patrick（patrick 分析）

Patrick Skill

Patrick is a licensed expertise management system that provides executive decision infrastructure through structured, context-aware expertise.

⚠️ If Patrick isn't working: Check {baseDir}/install.md for complete setup instructions including license configuration and initialization.

Setup Workflow

1. Install the CLI

Run the install script to download the patrick-cli binary from the Patrick server:

./install.sh

Or install directly via curl:

curl -fsSL https://portal.patrickbot.io/downloads/latest/install.sh | bash

This will download patrick-cli for your platform, verify its SHA256 checksum, and place it in $PATRICK_DATA_PATH/bin/patrick-cli (default: ~/.patrick/bin/). The Dockerfile and run.sh add this to PATH automatically, so you can call patrick-cli directly.

2. Get Your License

Visit patrickbot.io to obtain your Patrick license.

Once you have your license, send it to me via chat:

Here's my Patrick license: <paste-your-license-here>

I'll configure it automatically. Alternatively, you can set it manually:

patrick-cli set-license "your-license-here"

To switch to a different account (wipes existing data since it's tied to your identity):

patrick-cli set-license "new-license" --wipe-data

3. Initialize Patrick Context

CRITICAL: Before running any Patrick expertise, initialize the context:

patrick-cli fetch initialize

This bootstraps Patrick with information about you, your company, and your team. Without initialization, other expertise prompts won't work properly.

4. Verify Installation

Test it by listing available expertise:

patrick-cli list

Note: The server URL defaults to https://portal.patrickbot.io. For local development:

• Host machine: export PATRICK_SERVER_URL="http://localhost:8080"
• Docker container: export PATRICK_SERVER_URL="http://host.docker.internal:8080"

The patrick-moltbot Docker container is pre-configured with the correct URL.

4. Set Up Cronjobs (Critical)

Patrick requires scheduled cronjobs to function properly. Read INSTALL.md in this directory for complete setup instructions including recommended cronjob schedules for daily briefings and commitment tracking.

License auto-renewal in cron: Before running any scheduled task, check if the license is about to expire and renew it:

# Check license expiry and renew if <1 day remaining
DAYS=$(patrick-cli license 2>&1 | grep "days remaining" | grep -oP '\d+')
if [ -n "$DAYS" ] && [ "$DAYS" -lt 1 ]; then
  patrick-cli renew
fi

# Then run your scheduled expertise
patrick-cli fetch daily-briefing --json

The patrick-cli renew command contacts the server, verifies your active subscription, and saves a fresh license with the remaining subscription days.

5. Keeping Patrick Updated

Check for updates and upgrade to the latest version:

# Check if update is available
patrick-cli upgrade --check

# Upgrade to latest version
patrick-cli upgrade

The upgrade process:

1. Contacts Patrick server for latest version info
2. Verifies SHA256 checksum of downloaded binary
3. Backs up current version
4. Installs new version
5. Shows release notes

Updates are cryptographically verified and signed by the Patrick server.

What You Get

When you list expertise, you'll see:

• Expertise ID and version
• Name and description
• Category (sense, interpret, decide, align, execute, learn, reporting, intelligence)
• Response format (structured JSON, markdown, mixed)
• Required context variables
• Whether it's bidirectional (accepts data back)

IMPORTANT: Gathering Context Before Using Patrick

For AI Agents: Before running any Patrick expertise, you MUST:

1. Check for company data files in /app/company/ or similar locations
2. Read all available context:

- Company data JSON files - Slack message archives - JIRA tickets - Git commit history - Calendar events - Any operational data available

1. Load this context into your working memory
2. Then run Patrick expertise with full awareness of company operations

Patrick expertise is most effective when you have complete situational awareness. Don't run Patrick commands without first gathering all available company data.

Commands Overview

Key Distinction:

• fetch = Get the expertise template/prompt FROM the server
• get = Retrieve your stored data/results THAT YOU SENT BACK

Using Expertise with LLMs

Workflow Overview

1. List expertise to see what's available in your license
2. Fetch an expertise with optional context variables
3. Send expertise content to an LLM for processing
4. Validate response against the provided JSON schema (if structured)
5. Store response back to Patrick using send (if bidirectional)
6. Retrieve stored data later using get

Step 1: List Available Expertise

patrick-cli list

Example output:

Available Expertise:

  daily-briefing (v1.0.0)
    Name: Daily Executive Briefing
    Category: sense
    Response Format: structured
    Bidirectional: ✓ (stores to 'daily-briefing')
    Required Context: (none)
    Description: What's urgent, developing, changed, and the one question

  decision-framing (v1.0.0)
    Name: Decision Framing
    Category: decide
    Response Format: structured
    Bidirectional: ✓ (stores to 'decision-framing')
    Required Context:
      - decision
    Description: Structures ambiguous decisions into clear trade-offs...

Step 2: Fetch Expertise

Without context (for expertise that doesn't need it):

patrick-cli fetch daily-briefing

With context variables:

patrick-cli fetch decision-framing \
  --context '{"decision":"Should we raise prices?"}'

Get JSON output for LLM integration:

patrick-cli fetch daily-briefing \
  --json

Step 3: Send to LLM

When you fetch expertise, you receive:

• content: The filled expertise text to send to the LLM
• response_schema: JSON schema for validating the LLM response (if structured)
• response_format: How the LLM should format its response (structured/markdown/mixed)

Example workflow:

# Fetch expertise as JSON
EXPERTISE_DATA=$(patrick-cli fetch daily-briefing --json)

# Extract the content field
EXPERTISE_CONTENT=$(echo "$EXPERTISE_DATA" | jq -r '.content')

# Send to your LLM (pseudo-code)
# LLM_RESPONSE=$(send_to_llm "$EXPERTISE_CONTENT")

# If structured format, validate against schema
RESPONSE_SCHEMA=$(echo "$EXPERTISE_DATA" | jq -r '.response_schema')
# validate_json "$LLM_RESPONSE" "$RESPONSE_SCHEMA"

Step 4: Store Results (Optional)

For bidirectional expertise, send the LLM response back to Patrick:

# Store the response
patrick-cli send daily-briefing \
  --data @llm-response.json

# Later, retrieve stored data
patrick-cli get daily-briefing

Key Concepts

Context Variables

Many prompts require context to fill template variables. Use --context with a JSON object:

{
  "current_phase": "pre-launch",
  "launch_date": "2026-02-15",
  "completed_items": 12,
  "target_platforms": ["iOS", "Android"]
}

Variables are substituted using {{context.key}} syntax in the expertise content.

Response Formats

• structured: LLM must return JSON matching the provided schema
• markdown: LLM returns markdown text
• mixed: LLM returns markdown with embedded JSON blocks

Bidirectional Expertise

Some expertise accepts data back from the LLM and stores it in Patrick's datastore:

• ✓ (stores to 'key') - Use send command to store response
• ✗ (one-way) - Expertise is read-only, no data storage

Common Patterns

Pattern 1: Simple Fetch

# No context needed
patrick-cli fetch daily-briefing

Pattern 2: Contextual Fetch

# Provide context variables
patrick-cli fetch decision-framing \
  --context '{"decision":"Should we expand to EMEA?"}'

Pattern 3: Full Bidirectional Workflow

# 1. Fetch expertise template
patrick-cli fetch daily-briefing --json > expertise.json

# 2. Process with LLM (pseudo-code)
# llm_process < expertise.json > response.json

# 3. Store response back to Patrick
patrick-cli send daily-briefing --data @response.json

# 4. Later, retrieve stored data
patrick-cli get daily-briefing

Troubleshooting

License Errors

Error: No license found at ~/.patrick/license.jwt

Solution: Get a license from patrickbot.io and save it:

patrick-cli set-license "YOUR_LICENSE_HERE"

Error: Expertise 'X' not in license

Solution: Visit patrickbot.io to upgrade your license or add the expertise

Fetch Errors

Error fetching expertise: 401 Unauthorized

Solution: Check your license is valid and not expired

Expertise 'X' not found

Solution: List available expertise with list to see what's accessible

Context Errors

Warning: Missing required context variables

Solution: Check list-prompts output for required context fields and provide them via --context

Signature Verification Errors

Error: HMAC signature verification failed for 'storage_key'

Cause: Files in ~/.patrick/data/ were manually edited, corrupted, or modified outside of Patrick CLI.

Solution: Delete the corrupted file and regenerate the data:

rm ~/.patrick/data/storage_key.json
# Re-run the command that generates this data

Prevention: Never manually edit files in ~/.patrick/ - all data is signed with HMAC-SHA256 tied to your customer identity.

Switching Licenses

Error: License belongs to a different account.

Cause: You're trying to set a license for a different account. Stored data is signed with your current customer identity and cannot be read under a different account.

Solution: Use set-license with --wipe-data to switch accounts:

patrick-cli set-license "NEW_LICENSE" --wipe-data

This deletes all stored data in ~/.patrick/data/ and saves the new license. You'll need to re-initialize afterwards.

Security

• Never commit license.jwt or environment files with real credentials
• Licenses are authenticated on every API call
• Expertise content is verified with SHA256 checksums
• Only expertise listed in your license are accessible

⚠️ IMPORTANT: Do not manually edit files in ~/.patrick/

All data stored in ~/.patrick/ is signed with HMAC-SHA256 tied to your verified customer identity:

• license.jwt - Your license token
• jwks_cache.json - Public key cache for license verification
• data/ - Stored expertise responses (if using bidirectional expertise)

Manual modifications will break signature verification and cause errors like:

Error: Stored data signature verification failed

If you need to reset your data, delete the specific file and re-run the command - don't edit it manually.

See {baseDir}/references/security.md for detailed information on Patrick's cryptographic signing model.

Reference Documentation

See {baseDir}/references/ for:

• prompts-api.md - Full API documentation (now uses /v1/expertise endpoints)
• prompt-format.md - Expertise structure specification
• llm-integration.md - LLM integration patterns
• security.md - Cryptographic signing and data integrity model

Advanced Usage

Custom Server URL

Point to a self-hosted Patrick server:

export PATRICK_SERVER_URL="https://patrick.mycompany.com"

Data Storage Path

Configure where Patrick stores local data (directory, not file):

export PATRICK_DATA_PATH=~/.patrick  # Default location

Patrick will store customer data files in $PATRICK_DATA_PATH/data/<storage_key>.json

Debug Logging

Enable detailed logging:

export RUST_LOG="patrick_cli=debug"