- Inside the Sandbox: Shell Execution Kills Reliability
Exploring why allowing agents arbitrary host shell access leads to non-deterministic failures and why strict Docker isolation is mandatory for production.
2 min
- Strategies, Not Model Ifs
Named strategy presets beat model-id ifs: cell budgets, history caps, and hybrid phases on one object.
6 min English
- When Smaller Cells Make Agents Worse
Smaller cells cut cells/turn but can exhaust step budgets; match policy to horizon, not aesthetics.
5 min English
- RLM Is Not Automatically Token-Efficient
History thrash erases recursive-loop savings; compaction plateaus input without fixing write thrash alone.
5 min English
- Token Efficiency: Code Fences vs Large Contexts
Analyzing how single-cell turn discipline impacts context growth and token costs.
2 min
Back