cv-test-report

name	cv-test-report
description	Publish Curvine full-chain daily test reports to the Hextra Hugo site at CurvineIO/test-reports, using a standard Markdown template with no environment or secret leakage. Use when the user asks to publish a test report, convert harness output to Markdown, or update CurvineIO/test-reports.

cv-test-report

Turn a full-chain / daily harness report into one Markdown post and publish it to CurvineIO/test-reports. Live site: https://curvineio.github.io/test-reports/ (Hugo Hextra). Product GitHub: CurvineIO/curvine.

One report = one Markdown file. Hextra builds the sidebar and in-page TOC. Posts older than 30 days are pruned on deploy.

Public posts express test results only, in English. They must not describe or identify the test environment. Do not use Chinese or other CJK text.

When to Use

• User asks to publish / upload / post a test report
• User pastes a full-chain daily report and wants it on GH Pages
• User asks to convert harness output into the standard report format
• User mentions test-reports, cv-test-report, or the report site URL

Target repo

Item	Value
Product GitHub	CurvineIO/curvine
Pages repo	CurvineIO/test-reports
Pages URL	https://curvineio.github.io/test-reports/
Theme	Hextra (github.com/imfing/hextra, Hugo module)
Post path	content/reports/YYYY-MM-DD-<slug>.md
Retention	30 calendar days (filename / front-matter date)

gh repo clone CurvineIO/test-reports /path/to/test-reports

Do not put the site inside the curvine tree. Site chrome GitHub links must point to https://github.com/CurvineIO/curvine.

Forbidden in the published post (hard stop)

If any of the following appear, strip them or do not publish. Do not replace with aliases that still identify the environment (reserved, internal-archive, host roles, etc.).

Class	Examples (never publish)
Paths	/data/..., /var/lib/..., file://, UNC, home dirs, ~/...
Accounts	Unix users, Git authors as operators, kube contexts, SSO names
Network	IPv4/IPv6, hostnames, DNS, CIDR, VIP, 10.x / 192.168.x / 172.16.x
Environment	cluster / CSI context names, machine types, client/server roles+addresses, image digests, harness SHAs, run directories, archive roots
Secrets	tokens, kubeconfigs, passwords, keys
Evidence handles	Run IDs, fingerprints, artifact URLs, log path columns
Non-English copy	Chinese headings, table headers, alerts, or body text

Allowed: profile names, PASS/FAIL, counts, durations of a profile, GO/NO-GO, failure case names, one-line error text with paths/IPs removed, LTP suite stats, perf numbers, public GitHub Issue/PR numbers, Curvine product commit only if the user explicitly asks to name the tested revision (otherwise omit).

Title may include the report calendar date. Do not include start/end timestamps, total wall time, or where it ran.

Step 1: Collect the source (private)

Read the raw harness report locally. Use env fields only to decide the post date and GO/NO-GO. Those fields stay out of the Markdown. Translate source text into English before publishing.

Step 2: Write the standard post

Copy assets/daily-full-chain.md. Hierarchy rules: references/report-structure.md.

---
title: "Curvine full-chain daily test report - YYYY-MM-DD"
linkTitle: "YYYY-MM-DD full-chain"
date: YYYY-MM-DDT00:00:00Z
weight: -YYYYMMDD
tags: [full-chain, daily, no-go]
---

• date must be in the past (Hugo skips future posts).
• weight is -YYYYMMDD so the sidebar lists newest first.
• Filename: content/reports/YYYY-MM-DD-full-chain.md.
• Keep the H2/H3 order. Omit optional H3s that have no data.
• Never drop H2 Quality conclusion or Test results.
• Do not put an H1 in the body. Hextra already renders title as the page H1.
• Lead GO / NO-GO with a GitHub alert: > [!TIP] for GO, > [!CAUTION] for NO-GO.
• Table cells that are exactly PASS / FAIL / pass / fail / degraded / NOT_RECORDED get colored status dots automatically. Do not wrap them in backticks.

Markdown must render (hard stop)

Do not publish until a local Hugo build proves the HTML is valid. Goldmark + GFM tables are strict.

Rule	Why
No # heading in the body	Duplicates the Hextra H1
No backticks in the published post body	Inline code next to CJK or | often stays literal or eats spaces
No CJK characters	Site and skills are English-only
Tokens use bold: NO-GO, failed, 40	Renders in lists and tables
Tables: spaces around every |	| ITEM | not |ITEM|
Header cells = separator cells = every row	One missing | --- | dumps the whole table as a paragraph
No / or () in table headers	Use SPEED GiB/s, P50 ms, Sequential write 64KB
Blank line before and after each table	Required by GFM
Table before/after is a heading or paragraph, not a list item	Lists swallow tables

Validate before commit (from test-reports):

python3 - <<'PY'
from pathlib import Path
import re, sys
p = Path("content/reports/YYYY-MM-DD-full-chain.md")
body = p.read_text().split("---", 2)[-1]
if re.search(r"^# ", body, re.M):
    sys.exit("H1 in body")
if "`" in body:
    sys.exit("backticks in body")
if re.search(r"[\u4e00-\u9fff]", body):
    sys.exit("CJK in body")
lines = body.splitlines()
i = 0
while i < len(lines):
    if re.match(r"^\| .+\|$", lines[i]) and i + 1 < len(lines) and re.match(r"^\|[-: |]+\|$", lines[i + 1]):
        cells = lambda s: s.strip().strip("|").split("|")
        h, sep = cells(lines[i]), cells(lines[i + 1])
        if len(h) != len(sep):
            sys.exit(f"header/sep {len(h)} vs {len(sep)}: {lines[i]}")
        j = i + 2
        while j < len(lines) and lines[j].startswith("|"):
            if len(cells(lines[j])) != len(h):
                sys.exit(f"row {len(cells(lines[j]))} vs {len(h)}: {lines[j]}")
            j += 1
        i = j
        continue
    i += 1
print("markdown ok")
PY
hugo --gc --minify
# built HTML must have exactly one H1, real <table>s, hextra-toc, and no leftover "| ITEM |"

If Hugo is not on PATH, use the repo .tools/hugo. First build needs Go (hugo mod get). Fail the publish if the script exits non-zero or the HTML contains <p>\|.

Step 3: Redact, then scan

After filling the template, search the file for leaks:

rg -n -i \
  -e '/data|/var/lib|file://|home/' \
  -e '[0-9]{1,3}(\.[0-9]{1,3}){3}' \
  -e 'sha256:|ecs\.|kube|CSI context' \
  content/reports/YYYY-MM-DD-full-chain.md

Fix every hit that is not an explicit product commit the user asked to publish. Re-run until clean.

Do not add static/files/ attachments (they usually carry paths or env dumps).

Step 4: Publish

From test-reports (confirm with the user before commit if this session is in curvine):

git checkout main
git pull --ff-only
git add content/reports/YYYY-MM-DD-full-chain.md
git commit -m "$(cat <<'EOF'
docs: add full-chain report YYYY-MM-DD

EOF
)"
git push origin main

Step 5: Verify

gh run watch --repo CurvineIO/test-reports --exit-status
curl -sS -o /dev/null -w '%{http_code}\n' \
  https://curvineio.github.io/test-reports/reports/YYYY-MM-DD-full-chain/

Expect 200. Posts older than 30 days are removed on the next deploy.

Do Not

• Publish environment, paths, accounts, IPs, usernames, run IDs, or evidence URLs
• Write a Test scope and environment or Evidence index section
• Put # title, backticks, or CJK in the published post body
• Publish a table whose column counts do not match
• Attach raw JSON/logs
• Commit reports into curvine or curvineio.github.io
• Point site GitHub chrome at CurvineIO/test-reports (use CurvineIO/curvine)
• Write the post for the old Paper theme (no body H1 is still required; Hextra adds linkTitle, weight, and GO/NO-GO alerts)
• Treat a subset of green profiles as GO when any required profile failed

Related

• cv-create-issue — product regression after attribution
• cv-csi-test — CSI profile meaning
• cv-add-skills — skill layout

Checklist

• Front matter has linkTitle and weight: -YYYYMMDD
• GO/NO-GO uses > [!TIP] or > [!CAUTION]
• H2/H3 order matches the template; no body H1
• English only; no CJK; no backticks in the published post body; tokens are bold
• Every table column count matches; Hugo HTML has <table> and hextra-toc, not \| ITEM \|
• Results only; no env / path / IP / account / username
• rg leak scan clean
• No attachments
• Pushed to CurvineIO/test-reports main
• Pages URL returns 200