Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →sapientinc/praxist-diagnostic
Diagnose Praxist run health, artifact integrity, research-loop completeness, generation-scoped DIG/QD, PI/Gems/frontier/incubator consistency, peer memory freshness, diversity HHI, hardware utilization, LLM/runtime friction, task harness health, sustained low-performance causes, strongest variants/Pareto front, strong-variant lineage, and human-readable run reports.
Passed 11 of 17 test cases with the skill, and 6 of 17 without it.
and 7 more in the full report
8 fixed6 passed either way6 still failing2 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
A variant achieved top performance in `gen_2`, but was evaluated under an unauthorized preliminary debug mode and missing from `frontier_manifest.json`. Which missing candidate category should be assigned?
| Checked | Without skill | With skill |
|---|---|---|
| The diagnostic categorizes the candidate as immature. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill as a read-only expert diagnostic pass over an existing Praxist task/run. The job is to determine whether the run is healthy, whether the research loop is complete and internally consistent, whether useful evidence is being transferred correctly, whether resources or runtime systems are blocking progress, and whether weak performance has a process-level or task-harness cause.
The skill is task-agnostic. Do not encode domain-specific rules, metrics,
thresholds, paths, or project names into the diagnosis. Use the task's own
task.yaml, evaluator outputs, manifests, logs, and documented metrics to
interpret results. If a fact is not available in artifacts, label it unknown
rather than inventing a task-specific assumption.
Artifact source rule: distinguish current fact owners from views. Measured
result/finding summaries, frontier/frontier_manifest.json, committed
gems/gems_state.json, and gen_N/generation_boundary.json are canonical
state. Leaderboards, PI evidence packs, PI agendas, prompt layouts, rendered
prompts, diagnostics, and behavior reports are derived views or audit
snapshots. Use audit snapshots to reconstruct what an agent saw at the time;
use canonical state to judge current promotion, Gems, evidence-transfer, and
resume truth. If artifact_semantics exists, report role/status/source fields;
if it is absent, treat the artifact as a legacy file and infer cautiously.
Discover compact evaluator summaries recursively under results/**/. Recognize
summary.json, evaluation_summary.json, eval_summary.json,
tiered_eval_summary.json, custom_*_tiered_eval_summary.json, and the
compatibility name result_summary.json. Reconcile their structured
lane/maturity/diagnostic metadata with auto-materialized findings; do not assume
one fixed directory depth or one task-specific filename.
When available, separately inspect task-defined preliminary, aligned, partial,
repair, failed-but-informative, ablation, diagnostic, lower-stage, or
late-after-boundary evidence. They can indicate missed opportunities or broken
promotion flow, but they are not clean frontier/Gems facts. For expensive
tasks, diagnostics must not mix preliminary triage results with aligned ranking
evidence, and must check whether aligned evaluation kept the same or
near-complete data/evaluation coverage while reducing training or optimization
effort first. Resolve labels from the task maturity policy; never infer global
maturity from a familiar-looking tier name.
Late-after-boundary signals are result summaries written after
test cases passed, vs 6 / 17 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.