Search results

research(patterns): typicality bias in human eval data as a known failure mode for LLM evaluation

Status: Open.

#683 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(patterns): add llm-diversity pattern encoding Verbalized Sampling for output diversity in LLM apps

Status: Open.

#682 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(skills): apply Verbalized Sampling prompt pattern to directive brainstorming and ideation tasks

Status: Open.

#681 In deftai/directive;

· visionik opened on Apr 25, 2026

Reduce overhead for vBRIEF phase lifecycle completion

Status: Open.

#680 In deftai/directive;

· visionik opened on Apr 25, 2026

research(patterns): per-task VM isolation as architecture standard for agent platforms

Status: Open.

#679 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(patterns): add agent network egress security standards to coding/security guidance

Status: Open.

#678 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(patterns): add agent-sandbox pattern covering VM isolation, MITM proxy, and DNS exfil defense

Status: Open.

#677 In deftai/directive;

· visionik opened on Apr 25, 2026

docs(patterns): encode 'small model inversion' as a cost/performance pattern for LLM projects

Status: Open.

#676 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(patterns): add long-context LLM architecture pattern with complexity-class taxonomy

Status: Open.

#675 In deftai/directive;

· visionik opened on Apr 25, 2026

docs(safety): name and define 'context rot' as an explicit agent failure mode

Status: Open.

#674 In deftai/directive;

· visionik opened on Apr 25, 2026

feat(swarm): add recursive sub-agent delegation pattern to deft-swarm

Status: Open.

#673 In deftai/directive;

· visionik opened on Apr 25, 2026

docs(patterns): encode lazy-loading as named 'recursive context exploration' principle

Status: Open.

#672 In deftai/directive;

· visionik opened on Apr 25, 2026