Execute qualquer Skill no Manus
com um clique

Execute qualquer Skill no Manus com um clique

phase2-screening

Estrelas6

Forks1

Atualizado7 de julho de 2026 às 07:52

Run Phase 2 (SCREENING) of the A2MC calibration workflow as the offline agent — the human-in-the-loop analog of the orchestrator's `_run_screening()` / `reasoning.analyze_screening_results()`. Rank the ensemble against the validation targets (RMSRE, targets-satisfied), find the best / lowest-cost / most-targets cases, read the error/bias patterns, and route the round into Phase 3 diagnosis. Use when the user says "screen the ensemble", "rank against targets", "which case is best", "run Phase 2", "how many targets are met", after Phase 1 sensitivity.

Instalação

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Executar no Manus

Fonte

jingtao-lbl

jingtao-lbl/A2MC-elm

Abrir repositório GitHub Ver repositórios do creator

Download

Executar no Manus

Ocupações relacionadasSOC

Baseado na classificação ocupacional SOC

Desenvolvedores de softwareInformática e Matemática·SOC 15-1252

SKILL.md

readonly

name	phase2-screening
description	Run Phase 2 (SCREENING) of the A2MC calibration workflow as the offline agent — the human-in-the-loop analog of the orchestrator's `_run_screening()` / `reasoning.analyze_screening_results()`. Rank the ensemble against the validation targets (RMSRE, targets-satisfied), find the best / lowest-cost / most-targets cases, read the error/bias patterns, and route the round into Phase 3 diagnosis. Use when the user says "screen the ensemble", "rank against targets", "which case is best", "run Phase 2", "how many targets are met", after Phase 1 sensitivity.
modes	{"requires_fates":false,"nutrient_pathway":"any","scope":["calibration"],"summary":"Offline analog of Phase 2 (rank ensemble vs validation targets). Applies in every calibration mode."}

Phase 2: Screening (offline agent)

The offline analog of _run_screening(). Online, reasoning.analyze_screening_results() reads the ranked ensemble automatically; offline, YOU read it — identify best / lowest-cost / most-targets cases, the per-target bias (over/under-estimate), edge parameters in the top cases, and PFT trade-offs, then set up what Phase 3 must explain.

Floor, not ceiling. Ranking + best-case is the online floor. As the offline agent, go further: inspect the whole cost distribution (not just top-N), test whether a "best" case is real or a contamination artifact before trusting it, look for equifinality (different parameter sets, same cost), or compare this round's frontier against a prior round. Produce the online ranking, then interrogate it.

Inputs (from Phase 1): extracted per-case data + Morris rankings. Deliverable: ranked cases (RMSRE + targets satisfied), the best/lowest-cost/most-targets cases, error-pattern read → routed to phase3-diagnosis.

Step 1 — rank the ensemble (reads existing extracted outputs — no new simulations)

phases/phase2_screening/screen_ensemble.py --data-dir $A2MC_EXTRACTED_DATA --top-n 100. Per-target relative error → composite RMSRE via tools/optimize_function.py / tools/cost_functions.py (shared tools/evaluate_case.py); counts targets within tolerance (default ±20%). Outputs ranked indices, composite cost, per-target errors, n_satisfied.

Footgun — --max-case-num contamination guard. When experiment cases (e.g. _exp / #5001+) share the extract dir with the Morris ensemble, pass --max-case-num 4890 so screening ranks only the ensemble. Without it, foreign cases inflate the top-N. (Same guard the compare-calibration-rounds cap enforces.)

Footgun — index vs case number. Screening Set_ID/indices can be position+1, not the real case number. Use the JSON best_case_num + the _results.txt Sim_ columns, not the indices file, before quoting a "best case" (see diagnose-forensics).

Step 2 — read the patterns (you are the reasoning)

Mirror analyze_screening_results: pft_performance (cases within uncertainty, quality), target_bias (UNDEREST/OVEREST + consistency), edge_parameters (top cases pinned at bounds), pft_tradeoffs (shared-parameter conflicts), priority_targets for diagnosis. Optional QA: phases/phase2_screening/compare_biomass_topcases.py --top-n 50 (top-N biomass vs targets panel).

Step 3 — log and route

Log via calibration-log (phase log → PhaseLogger.log_screening): best/lowest-cost/most-targets cases, targets met, bias patterns, priority targets. Route to phase3-diagnosis with the priority targets + candidate base cases.

Related skills / next phase

Canned single-round report (graphs + eval + μ)* → summarize-calibration-round. Cross-round frontier → compare-calibration-rounds. This skill decides + routes; those render.
A "best" case looks too good → diagnose-forensics (triage before trusting).
Next: phase3-diagnosis.

Changelog

2026-07-02: Created — offline Phase 2 routine mirroring reasoning.analyze_screening_results(); drives screen_ensemble, delegates figures to summarize-/compare-calibration-round, routes to phase3-diagnosis.

Mais deste repositório

mesmo repositório

a2mc-init

jingtao-lbl/A2MC-elm

First-run interactive setup for the offline agent — the official "getting started" flow the FIRST time someone uses A2MC in a repo/site. Interview the user (is a2mc_config.sh set up, where is the E3SM/ELM-FATES checkout, FATES on/off, carbon-only vs nutrient-enabled PARTEH, ECA vs RD, site name/location, PFTs, calibration targets), verify the checkout + RAG milestone, then create and populate the use case (site config + validation targets + parameter list) and hand off to phase0-design. Use when the user says "set up A2MC", "first time using A2MC", "help me get started / onboard me to A2MC", "initialize a new site / use case", "configure A2MC for my site", "I want to calibrate a new site". DISTINCT from onboard-session (which resumes an ALREADY-configured setup).

2026-07-076

add-skill

jingtao-lbl/A2MC-elm

Scaffold + register (or retire + de-register) an A2MC skill so the registry steps are never half-done. Use when the user or agent wants to create a new capability — "add a skill", "create/scaffold a skill", "make this reusable as a skill", "distill X into a skill" — OR to remove one — "remove/retire/delete the X skill". Writes (or deletes) the SKILL.md, and registers (or de-registers) it across all THREE human-facing registries (the README 'Current skills' table, docs/a2mc_reference/skills_catalog.md, AND the AGENTS.md 'At a glance' table), runs the mechanical drift check, and stops for human review before commit. Human-gated.

2026-07-076

arm-hpc-monitoring

jingtao-lbl/A2MC-elm

Set up real-time monitoring of an active A2MC HPC ensemble or experiment on Perlmutter at session start (CLAUDE.md Rule

2026-07-076

calibration-log

jingtao-lbl/A2MC-elm

Log interactive calibration work for a site under use_cases/{site}/memory/logs/. Two log types — a PHASE log (a diagnosis, screening, experiment design/result, refinement — recorded the same way the autonomous agent does, via PhaseLogger, so both modes' logs synthesize together) or a free-form SESSION log (exploratory work that does not map to a specific phase). Use when the user wants to record calibration or exploration work — "log this", "log this phase / diagnosis / experiment", "log this calibration session", "write a session log", "record what I explored".

2026-07-076

compare-calibration-rounds

jingtao-lbl/A2MC-elm

Compare A2MC calibration rounds (R1, R2, …) against each other and the validation targets — top-N best-achievable biomass per round, per-target Morris μ* sensitivity, P-pool / cross-regime overlays. Use when the user asks to "compare rounds", "update the multi-round figure", "refresh the R1-RN comparison", "which round is best", "top sensitive parameters per round", or to regenerate any `multiround_*` / cross-round figure after new cases complete. Codifies the multiround bundle pipeline + the footguns (screening contamination, case-name patterns, μ* ranking, param-set mismatch, partial-ensemble caveats).

2026-07-076

inject-knowledge

jingtao-lbl/A2MC-elm

Inject a human-originated domain fact (a discovery, a parameter insight, a mechanism/relationship) into A2MC's curated knowledge so the calibration agent surfaces it — placing it correctly across the up-to-three channels (site discoveries.json, generic parameters.json, curated_relationships.yaml), validating, and rebuilding the graph. Use when the user says "add this finding/parameter knowledge to A2MC's AI", "inject the X discovery into the KB", "make the agent aware of X", "author a curated relationship for X", or wants a manuscript/literature insight to reach the reasoning pipeline. This is the HUMAN-originated counterpart to curate-knowledge (which promotes run-originated proposals); both are Tier-3 curated writes and you (the human-in-the-loop) own the truth call. Distilled from graphrag_curated_yaml_roadmap.md + dev log 20260519d (the clumping_index injection).

2026-07-076

name	phase2-screening
description	Run Phase 2 (SCREENING) of the A2MC calibration workflow as the offline agent — the human-in-the-loop analog of the orchestrator's `_run_screening()` / `reasoning.analyze_screening_results()`. Rank the ensemble against the validation targets (RMSRE, targets-satisfied), find the best / lowest-cost / most-targets cases, read the error/bias patterns, and route the round into Phase 3 diagnosis. Use when the user says "screen the ensemble", "rank against targets", "which case is best", "run Phase 2", "how many targets are met", after Phase 1 sensitivity.
modes	{"requires_fates":false,"nutrient_pathway":"any","scope":["calibration"],"summary":"Offline analog of Phase 2 (rank ensemble vs validation targets). Applies in every calibration mode."}

Phase 2: Screening (offline agent)

Floor, not ceiling. Ranking + best-case is the online floor. As the offline agent, go further: inspect the whole cost distribution (not just top-N), test whether a "best" case is real or a contamination artifact before trusting it, look for equifinality (different parameter sets, same cost), or compare this round's frontier against a prior round. Produce the online ranking, then interrogate it.

Step 1 — rank the ensemble (reads existing extracted outputs — no new simulations)

Footgun — --max-case-num contamination guard. When experiment cases (e.g. _exp / #5001+) share the extract dir with the Morris ensemble, pass --max-case-num 4890 so screening ranks only the ensemble. Without it, foreign cases inflate the top-N. (Same guard the compare-calibration-rounds cap enforces.)

Footgun — index vs case number. Screening Set_ID/indices can be position+1, not the real case number. Use the JSON best_case_num + the _results.txt Sim_ columns, not the indices file, before quoting a "best case" (see diagnose-forensics).

Step 2 — read the patterns (you are the reasoning)

Step 3 — log and route

Related skills / next phase

Canned single-round report (graphs + eval + μ)* → summarize-calibration-round. Cross-round frontier → compare-calibration-rounds. This skill decides + routes; those render.
A "best" case looks too good → diagnose-forensics (triage before trusting).
Next: phase3-diagnosis.

Changelog

2026-07-02: Created — offline Phase 2 routine mirroring reasoning.analyze_screening_results(); drives screen_ensemble, delegates figures to summarize-/compare-calibration-round, routes to phase3-diagnosis.