html-experiment-report

name	html-experiment-report
description	Use when producing a browser-readable HTML experiment or Eval report; first decide the primary figure, then plan and run an evidence-backed report renderer while keeping domain authority in the original tool.

HTML Experiment Report

Tool Commands

Use the command packet before applying this skill's workflow:

python3 tools/agent_tools/skill_tool_commands.py show --skill html-experiment-report --format text

Execute the required and task-matching conditional commands that the packet prints.

1. Read agents/skills/html-experiment-report.md.
2. Use $structure-planning before renderer or experiment implementation to fix the first figure, source-to-structure map, metric contract, section order, and invalid interpretations.
3. Use $html-output for polished HTML layout, optional $imagegen assets, existing preview-server reuse, python3 -m http.server --bind 0.0.0.0 publication, curl -fsS validation, and local/external browser URLs.
4. Survey existing assets first: relevant skills, tool catalog entries, workflow docs, report helpers, previous run artifacts, and existing experiment scripts.
5. Write a responsibility analysis before implementation: raw evidence owner, renderer owner, domain decision owner, generated artifact path, and the invalid authority drift to avoid.
6. Name the first figure through the structure contract before running the experiment. Record its question, required data, metric denominator, directionality, and invalid interpretations.
7. Derive the experiment plan from that figure: evidence-producing command, renderer command, HTML output path, blocked-provider behavior, and validation gates.
8. Reuse existing producers and helpers. Add a report-specific renderer or adapter only when the HTML artifact needs one.
9. Keep generated SQLite, JSON, and HTML artifacts under ignored run paths such as reports/agents/<run-id>/; do not make generated HTML a policy truth.
10. Put the primary figure first in the HTML, followed by source packet, observations, interpretation, limitations, provenance, and next action.
11. For semantic-index provider comparison, use the figure Provider Delta To Shared Candidate Logic and state that LLM latent vectors may change retrieval/ranking deltas but do not create labels, ownership, or merge authority.
12. Validate with targeted renderer tests, docs checks for changed Markdown, and catalog/dependency checks for changed tool or skill wiring.
13. Record closeout tokens: html_experiment_report=complete, html_primary_figure=<figure-name>, structure_contract=<path-or-inline>, html_source_artifact=<path>, html_report_artifact=<path>, html_domain_authority=<tool-or-doc>, and html_output=complete, html_server_mode=<reuse|started|not_requested|blocked>, and html_invalid_interpretations_recorded=yes.