agent: optimize and evaluate local-model Kubernetes exploration - #185

Merged

hellices merged 34 commits into

mainfrom

agents/prompt-optimization-local-models

Aug 4, 2026

Commits on Aug 4, 2026

feat(agent): eval-measured diagnosis rules in the small-profile prompt (#177)

hellices
and
Copilot
committed
7cd2815
feat(agent): three measured prompt iterations against qwen3:4b failures (#177)

hellices
and
Copilot
committed
52a6ade
feat(agent): small-profile worked example teaches the 137 discrimination (#177)

hellices
and
Copilot
committed
4044fe7
feat(agent): project current pod health and PVC failures (#177)

hellices
and
Copilot
committed
0b5619a
feat(agent): diagnose Deployment rollouts in one tool call (#177)

hellices
and
Copilot
committed
29df1db
test(evals): credit compound rollout diagnosis evidence (#177)

hellices
and
Copilot
committed
e49eb9d
test(evals): recognize compound PVC evidence and equivalent rollout wording (#177)

hellices
and
Copilot
committed
9815224
fix(agent): preserve completed and concurrent diagnostic evidence (#177)

hellices
and
Copilot
committed
a7c2acd
docs: design standardized AKS model evaluation matrix (#176)

hellices
and
Copilot
committed
f3c50d5
fix(evals): support slow local-model inference timeouts (#176)

hellices
and
Copilot
committed
bb73ac0
docs: center model matrix on personal local devices (#176)

hellices
and
Copilot
committed
0a285f8
feat(evals): add persistent conversational journey harness (#176)

hellices
and
Copilot
committed
ced122f
fix(evals): preserve core resource aliases in live journeys (#176)

hellices
and
Copilot
committed
25d84a2
docs: publish detailed conversational and live AKS eval results (#176)

hellices
and
Copilot
committed
0071ae4
fix(evals): enforce live journey boundaries and rollout identity (#176)

hellices
and
Copilot
committed
57c5a54
test(evals): require invalid image claim in live correction (#176)

hellices
and
Copilot
committed
df8a7c5
fix(evals): address Copilot review findings (#176)

hellices
and
Copilot
committed
6b210a7
fix(evals): address Copilot review round two (#176)

hellices
and
Copilot
committed
0a0d343
fix(agent): preserve rollout attribution at small-profile cap (#176)

hellices
and
Copilot
committed
9366cc9
fix(evals): degrade optional live evidence failures safely (#176)

hellices
and
Copilot
committed
1118935
fix(evals): grade conversational evidence per turn (#176)

hellices
and
Copilot
committed
5d37d96
fix(evals): tighten strict journey and PVC evidence reporting (#176)

hellices
and
Copilot
committed
758e398
fix(evals): verify live namespace ownership before reads (#176)

hellices
and
Copilot
committed
b823012
test(evals): require explicit triage prioritization (#176)

hellices
and
Copilot
committed
57d87b5
fix(evals): measure triage priority and namespace discipline (#176)

hellices
and
Copilot
committed
501a97d
fix(agent): preserve workload status on child list failures (#176)

hellices
and
Copilot
committed
8bf3433
fix(evals): bound rollout summaries and live defaults (#176)

hellices
and
Copilot
committed
5cdd544
fix(agent): keep compact evidence for every rollout pod (#176)

hellices
and
Copilot
committed
cdf00fe
fix(evals): reject ambiguous targets and malformed namespaces (#176)

hellices
and
Copilot
committed
6ce58d2
fix(evals): normalize cluster-scoped resource aliases (#176)

hellices
and
Copilot
committed
e988693
fix(agent): isolate rollout pod read failures (#176)

hellices
and
Copilot
committed
9aaf1fe
fix(evals): enforce live boundaries and priority evidence (#176)
hellices
committed
a097b54
fix(agent): keep probe guidance grounded (#176)

hellices
and
Copilot
committed
618bc2e
fix(evals): require a live evaluation run id (#176)

hellices
and
Copilot
committed
093fcf7

Navigation Menu

agent: optimize and evaluate local-model Kubernetes exploration - #185

agent: optimize and evaluate local-model Kubernetes exploration#185
hellices merged 34 commits into
mainfrom
agents/prompt-optimization-local-models

Commits on Aug 4, 2026

feat(agent): eval-measured diagnosis rules in the small-profile prompt (#177)

feat(agent): three measured prompt iterations against qwen3:4b failures (#177)

feat(agent): small-profile worked example teaches the 137 discrimination (#177)

feat(agent): project current pod health and PVC failures (#177)

feat(agent): diagnose Deployment rollouts in one tool call (#177)

test(evals): credit compound rollout diagnosis evidence (#177)

test(evals): recognize compound PVC evidence and equivalent rollout wording (#177)

fix(agent): preserve completed and concurrent diagnostic evidence (#177)

docs: design standardized AKS model evaluation matrix (#176)

fix(evals): support slow local-model inference timeouts (#176)

docs: center model matrix on personal local devices (#176)

feat(evals): add persistent conversational journey harness (#176)

fix(evals): preserve core resource aliases in live journeys (#176)

docs: publish detailed conversational and live AKS eval results (#176)

fix(evals): enforce live journey boundaries and rollout identity (#176)

test(evals): require invalid image claim in live correction (#176)

fix(evals): address Copilot review findings (#176)

fix(evals): address Copilot review round two (#176)

fix(agent): preserve rollout attribution at small-profile cap (#176)

fix(evals): degrade optional live evidence failures safely (#176)

fix(evals): grade conversational evidence per turn (#176)

fix(evals): tighten strict journey and PVC evidence reporting (#176)

fix(evals): verify live namespace ownership before reads (#176)

test(evals): require explicit triage prioritization (#176)

fix(evals): measure triage priority and namespace discipline (#176)

fix(agent): preserve workload status on child list failures (#176)

fix(evals): bound rollout summaries and live defaults (#176)

fix(agent): keep compact evidence for every rollout pod (#176)

fix(evals): reject ambiguous targets and malformed namespaces (#176)

fix(evals): normalize cluster-scoped resource aliases (#176)

fix(agent): isolate rollout pod read failures (#176)

fix(evals): enforce live boundaries and priority evidence (#176)

fix(agent): keep probe guidance grounded (#176)

fix(evals): require a live evaluation run id (#176)