Agent-Skills.md

Agent Skills: Client Intelligence Skill

Generate comprehensive project intelligence for Improvado clients. Use when user says "analyze client [name]", "client overview", "project intelligence for [client]", "what's happening with [client]". Automatically finds client folder, queries all context (Gong, Jira, emails), identifies 3-9 projects (MECE), generates detailed reports per project, and creates executive summary.

UncategorizedID: tekliner/improvado-agentic-frameworks-and-skills/client-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/client-intelligence

Skill Files

Browse the full folder contents for client-intelligence.

Download Skill

Loading file tree…

skills/client-intelligence/SKILL.md

Skill Metadata

Name
client-intelligence
Description
Generate comprehensive project intelligence for Improvado clients. Use when user says "analyze client [name]", "client overview", "project intelligence for [client]", "what's happening with [client]". Automatically finds client folder, queries all context (Gong, Jira, emails), identifies 3-9 projects (MECE), generates detailed reports per project, and creates executive summary.

Client Intelligence Skill

Core Purpose: Generate comprehensive project portfolio overview for any Improvado client using two-phase LLM analysis (identify projects → generate reports → create summary).

Quick Reference

Use this skill when:

  • "What's the status with [client]?"
  • "Generate project intelligence for [client]"
  • "Client overview for [company name]"
  • "Analyze all projects for [client]"
  • "Create executive summary for [client]"

Workflow:

graph LR
    Find[Find Client Folder] --> Load[Load All Context]
    Load --> Phase1[Phase 1: Identify 3-9 Projects]
    Phase1 --> Phase2[Phase 2: Generate Reports]
    Phase2 --> Summary[Create Executive Summary]
    Summary --> Commit[Commit & Share GitHub Links]

Prerequisites

Required modules (already exist):

  • client_cases/00_client_context_loader.py - Loads Gong, Jira, Email, files
  • client_cases/01_project_intelligence_pipeline.py - Phase 1/2 pipeline
  • client_cases/0_TEMPLATE/00_PROJECT_INTELLIGENCE_TEMPLATE.md - Report template

Environment:

  • GOOGLE_API_KEY or GEMINI_API_KEY in .env
  • ClickHouse access for internal analytics (Gong, Jira, Email)

Step-by-Step Execution

Step 1: Find/Verify Client Folder

Search for client in ClickHouse:

# Using python (ch command may not be in PATH)
claude_venv/bin/python -c "
from data_sources.clickhouse import internal
df = internal.query('''
    SELECT dts_agency_id, dts_agency_title, company_domain_id
    FROM internal_analytics.biz_active_customers
    WHERE lower(dts_agency_title) LIKE '%CLIENT_NAME%'
    LIMIT 5
''')
print(df.to_string())
"

Check existing folder:

ls -la client_cases/ | grep -i "client_name"

Folder format: im_{agency_id}_{hash}___ClientName

If folder doesn't exist: Create using template from client_cases/0_TEMPLATE/

Step 2: Load All Context

The pipeline automatically loads:

  1. Gong calls (last 365 days) - from dim_gong_calls
  2. Jira tickets with comments (last 365 days) - from mrt_jira_tickets
  3. Email threads (last 180 days) - from stg_gmail_cs_messages
  4. Client files (Excel, CSV, TXT, SQL) - from client folder

Manual context check:

# Test context loading
claude_venv/bin/python client_cases/00_client_context_loader.py \
  client_cases/im_XXXX_XXX___ClientName/ \
  --domain "clientname" \
  --stats-only

Step 3: Run Phase 1 - Identify Projects

claude_venv/bin/python client_cases/01_project_intelligence_pipeline.py \
  clientname \
  --phase 1 \
  --docs client_cases/im_XXXX_XXX___ClientName/ \
  --output client_cases/im_XXXX_XXX___ClientName/project_intelligence/

Expected output: 00_projects_identified.json with 3-9 projects (MECE)

Review with user before proceeding:

Found N projects for [Client]:
1. [Project Name] - [Status] - [Key Blocker]
2. ...

Proceed with Phase 2 to generate detailed reports?

Step 4: Run Phase 2 - Generate Reports

claude_venv/bin/python client_cases/01_project_intelligence_pipeline.py \
  clientname \
  --phase 2 \
  --output client_cases/im_XXXX_XXX___ClientName/project_intelligence/

Expected output: NN_project_name_report.md for each project

Step 5: Create Executive Summary

After Phase 2, create 00_CLIENT_EXECUTIVE_SUMMARY.md that combines all projects:

Template structure:

# [Client]: Executive Project Portfolio Summary

## Thesis
[Client] engagement encompasses [N] projects spanning [domains],
with data flowing through [path], currently facing [M] blockers.

## Portfolio Overview Diagram
[Mermaid diagram showing all projects and data flow]

## Portfolio Health Summary
| Status | Count | Projects |
|--------|-------|----------|
| In Progress | X | ... |
| Waiting | Y | ... |

## Project Portfolio Matrix
[Table with all projects, Jira, status, leads]

## Critical Blockers Across Portfolio
[Top 5 blockers with actions]

## Project Summaries
[Brief summary of each project with next actions]

## Document Links
[GitHub links to all reports]

Step 6: Fix Mermaid Diagrams

CRITICAL: GitHub doesn't render emojis in Mermaid diagrams!

Fix all diagrams:

  • Remove emojis from subgraph labels
  • Use underscores in subgraph names (not quotes)
  • Simplify edge labels

Example fix:

- subgraph "📋 REQUIREMENTS"
+ subgraph REQUIREMENTS

- B1[🔴 Critical Blocker]
+ B1[Critical Blocker]

- B1 --> |blocks| OUTPUT
+ B1 --> OUTPUT

Step 7: Commit and Share

cd client_cases/im_XXXX_XXX___ClientName/project_intelligence/
git add .
git commit -m "docs(client): add project intelligence reports for [Client]"
git push

Generate GitHub links:

https://github.com/tekliner/tcs-chrome-extension/blob/main/client_cases/im_XXXX_XXX___ClientName/project_intelligence/00_CLIENT_EXECUTIVE_SUMMARY.md

Key Database Tables

All on Palantir shard (internal_analytics):

| Table | Purpose | Key Columns | |-------|---------|-------------| | dim_gong_calls | Call recordings | gong_call_url, gong_call_title, gong_call_content_brief | | mrt_jira_tickets | Support tickets | jira_ticket_id, jira_ticket_summary, jira_comment_body_array | | stg_gmail_cs_messages | Email threads | gmail_msg_subject, gmail_msg_body, gmail_msg_from_email | | biz_active_customers | Agency lookup | dts_agency_id, dts_agency_title, company_domain_id |

MECE Project Identification Rules

Mutually Exclusive:

  • Each project has distinct deliverable
  • No overlapping scope
  • Clear boundaries

Collectively Exhaustive:

  • All customer activities covered
  • No gaps in coverage
  • Include blocked/waiting work

Project Boundaries (what defines a project):

  • Different Jira epic/ticket series
  • Different data sources
  • Different stakeholders
  • Different timeline/phase

Target: 3-9 projects per client (not 1, not 20)

Report Template Sections

Each project report includes:

| Section | Purpose | Required | |---------|---------|----------| | Thesis | Single-sentence summary | ✅ | | Overview Diagram | Mermaid data flow + blockers | ✅ | | Key Blockers | Status table + details | ✅ | | Quick Reference | Key facts table | ✅ | | §1.0 Requirements | Source documents | ✅ | | §2.0 Specifications | Metrics, dimensions | If applicable | | §3.0 Architecture | Data flow diagram | If applicable | | §4.0 Implementation | Component status | ✅ | | §5.0 Stakeholders | Teams involved | ✅ | | §6.0 Timeline | Decision history | ✅ | | §7.0 Agent Next Steps | Actions + SQL queries | ✅ | | Ground Truth Index | All citations | ✅ |

Citation Format

Always cite sources:

  • [JIRA:PS-4406] - Jira ticket
  • [CALL:2025-12-29|Weekly Sync] - Gong call
  • [EMAIL:2025-12-31|Subject] - Email thread
  • [EXCEL:filename|sheet] - Excel file
  • [SQL:model_name.txt] - SQL model

Example Execution

# Full pipeline for Earth Justice
CLIENT="earthjustice"
FOLDER="client_cases/im_XXXX_XXX___ExampleClient"

# Phase 1: Identify projects
claude_venv/bin/python client_cases/01_project_intelligence_pipeline.py \
  $CLIENT --phase 1 --output $FOLDER/project_intelligence/

# Review projects with user...

# Phase 2: Generate reports
claude_venv/bin/python client_cases/01_project_intelligence_pipeline.py \
  $CLIENT --phase 2 --output $FOLDER/project_intelligence/

# Create executive summary manually (or extend pipeline)

# Fix Mermaid, commit, push

Troubleshooting

Rate limits (429):

  • Reduce max_tokens in pipeline (default 200K)
  • Wait 60 seconds between phases

No Gong calls found:

  • Check domain pattern matches sfdc_company_name_array
  • Try variations: "company", "company.com", "company name"

JSON parse errors:

  • Pipeline has recovery logic for truncated responses
  • Check /tmp/gemini_response_*.txt for raw output

Mermaid not rendering:

  • Remove ALL emojis from diagrams
  • Use simple subgraph names without quotes

Success Criteria

✅ Client folder identified correctly ✅ 3-9 projects identified (MECE) ✅ Each project has detailed report with citations ✅ Executive summary combines all projects ✅ All Mermaid diagrams render on GitHub ✅ GitHub links provided to user

Meta: This skill reuses existing modules (00_client_context_loader.py, 01_project_intelligence_pipeline.py) following DRY principles. Only the executive summary generation needs to be added inline after Phase 2 completes.