|
|
b7785c106b
|
Alfred Agent v2.0 — production-grade runtime with 14 service modules
Core rewrite:
- agent.js: streaming, mid-loop compaction, context tracking, skill matching
- providers.js: streamQuery() method with SSE events for Anthropic
- index.js: v2 HTTP server with 12 endpoints (/health, /chat/stream, /tokens, etc.)
- cli.js: --stream flag, /tokens, /context, /skills commands
New services:
- tokenEstimation.js: multi-strategy token counting with context warnings
- messages.js: typed message system (user/assistant/system/compact/tombstone)
- compact.js: 4-tier compaction engine (micro → auto → memory → cleanup)
- contextTracker.js: file/git/error/discovery tracking with scoring
- steering.js: per-tool safety rules (OWASP-aligned)
- skillEngine.js: SKILL.md parser with keyword triggers and hot reload
- agentFork.js: sub-agent spawning with persistent task tracking
- redact.js: Aho-Corasick secret scrubbing from tool outputs
- intent.js, memory.js, permissions.js, costTracker.js, modelRouter.js, doctor.js
|
2026-04-08 19:01:30 -04:00 |
|
Alfred
|
e89ce12316
|
Initial commit: Alfred Agent - multi-provider AI agent harness
- 7 source files: providers, tools (14), prompt, session, agent (core loop), cli, http
- Multi-provider: Anthropic, OpenAI, Groq
- 14 built-in tools for file ops, web, code execution
- HTTP API + CLI interface
|
2026-04-07 11:39:57 -04:00 |
|
