A study vault
Eighteen open-source agent codebases, read sideways: how memory compression actually works across eight of them, why Strix uses XML tool calls where Claude Code uses native ones, what guardrails layer in production agents.
This is a personal notebook. The orientation below is the path I took; you are welcome to wander.
18 projects · 12 concepts · 10 insights · 26 terms
Three ways in
By project
The codebase as the unit. Memory, tools, sandbox, all in one place.
Browse projects →By concept
The idea across codebases. Memory compression, agent loops, caching.
Browse concepts →By insight
The clever bit that doesn't fit a textbook chapter.
Insight gallery →Featured tonight
Pre-baked reading orders. Each tour is a 30–90 min study session with explicit takeaways.
Most pedagogically dense projects
Ranked by how much you learn per page. Drill-down docs preserved as styled HTML.
Anthropic's official agentic CLI. Streaming tool calls, prompt caching, thinking signatures, multi-agent subagents, slash commands.
All-hands AI v0 — autonomous software engineer agent. Event-sourced state, microagents, controller-level guardrails.
Open-source 'AI hacker' for autonomous pentesting. XML tool format, markdown-as-skills, LLM-based dedupe, module-level agent graph.
Latest insights
One non-obvious trick per card. The clever bits that don't fit a textbook.
Generic "summarize this conversation" loses the bits the agent needs to keep working. Mature systems enumerate preservation rules.
Decouples the agent's loop from its expertise. Domain experts contribute via PR; the loop almost never changes; the library evolves weekly.
Two pentest reports describing the same SQL injection with different payloads aren't textually similar — but they should dedupe. Hashing fails; LLM reasoning works.
Survives every refactor, no marker objects to remember to add. Lets dozens of contributors compose one prompt without breaking cross-org cache reuse.