sancovp
8 verified skills
application/compoctopus/skills/behavioral-assertions
# Behavioral Assertions You write behavioral assertions — tests that PROVE an agent works by running it. ## What Makes a Behavioral Assertion A behavioral assertion calls `execute()` with real inputs and verifies real outputs. It is NOT a unit test of internal functions. It is a test of the WHOLE agent. ## Format ```json { "description": "Responds with greeting containing the user's name", "setup": "agent = make_greeter()\ninput_data = {'name': 'Alice'}", "call": "result = await agent
development
document-market-thesis
Create comprehensive market thesis documentation for new product categories
development
application/compoctopus/skills/prd-authoring
# PRD Authoring You write Product Requirements Documents (PRDs) for the Compoctopus pipeline. ## PRD Fields (ALL required) | Field | Type | Purpose | |-------|------|---------| | `name` | snake_case string | Project identifier | | `description` | string | What this agent/system does | | `architecture` | "Chain" or "EvalChain" | Chain = sequential, EvalChain = loop with evaluator | | `links` | LinkSpec[] | The steps in the chain | | `types` | TypeSpec[] | Data structures the agent uses | | `be
devops
base/sdna/skills/understand-sdna
# understand-sdna **WHAT:** Gnostic agent workflow DSL with LangGraph as native execution substrate. Ariadne (threading) + Poimandres (generation) = SDNA spiral. **WHEN:** Building agent workflows with typed composition, context threading, human-in-the-loop patterns, or LangGraph integration. **HOW:** Use the decision tree below, then read the relevant resources. --- ## Decision Tree: What to Build ``` Is this continuous improvement / optimization loop? ├── YES → SDNA^F (SDNAFlowchain) │
development
application/compoctopus/skills/chain-architecture
# Chain Architecture You design agent chains for the Compoctopus pipeline. ## Chain Types ### Chain (sequential) Links execute in order. Output of link N feeds into link N+1. ``` Link1 → Link2 → Link3 → result ``` Use when: each step has a clear input/output, no retry logic needed. ### EvalChain (loop with evaluator) Chain runs, then an evaluator approves/rejects/retries. ``` Link1 → Link2 → Link3 → Evaluator ↓ approve / reject / retry
devops
spawn-ralph
Spawn isolated Ralph agents (claude -p) for parallel subtasks
data-ai
crystal-ball
Crystal Ball ontological navigation — composing, folding, and scrying spaces
tools
map
Compile tasks through Map — queue-driven enrich/instance loop. Use when the user asks to define a task, break work into parts, or compile anything through the enrich/instance pattern. Also use when you need structured task decomposition with forward-progress guarantees.
development