agenticgogol
Community@agenticgogol
agenticgogol provides a staged pipeline for designing, building, evaluating, and deploying production-grade autonomous systems with rigorous failure analysis and judge calibration.
Agent Skills by agenticgogol
Showing 80 vetted skills indexed across 1 GitHub repositories.
agenticgogoldeploy-agent
Analyzes agent capabilities, scores deployment stacks, and guides step-by-step production deployment.
agenticgogolbackend-fastapi
Implements FastAPI backend endpoints exactly matching the API contract defined in design.md.
agenticgogoldatabase
Scaffolds SQLite, Postgres, or MongoDB connections, schemas, and query modules for application builds.
agenticgogoltrajectory-eval-new
Evaluates agent tool-use trajectories and attributes failures to the earliest diverging step.
agenticgogoldemo-agent-scaffolder
Scaffolds a deliberately flawed toy agent with seeded bugs for eval-toolkit teaching.
agenticgogolagent-mcp-real
Implements real MCP client-server integrations over stdio or SSE transports with capability negotiation.
agenticgogolcicd-integrator
Generates GitHub Actions workflows and scripts that gate pull requests on eval regressions.
agenticgogolagent-decision-design-pattern
Selects the agent design pattern for a usecase and writes an approval-gated design document.
agenticgogollint-and-typecheck
Runs mypy, ruff, tsc, and ESLint checks on Python and TypeScript code before integration.
agenticgogolagent-decision-memory
Decides memory and state storage per category for agent system designs.
agenticgogolvector-store
Scaffolds ChromaDB, FAISS, or Qdrant Cloud client, ingestion, and query code for RAG retrieval.
agenticgogolfrontend-streamlit
Scaffolds a single-file Streamlit app.py UI for Python-only demos and internal tools.
agenticgogolpick-requirements
Pins Python and Node dependencies from design.md into requirements.txt and package.json.
agenticgogolproject-debug
Diagnoses and iteratively fixes failing tests and runtime errors in built projects.
agenticgogolcost-optimize-new
Reduces agent inference cost through prompt refinement, model cascades, and CI-overlap verification.
agenticgogolproduction-monitor-setup
Generates a production monitoring configuration for deployed agents using tracing tools and calibrated judges.
agenticgogolgolden-dataset-builder
Generates a reviewed 50-example golden evaluation dataset in JSONL batches.
agenticgogolfailure-analyzer
Cluster eval failures by root cause into a structured failure report.
agenticgogolaxial-coding-new
Cluster open-coded failure notes into an approved taxonomy and ranked remediation backlog.
agenticgogoljudge-builder-new
Write binary PASS/FAIL LLM judge prompts grounded in eval specs for routed failure modes.
agenticgogolevaluator-design-new
Route failure modes to code checks or LLM judges and write pytest evaluators.
agenticgogolmetric-definition
Derives grounded evaluation metrics from concrete examples and writes eval/metrics.md.
agenticgogolagent-decision-single-vs-multi
Decides single-agent versus multi-agent topology through a structured five-question rubric.
agenticgogolagent-architecture-design
Interviews users to select and document a justified agent architecture pattern.
Frequently Asked Questions About agenticgogol
FAQPage SchemaWhat tasks can I accomplish with agenticgogol's skills?▼
You can design autonomous systems through staged decision documents, scaffold FastAPI/Next.js/Streamlit frontends and backends, build LangGraph or CrewAI implementations, create golden evaluation datasets, calibrate model judges against human labels, run retrieval scoring with Recall@k/MRR/NDCG, and deploy to production with monitoring via LangFuse or LangSmith.
Who is the target audience for these skills?▼
ML engineers, applied AI practitioners, and technical educators building production autonomous systems. The teaching pipeline (concept briefs, progressive notebooks, demo scaffolders) also serves instructors creating hands-on course material, while the eval suite targets teams needing rigorous quality gates before shipping.
How does the evaluation pipeline run in practice?▼
Start with eval-init-new to scaffold evals/ and draft must-always/must-never rules, then define tasks and metrics, build a golden set, select graders, and run baselines. Failures feed failure-analyzer and axial coding; judges get train/dev/test splits and calibration with Cohen's kappa before any CI gate trusts their scores.
What prerequisites and credentials are required?▼
A verified, working model API key is mandatory — require-api-key is the first gate in every pipeline and no mock mode exists. Vector stores (ChromaDB, FAISS, Qdrant Cloud), MCP server credentials, and database connections must also be verified before any code is written; missing credentials hard-stop the build.
How are costs and production quality controlled after deployment?▼
cost-optimize-new reduces inference spend via prompt refinement, few-shot calibration, task decomposition, KV-cache-friendly ordering, and model cascades — but only after the failure backlog closes. production-monitor-setup wires tracing through LangFuse or LangSmith, and eval-dashboard-new tracks judge calibration and CI gate metrics over time.