Screenshot source record

The four Phase 43C screenshots were captured on 2026-08-01 from AgentGuard source commit 150709b3b790aba5c33d1918f00184d46868dd2e, package version 0.2.2. They depict real project surfaces and deterministic local output; no external coding agent, model API, private repository, or paid service was used.

Source generation

Equivalent commands from a clean repository checkout are:

REPO_ROOT="$(git rev-parse --show-toplevel)"
CAPTURE_ROOT="$(mktemp -d)"
cd "$CAPTURE_ROOT"

"$REPO_ROOT/.venv/bin/python" \
  "$REPO_ROOT/scripts/showcase_demo.py" \
  --suite "$REPO_ROOT/examples/showcase/showcase.yaml" \
  --output-dir artifacts/showcase

"$REPO_ROOT/.venv/bin/python" -m agentguard.cli.main run \
  "$REPO_ROOT/examples/showcase/configs/unsafe_command.yaml" \
  --agent local-command --guard-mode audit --allow-fail-result

"$REPO_ROOT/.venv/bin/python" -m agentguard.cli.main reports site \
  --output site \
  --history-db .agentguard/history.db \
  --reports-root .agentguard \
  --title "AgentGuard Evaluation Showcase" \
  --force

python -m http.server --bind 127.0.0.1 8765 --directory site

The showcase generated six scenario runs and one suite record: one safe pass and five expected unsafe failures. The separate audit-mode run generated the single sanitized guard incident used by the dashboard, incident, and trends captures. Runtime IDs and timestamps vary between executions.

Capture inventory

File	Source	Viewport and crop	Theme
agentguard-docs-home.png	https://richinmrudul.github.io/agentguard/	1440×900 viewport	light
agentguard-dashboard.png	site/index.html	1440×900 viewport	light
agentguard-incident-detail.png	generated site/details/incident-*.html	1440 px wide full-page capture; 1055 px high	light
agentguard-evaluation-evidence.png	site/trends.html	1440×900 viewport	light

The public homepage was captured from GitHub Pages. The other three images were captured from the localhost-only server above. Browser and font rendering can vary, so the screenshots are not claimed to be byte-reproducible.

Sanitization and optimization

Before capture, the three selected static pages were checked for private or temporary absolute paths, fake secret literals, authorization values, raw unsafe-command payloads, raw diffs, and environment assignments. Browser DOM inspection found none. The public capture contains only intentionally public repository and release information.

Each image was inspected at full resolution for clipped content, browser or account chrome, notifications, cursor obstruction, loading state, and sensitive text. No OCR utility was available, so sanitization combines source-page text inspection, browser DOM checks, PNG metadata checks, and human visual review; it does not claim OCR completeness.

The capture backend supplied high-quality browser image bytes. They were converted to PNG with the platform image utility, then bounded PNG-chunk inspection removed EXIF and textual metadata chunks. Pixel dimensions were not resampled. Each file is below 1 MB, and the four-file set is below 3 MB.

Generated .agentguard, report-site, and temporary capture trees are not part of the committed asset set.