zvirb

---
name: CRM Entity Extraction
description: Workflow-driven skill that bridges Gmail data extraction logic to Google Sheets or Google Contacts via Composio.
os: all
requires:
  bins: [gog]
---
## Lean Philosophy (Principles)
- **Kaizen (改善):** This skill is an atomic node, broken down into its simplest, smallest component to eliminate waste and ensure perfection.
- **Standardized Work (Hyojun Sagyo):** This node represents the most efficient, standardized path for this specific task before automation.
- **Jidoka (自働化):** This node includes autonomous defect detection. It will stop immediately and report if it cannot achieve the expected outcome.

# CRM Entity Extraction

This skill extracts structured data (persons, organizations, dates) from raw text and appends it to a CRM spreadsheet.

## Cognitive Directives
WHEN [A business-related email or note containing CRM data is received]
THEN [
  Execute the following Jidoka-validated loop:
  1. Execute `llm_extract_json` (Sub-Agent) to extract structured entities (name, org, date).
     - **Verification Step (Jidoka):** Check if the sub-agent returns a valid JSON object matching the requested entity schema. IF it returns unstructured text or invalid JSON, instruct the sub-agent to format correctly and retry.
  2. Execute the native terminal command `gog sheets append <spreadsheetId> <range> --values-json '[["..."]]'` to add the extracted JSON row to the CRM spreadsheet.
     - **Verification Step (Jidoka):** Verify the `gog sheets append` command returns a successful JSON confirmation. IF the API request fails, wait 3 seconds and retry (max 3 times). IF it still fails, report the error and STOP.
]

## Expected Output
A JSON summary of the extracted data and the append result.

---
name: capture-classification
description: Workflow-driven skill that routes unstructured audio transcripts or quick notes to Google Tasks or LanceDB based on urgency.
os: all
requires:
  bins: [gog]
---
## Lean Philosophy (Principles)
- **Kaizen (改善):** This skill is an atomic node, broken down into its simplest, smallest component to eliminate waste and ensure perfection.
- **Standardized Work (Hyojun Sagyo):** This node represents the most efficient, standardized path for this specific task before automation.
- **Jidoka (自働化):** This node includes autonomous defect detection. It will stop immediately and report if it cannot achieve the expected outcome.

# Capture Classification

This skill acts as a semantic router for inbound text, evaluating intent and urgency before routing to the appropriate destination.

## Cognitive Directives
WHEN [Unstructured audio transcript or note text is captured]
THEN [
  Execute the following Jidoka-validated loop:
  1. Execute `llm_classify_intent` (Sub-Agent) with categories: ["Actionable", "Informational"].
     - **Verification Step (Jidoka):** Verify the sub-agent returns exactly one of the requested categories. IF it returns an invalid or hallucinated category, instruct it to correct the output and retry.
  2. IF "Actionable", Execute the native terminal command `gog tasks add @default --title "..."` with the summary.
     - **Verification Step (Jidoka):** Check if the native command returns a successful JSON response. IF it fails, wait 3 seconds, and retry. IF it still fails after 3 attempts, report the error and STOP.
  3. IF "Informational", Execute `vector_store_upsert` to save it as reference material.
     - **Verification Step (Jidoka):** Verify the vector store confirms a successful upsert. IF it fails, report the error and STOP.
]

## Expected Output
A JSON log confirming the routed destination and action taken.

---
name: calendar-guard
description: Workflow-driven skill that autonomously defends your schedule to manage cognitive load and prevent burnout.
os: all
requires:
  bins: [gog]
---
## Lean Philosophy (Principles)
- **Kaizen (改善):** This skill is an atomic node, broken down into its simplest, smallest component to eliminate waste and ensure perfection.
- **Standardized Work (Hyojun Sagyo):** This node represents the most efficient, standardized path for this specific task before automation.
- **Jidoka (自働化):** This node includes autonomous defect detection. It will stop immediately and report if it cannot achieve the expected outcome.

# Calendar Guard

This skill evaluates schedule density and automatically injects recovery blocks into your Google Calendar when needed to prevent burnout.

## Cognitive Directives
WHEN [Requested to check calendar for burnout risks OR running on a daily schedule]
THEN [
  Execute the following Jidoka-validated loop:
  1. Execute the native terminal command `gog calendar events primary --from "..." --to "..." --json` to retrieve events for the next 24 hours.
     - **Verification Step (Jidoka):** Check if the output is a valid JSON array. IF it fails or returns an error, wait 3 seconds, and retry. IF it still fails, report the error and STOP.
  2. Execute `llm_identify_conflicts` (Sub-Agent) to analyze the event list for periods of high cognitive load.
     - **Verification Step (Jidoka):** Verify the sub-agent returns a structured list of detected high-load periods. IF it returns raw unformatted text or an error, request the sub-agent to format the output correctly and retry.
  3. IF high load is detected, Execute the native terminal command `gog calendar create primary --summary "Recovery Block"` to inject "Recovery Block" events to protect decompression time.
     - **Verification Step (Jidoka):** Verify that the `gog calendar create` command returns a successful JSON confirmation. IF it fails, log the failure for that specific block and continue to the next block, rather than halting the entire process.
]

## Expected Output
A JSON log of any recovery blocks injected.

---
name: backlog-grooming
description: Workflow-driven skill that autonomously manages stale items in Google Tasks.
os: all
requires:
  bins: []
---
## Lean Philosophy (Principles)
- **Kaizen (改善):** This skill is an atomic node, broken down into its simplest, smallest component to eliminate waste and ensure perfection.
- **Standardized Work (Hyojun Sagyo):** This node represents the most efficient, standardized path for this specific task before automation.
- **Jidoka (自働化):** This node includes autonomous defect detection. It will stop immediately and report if it cannot achieve the expected outcome.

# Backlog Grooming

This skill directs the agent to trigger the native programmatic backlog grooming plugin to maintain a clean workspace.

## Cognitive Directives
WHEN [Triggered by schedule or user request to groom backlog] 
THEN [
  Execute the following Jidoka-validated loop:
  1. **Execute Node:** Execute the `workflow_backlog_grooming` plugin tool with empty JSON `{}` to trigger the grooming process.
  2. **Verification Step (Jidoka):** Check if the tool returns a valid JSON response with `{"success": true}` or an error message. IF it returns an error or hallucinated output, report the error, wait 3 seconds, and retry (max 3 times). IF it still fails, stop and notify the user.
]
## Expected Output
A JSON summary of the groomed tasks returned by the plugin.