All pull requests

Pull requests

• ci: connect SkillEvaluator to NVSkills catalog validation

  #440·chrisknvidia opened Nov 9, 2026Collaborator·9 of 12 tasks·

  ·
  1

• feat(ci): add LLM security, gateway agents and benchmark PRs

  #439·rng1995 opened Nov 9, 2026Collaborator·15 tasks done·

  ·
  14

• feat(tier3): report activation independently of task outcomes

  #438·rng1995 opened Nov 9, 2026Collaborator·8 of 9 tasks·

  ·
  1

• fix(tier2): expose collection coverage and transactional rebuilds

  #436·rng1995 opened Nov 9, 2026Collaborator·8 of 9 tasks·

  ·
  1

• feat(datasets): add lossless preview-first dataset migration

  #435·rng1995 opened Nov 9, 2026Collaborator·8 of 9 tasks·

  ·
  5

• fix(cli): preflight before paid autopilot dataset generation

  #434·rng1995 opened Nov 9, 2026Collaborator·8 of 9 tasks·

  ·
  1

• fix(reporting): align routing summaries and unavailable-score conclusions

  #433·rng1995 opened Nov 9, 2026Collaborator·8 of 9 tasks·

  ·
  1

• docs(grading): define evidence-grounded qualitative review contract

  #432·rng1995 opened Nov 9, 2026Collaborator·7 of 9 tasks·

  ·
  6

• Add advisory enforcement mode to Tier 1 reusable workflow

  #431·pt-bhavsar opened Nov 9, 2026·8 of 9 tasks·

  ·
  1

  18

• feat(tier3): add opt-in tolerance for baseline-only agent timeouts

  #430·rng1995 opened Nov 9, 2026Collaborator·

  ·

• feat(tier3): save RAGAS goal-accuracy requests and replies in judge_attempts.json

  #429·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier3): count edits, searches and MCP calls as productive and leave planning calls out of tool efficiency

  #428·rng1995 opened Nov 9, 2026Collaborator·

  ·

• feat(tier3): report token usage per case and agent

  #427·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier2): raise the intra-skill chunk ceiling to 1,024

  #426·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier3): score skill_efficiency and unlabeled skill_execution as N/A, not 1.0

  #425·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(reporting): name the skill in single-skill JSON reports and record the evaluator version

  #424·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier2): chunk shell scripts in one pass instead of once per function header

  #423·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier2): parse skill frontmatter with bounded YAML

  #422·rng1995 opened Nov 9, 2026Collaborator·

  ·
  2

• fix(tier1): report dependency advisories as structured findings

  #421·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier3): keep the Rich progress frame at a fixed height

  #420·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(tier3): reuse a run's saved feedback when view regenerates its report

  #419·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(reporting): show judge-mixed lift exclusions on Tier 3 evaluator cards

  #418·rng1995 opened Nov 9, 2026Collaborator·

  ·

• docs(reports): explain completed-agent results in incomplete multi-agent Tier 3 runs

  #417·rng1995 opened Nov 9, 2026Collaborator·

  ·

• fix(reporting): show recovered timeout attempts accurately in Tier 3 attempt details

  #416·rng1995 opened Nov 9, 2026Collaborator·

  ·
  1

• fix(tier1): retry an invalid rubric judge reply once and name the problem

  #415·rng1995 opened Nov 9, 2026Collaborator·

  ·