report-generation

name	report-generation
description	Convert ts-agents workflow outputs, run manifests, plots, JSON, CSV files, and short Markdown reports into stakeholder-facing Markdown, Quarto, HTML, or PDF reports. Use when the user asks for a report, executive summary, research appendix, or reusable analysis deliverable from existing ts-agents artifacts.
compatibility	Designed for ts-agents CLI artifacts. Also usable by coding agents that can read files and run shell commands.
metadata	{"domain":"time-series","tasks":["reporting","artifact-synthesis","quarto","stakeholder-communication"],"ts_agents":{"preferred_tools":[],"artifact_checklist":["run_manifest.json","report.md"]}}

Report generation from ts-agents artifacts

When to use

Use this skill when a workflow has already produced artifacts, or when the user wants a polished report as the final deliverable rather than only terminal output. Prefer this skill after inspect-series, forecast-series, or activity-recognition runs.

Goal

Create a concise, reproducible report that links claims to files:

• the command or workflow that produced the result
• key metrics and quality flags from JSON outputs
• selected plots and tables
• limitations and recommended next steps

Minimal artifact checklist

• A workflow output directory containing run_manifest.json, when available
• The workflow report.md, when available
• Referenced plot/table artifacts, for example PNG, CSV, or JSON files
• A final report.md or report.qmd; optionally rendered HTML or PDF

Workflow

1. Inspect the workflow and output directory

Start from a workflow output directory such as outputs/forecast/<run-id>/, outputs/activity/<run-id>/, or an explicit output directory supplied by the user. Read run_manifest.json first when it exists, then inspect the files it references.

If the user has not run a workflow yet, discover the relevant workflow contract:

uv run ts-agents workflow show inspect-series --json
uv run ts-agents workflow show forecast-series --json
uv run ts-agents workflow show activity-recognition --json

2. Choose the report format

Use Markdown for quick internal handoff. Use Quarto HTML for shareable reports. Use Quarto PDF only when the environment has a working Quarto/PDF toolchain.

Recommended default:

---
title: "Time-Series Analysis Report"
format:
  html:
    toc: true
    code-fold: true
---

3. Compose the report

Use this structure unless the user asks for a different one:

• Summary: state the main result in 2-4 bullets.
• Inputs and run metadata: name the source data, workflow, command options, output directory, and any quality flags or warnings.
• Findings: include the most important metrics, plots, and tables. Link every figure to a real artifact path.
• Limitations: list data, model, dependency, validation, or workflow limitations that affect interpretation.
• Next steps: recommend the next analysis or productionization step.

4. Render only after checking paths

Before rendering, verify that every referenced image/table exists and is readable. If Quarto is installed, render the report:

quarto render outputs/reports/report.qmd

If Quarto is not available, deliver Markdown plus the artifact directory path.

Workflow-specific guidance

Forecasting

Use forecast_comparison.json for metrics and model ranking, forecast_comparison.png for the visual comparison, forecast.csv for the selected forecast, and report.md for the baseline narrative.

Activity recognition

Use window_selection.json for candidate-window scores, eval.json for final metrics and confusion matrix values, window_scores.png and confusion_matrix.png for figures, and report.md for the summary.

Inspect series

Use summary.json and ledger.json for diagnostics, autocorrelation.png for lag structure, and report.md for the quick interpretation.

Guardrails

• Do not invent metrics or conclusions that are not present in artifacts.
• Do not paste large arrays into the report; summarize and link to JSON/CSV.
• Preserve run provenance: command options, output directory, run ID, warnings, quality flags, and timestamps when available.
• If a workflow status is degraded, explain why before making recommendations.
• For stakeholder reports, prefer one strong figure per section over a gallery of every generated artifact.