L

llm-agent

Autonomous hh.ru job-application agent built on the Mastra agent framework. Playwright attaches to an already-running Chrome over CDP, then the agent discovers and scores vacancies against profile.yaml, writes a Russian cover letter, submits, and delivers CV plus a follow-up message in chat. SQLite enforces one application per company; SUBMIT_ENABLED=false gives a full dry run.

Sandbox for harness and loop engineering with the @earendil-works/pi-coding-agent SDK: compares model providers, thinking levels, tool allowlists, extensions and session/runtime control, starting from a minimal custom-model example. Node 22.19+ and an API key for any one of Anthropic, OpenAI or Gemini.

Minimal LangChain v1 agent scaffold running against local Ollama models such as GLM-4.6, Llama 2 and Mistral. Demonstrates the agent-tool architecture with an extendable weather tool, .env-driven configuration, uv-managed Python 3.12 packaging, and Chrome DevTools MCP wiring for browser automation.

Environment, prompt and journal repository for running the NousResearch Hermes Agent CLI in a remote workspace: install locations, ~/.hermes provider configuration via hermes setup, and a running log of changes to the workspace. Documentation only - no application code.

Go, TDD-driven self-improving agent that applies to hh.ru vacancies using Cloudflare Workers AI for the main loop and the Cloudflare AI Gateway for complex tasks, with a rescue-model fallback. Ships a doctor health-check, dry-run mode, SQLite state, Russian cover-letter generation and an evolving skills directory.

Python hh.ru application agent built on pydantic-ai and the gsd-browser CLI. Reads what is actually rendered rather than calling the site API, screens vacancies against profile.yaml, and acts through a typed tool surface. Typer CLI, Textual TUI, structlog audit trail, SQLite dedup; the search-and-form architecture generalises beyond hh.ru.

A persistent-memory AI agent with cryptographically signed, hash-chained, tamper-evident memory. Pre-commit quality scoring quarantines prompt-injection; protected zones guard core records. Scans its own history for recurring gaps and proposes improvements for human review. Provider-agnostic (Claude, GPT, Gemini, DeepSeek, OpenRouter, Ollama).