jira | Shared dependency — ticket lookup | | /improvado-mcp | ClickHouse queries for all data |

Architecture (Miras Pipeline Philosophy)

algorithms/A8_G&A_div/company_intelligence/
├── pipeline.py     — Algorithm (WHAT we do, ~50 lines)
├── impl.py         — Implementation (HOW we do it)
└── templates/
    ├── prospect_prompt.md          — 14-section Deal Brief prompt
    └── active_customer_prompt.md   — 3-phase Client Intelligence prompt

Philosophy: pipeline.py reads like documentation. impl.py has all the details. See: vault/Miras Knowledge Platform/.../Miras Data Pipeline Philosophy.md

Primary source: data_sources/obsidian/vault/Improvado/01-Revenue/11-Sales/Customer Intelligence Lifecycle - Prospect to Active.md Created: 2026-02-08 Author: Claude Code Session: 72cb0617-0c02-460e-b750-199d63378704 Version: 1.0.0 — Initial unified pipeline merging Deal Brief + Client Intelligence

Company Intelligence Skill Skill | Agent Skills