Brandify Agent

Brandify Agent — Agent Log to HTML Report

Converts Claude Code stream-json agent logs into styled, interactive HTML reports using the Fenrir Ledger visual system. Supports local HTML viewing and publishing as chronicles.

Usage

/brandify-agent <session-id-or-log-path>
/brandify-agent <session-id-or-log-path> --publish
/brandify-agent --regen-assets

Arguments

Workflow

Step 1 — Download log from GKE cluster (ALWAYS try this first)

When given a session ID, always attempt to download the latest log from the GKE cluster before falling back to any cached local copy. This ensures the report reflects the most recent agent output, even if the agent is still running.

REPO_ROOT=$(git rev-parse --show-toplevel)
SESSION_ID="<session-id>"
LOG_DIR="$REPO_ROOT/tmp/agent-logs"
LOG_FILE="$LOG_DIR/$SESSION_ID.log"

# Always try to download fresh from GKE first
mkdir -p "$LOG_DIR"
kubectl logs "job/agent-$SESSION_ID" -n fenrir-agents --timestamps=false > "$LOG_FILE.tmp" 2>/dev/null
if [ -s "$LOG_FILE.tmp" ]; then
  mv "$LOG_FILE.tmp" "$LOG_FILE"
  echo "[ok] Downloaded fresh log from GKE: $LOG_FILE"
elif [ -f "$LOG_FILE" ]; then
  rm -f "$LOG_FILE.tmp"
  echo "[info] GKE download failed, using cached log: $LOG_FILE"
else
  rm -f "$LOG_FILE.tmp"
  echo "[error] No log available — GKE download failed and no cached copy"
  exit 1
fi

If the argument is a full file path (contains / or ends with .log), use it directly (skip the download step — the user provided a specific file).

Step 2 — Generate the report

HTML mode (default):

SCRIPT="$REPO_ROOT/.claude/skills/brandify-agent/scripts/generate-agent-report.mjs"
node "$SCRIPT" --input "$LOG_FILE"

Output goes to the same directory with .html extension (replacing .log). Shared assets (agent-report.css, agent-report.js) are auto-generated alongside.

Publish mode (--publish):

SCRIPT="$REPO_ROOT/.claude/skills/brandify-agent/scripts/generate-agent-report.mjs"
BLOG_DIR="$REPO_ROOT/development/ledger/content/blog"
node "$SCRIPT" --input "$LOG_FILE" --publish --blog-dir "$BLOG_DIR"

Output: content/blog/agent-{session-slug}.mdx with frontmatter (title, date, rune, excerpt, slug, category: "agent"). Viewable at /chronicles/agent-{slug}. Uses <details>/<summary> for collapsible turns (no JS needed). Agent chronicles display with an "Agent" badge on the index and detail pages.

Step 3 — Report

**Report generated:** `<html-path-or-mdx-path>`
**Turns:** N | **Tool calls:** N | **Errors:** N
**Verdict:** PASS/FAIL/none
Open: `open <html-path>` or view at `/chronicles/agent-{slug}`

Metadata Detection

The report extracts session ID, branch, and model from multiple sources (in priority order):

1. Entrypoint text lines — GKE startup output before JSON events (Session:, Branch:, Model:)
2. JSON system/init event — Always present in stream-json logs (contains model, session_id)
3. Filename — Dispatch session ID pattern: issue-<N>-step<S>-<agent>-<hash>.log
4. Git commands in log — Branch name from git branch --show-current tool results

Regenerate Assets

To regenerate shared CSS/JS without processing a log:

node "$SCRIPT" --regen-assets --output-dir tmp/agent-logs

Output Structure

tmp/agent-logs/
  agent-report.css        ← shared theme (regenerable)
  agent-report.js         ← shared toggle logic (regenerable)
  issue-682-step2-loki-xxxx.log   ← raw log (input)
  issue-682-step2-loki-xxxx.html  ← styled report (output)

Features

• Fenrir Ledger dark theme (void black, gold accents, Cinzel fonts)
• Collapsible turns with thinking, text, and tool blocks
• Color-coded tool badges (Bash=teal, Read=blue, Edit=amber, Write=gold, Todo=purple)
• Entrypoint section with color-coded status lines
• Summary bar with turn/tool/error counts
• Auto-detected QA verdict with PASS/FAIL styling
• Expand All / Collapse All controls