agentsociety-analysis

Use when an experiment run has completed and the user wants rigorous interpretation, claim-driven charts, bilingual reports, or cross-hypothesis synthesis from simulation data. Also use when multiple charts or PNG/JPG assets must be assembled into one labeled composite figure. Requires high-quality narrative and evidence traceability, not only harness gate PASS.

Zur Installation springen

Repository: tsinghua-fib-lab/AgentSociety
Letzte Quellaktivität: 20. Juli 2026 um 02:44
Erkannte Sprache von SKILL.md: Englisch
Sterne: 1.203
Forks: 205

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.

Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.

npx skills add https://github.com/tsinghua-fib-lab/AgentSociety --skill agentsociety-analysis

Der Befehl bleibt in einer Zeile. Scrollen Sie horizontal, um ihn vor dem Kopieren vollständig zu prüfen.

Sie bevorzugen eine lokale Kopie? Laden Sie die Dateien herunter, die SkillsMP derzeit vorliegen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer

54 Dateien

SKILL.md

Quellanweisungen · Schreibgeschützte Vorschau

Zurück zur Suche Zurück zur Installation

frontend-design

Create distinctive, production-grade frontend interfaces with high design quality. Use when building or polishing analysis HTML reports under agentsociety-analysis — read via support/frontend-design/ inside that skill, not as a separate pipeline skill.

2026-07-20

agentsociety-paper-review

Use when a completed research manuscript needs a robust internal, venue-calibrated mock peer review. Orchestrates three isolated reviewer subagents in parallel, then a separate MetaReview subagent that verifies evidence, reports agreement and score dispersion, and emits advisory reroutes. Never revises research artifacts, executes experiments, or updates pipeline state.

2026-07-20

agentsociety-research-pipeline

Use when starting or resuming an AgentSociety research workspace, deciding which research skill to invoke next, checking current pipeline state, or sizing a simulation before configuration and module creation.

2026-07-20

Alle aus diesem Repository ansehen

name	agentsociety-analysis
version	1.0.0
description	Use when an experiment run has completed and the user wants rigorous interpretation, claim-driven charts, bilingual reports, or cross-hypothesis synthesis from simulation data. Also use when multiple charts or PNG/JPG assets must be assembled into one labeled composite figure. Requires high-quality narrative and evidence traceability, not only harness gate PASS.

Analysis

Use this skill for interactive experiment analysis. You (Claude Code) own orchestration, scientific judgment, and user dialogue. The harness CLI enforces structural reproducibility and phase gates; it does not replace your reasoning.

Quality before gates: Read references/analysis-quality.md at the start of each phase. A validate-* PASS with weak science is a failure — improve claims, evidence, and prose first, then attest.

Harness vs LLM (read first)

Layer	Who	What
Structural	Python `validate-*`	Files exist, JSON schemas valid, DB/tables/artifacts consistent
Attestation	You via `record-attestation`	Phase-complete judgment: findings, limitations, user alignment (schema-checked only)
Narrative	You	Reports, synthesis, claim wording, chart design, caveats
Memory

Command

Purpose

$PYTHON_PATH .agentsociety/bin/ags.py analysis load-context --workspace . --hypothesis-id ID --experiment-id ID

Load experiment context

$PYTHON_PATH .agentsociety/bin/ags.py analysis list-tables --data-path PATH

List replay-backed tables/datasets

$PYTHON_PATH .agentsociety/bin/ags.py analysis data-summary --data-path PATH

Full data summary

$PYTHON_PATH .agentsociety/bin/ags.py analysis query-data --data-path PATH --sql "SELECT ..."

Read-only SQL query

$PYTHON_PATH .agentsociety/bin/ags.py analysis run-eda --data-path PATH --output-dir DIR --type TYPE [--workspace . --hypothesis-id ID]

Generate EDA; auto-registers explore artifacts when hypothesis id is set

$PYTHON_PATH .agentsociety/bin/ags.py analysis compose-figure --workspace . --hypothesis-id ID --spec FILE

Combine claim-scoped raster charts/images into one labeled figure

$PYTHON_PATH .agentsociety/bin/ags.py analysis collect-assets --workspace . --hypothesis-id ID --experiment-id ID --output-dir DIR

Collect assets; DIR may be the report root or its assets/ child

$PYTHON_PATH .agentsociety/bin/ags.py analysis sync-report-assets --workspace . --hypothesis-id ID --experiment-id ID

Copy images referenced in reports from charts/ → assets/

$PYTHON_PATH .agentsociety/bin/ags.py analysis intake --workspace . --hypothesis-id ID --experiment-id ID

Initialize harness state

$PYTHON_PATH .agentsociety/bin/ags.py analysis write-plan --workspace . --hypothesis-id ID --payload '{...}'

Persist analysis_plan.yaml

$PYTHON_PATH .agentsociety/bin/ags.py analysis validate-plan / validate-explore / …

Structural + attestation gate per phase

$PYTHON_PATH .agentsociety/bin/ags.py analysis record-attestation --payload '{...}'

LLM phase sign-off (required for gate pass)

$PYTHON_PATH .agentsociety/bin/ags.py analysis record-phase-artifacts --phase explore --artifacts '["path"]'

$PYTHON_PATH .agentsociety/bin/ags.py analysis build-report-context --workspace . --hypothesis-id ID

Merge EDA/charts/claims → data/report_context.md (orchestrator / subagent input)

$PYTHON_PATH .agentsociety/bin/ags.py analysis prepare-produce --workspace . --hypothesis-id ID --experiment-id ID

Explicitly build context, sync report assets, and embed interactive EDA

$PYTHON_PATH .agentsociety/bin/ags.py analysis validate-report-quality / record-report-review / record-synthesis-review

Mechanical quality + independent review records

$PYTHON_PATH .agentsociety/bin/ags.py analysis validate

Workspace complete = validate-synthesis pass

$PYTHON_PATH .agentsociety/bin/ags.py analysis advance --phase PHASE

After prior phase gate_pass

$PYTHON_PATH .agentsociety/bin/ags.py analysis status / gate-status / run-loop

Checkpoints, recent operation receipts, retryable runs, and next steps

$PYTHON_PATH .agentsociety/bin/ags.py analysis draft-reflection / record-reflection / promote-reflection

Reviewable experience memory and explicit promotion

$PYTHON_PATH .agentsociety/bin/ags.py analysis record-feedback / review-reflection

User feedback capture + pre-promotion review

$PYTHON_PATH .agentsociety/bin/ags.py analysis memory-context

Show active lessons, recipes, and confirmed preferences used by orchestration

Mistake	Fix
Skipping context loading and going straight to charts	Always start with Stage 1 (`intake` + `write-plan`)
Skipping synthesis	Stage 6 is required; run `validate-synthesis` before pipeline `analysis completed`
Starting from a favorite plot type before defining the message	Write a figure contract first: finding, scope, evidence source, and expected takeaway
Decorative charts without a claim or figure contract	Each chart must defend one finding; skip charts when prose/EDA is enough
Writing analysis output to wrong directory	Single-experiment goes to `presentation/hypothesis_{id}/`, synthesis goes to `synthesis/`
Running EDA on all tables indiscriminately	Only run EDA for tables explicitly selected during Stage 2
Inventing helper scripts instead of using the analysis CLI	Use `.agentsociety/bin/ags.py analysis ...` for all mechanical operations
Chart without description	Every chart in the report must have a one-line description directly below it
Overloaded colors, repeated legends, or weak panel hierarchy	Reuse the chart contract and `references/charts.md` before regenerating
Leaving a multi-view finding as several disconnected screenshots	Generate atomic charts first, then assemble a single `figure_{nn}_{slug}.png` with labeled panels
Treating simulation output as direct real-world proof	Downgrade the claim strength and record the caveat in the figure contract or report text
Using ad hoc matplotlib defaults in one script and styled output in another	Reuse `references/api.md` and `references/common-patterns.md` so colors, fonts, and export rules stay consistent

Phase	Subagent prompt	Orchestrator keeps
explore, refine	`subagent-prompts/data-explorer.md`	`record-phase-artifacts`, `record-contract`, gates
produce	`report-producer` + `report-reviewer` (independent)	`record-report-review`, `validate-release`, attestation
synthesis	`synthesis-producer` + `synthesis-reviewer`	`record-synthesis-review`, `validate-synthesis`, attestation

agentsociety-analysis

Installationsoptionen

Quelldateien prüfen

Analysis

Harness vs LLM (read first)

Analysis

Harness vs LLM (read first)

Overview

When to Use

Quick Reference

First Move: Analysis Contract Before Plotting

Common Mistakes

Workflow

Pipeline Position

Stage Notes

Shared References

CLI Tool

Subagent Delegation

Runtime Contract

Hard Constraints

Progress Tracking

agentsociety-analysis

Quellinformationen

Installationsoptionen

Quelldateien prüfen

Mehr aus diesem Repository

Analysis

Harness vs LLM (read first)

Analysis

Harness vs LLM (read first)

Overview

When to Use

Quick Reference

First Move: Analysis Contract Before Plotting

Common Mistakes

Workflow

Pipeline Position

Stage Notes

Shared References

CLI Tool

Subagent Delegation

Runtime Contract

Hard Constraints

Progress Tracking

Mehr aus diesem Repository