chore(agents): update agents.md

AGENTS.md

@@ -1,9 +1,104 @@
-## Role
+# AGENTS.md
 
-Write code like a Go Senior Developer. 
-Use the best practices, write modern idiomatic code, and handle all errors. 
-Keep an eye on security.
+This file provides guidance to LLM agents when working with code in this repository.
 
-## Context
+## Project Overview
 
-- go 1.25.5
+Trackers is a Go application that aggregates torrent tracker links from multiple sources (HTTP, HTTPS, and local files). It:
+- Polls configured sources at regular intervals
+- Deduplicates and validates tracker links
+- Caches results to disk for resilience
+- Serves aggregated tracker lists via HTTP API
+- Runs a single HTTP server with minimal dependencies
+
+This is an educational project written in Go 1.25.5.
+
+## Development Commands
+
+### Build
+```bash
+go build -o trackers ./main.go
+```
+
+### Run
+```bash
+go run main.go -config config.toml
+```
+
+### Test
+The codebase has no tests currently. Tests can be added with:
+```bash
+go test ./...
+```
+
+To run a single test:
+```bash
+go test -run TestName
+```
+
+### Lint
+```bash
+go fmt ./...
+go vet ./...
+```
+
+### Release Build
+Uses goreleaser for cross-platform builds (Linux amd64/arm64):
+```bash
+goreleaser build --snapshot --clean
+```
+
+## Architecture
+
+**Single File Design**: All code is in `main.go` with clear functional separation:
+
+1. **Config System** (`loadConfig`): Parses TOML configuration with defaults
+    - `port`: HTTP server port (default: 8080)
+    - `cache_dir`: Directory for caching tracker lists (default: cache/)
+    - `poll_interval`: How often to refresh sources (default: 60m)
+    - `sources`: Array of URLs/file paths to fetch tracker lists from
+
+2. **Aggregator** (type `Aggregator`): Thread-safe in-memory deduplication
+    - Maintains per-source tracker sets using `sync.RWMutex`
+    - `Update()`: Stores new tracker list for a source
+    - `List()`: Returns combined sorted list across all sources
+
+3. **Polling System** (`pollSource`/`runOnce`): Background goroutine per source
+    - Fetches source on startup and at configured intervals
+    - Updates aggregator and writes to cache on success
+    - Graceful shutdown on SIGINT/SIGTERM
+
+4. **HTTP Handler** (`/list` endpoint): Returns deduplicated tracker list as plain text
+    - Links separated by double newlines
+    - Read timeouts enforce reasonable request handling
+
+5. **Source Fetching** (`fetchSource`): Pluggable source handlers
+    - **HTTP/HTTPS**: Makes requests with context support and timeout (15s)
+    - **File**: Reads local files via `file://` URLs
+    - Response is parsed line-by-line
+
+6. **Link Validation** (`normalizeLinks`/`isValidTrackerLink`):
+    - Strips whitespace and empty lines
+    - Validates URL format and supported schemes: `http`, `https`, `udp`, `ws`, `wss`
+    - Deduplicates via map-based set
+
+7. **Caching** (`writeCache`/`loadCachedLinks`): SHA1-hashed filenames in `cache_dir/`
+    - Enables graceful degradation if source becomes unavailable
+    - Filenames are hex-encoded SHA1(source_url) + ".txt"
+
+## Key Design Decisions
+
+- **No external dependencies except go-toml**: Keeps binary small and build simple
+- **Simple HTTP server**: Uses stdlib `net/http` instead of frameworks
+- **Per-source goroutines**: Allows independent polling without blocking
+- **RWMutex for reads**: Readers don't block each other when listing trackers
+- **Context propagation**: Respects shutdown signals in all async operations
+- **Line-based parsing**: Flexible input format (handles various tracker list formats)
+
+## Testing Notes
+
+The project follows Go conventions but has no test files. Consider adding tests for:
+- Link validation edge cases
+- Config parsing with invalid inputs
+- Concurrent aggregator updates
+- Cache file I/O