Blog Update Skill

Convert technical agent sessions into Fuwari blog posts. V10 uses a minimal, safety-first workflow: one main editor drafts a complete article, deterministic validators check it, and existing posts are changed only after a user-approved diff.

Usage

# Create or preview a draft article
/blog-update <topic> --tags [tag1,tag2] --category <category>

# Infer the topic from the visible session, then ask for confirmation
/blog-update

# Merge into an existing article after showing a diff
/blog-update <topic>

File Structure

blog-update/
├── skill.md                      # Canonical V10 skill spec
├── SKILL.md                      # Loader-compatible copy in this worktree
├── prompts/
│   ├── implement-agent.md        # Optional extraction assist for long/noisy sources
│   ├── merge-agent.md            # Optional complex merge assist
│   └── review-agent.md           # Optional semantic/evidence review
├── scripts/
│   ├── scan-sensitive-content.py # Blocking safety scan, stdlib
│   ├── validate-post.py          # Markdown structure/session leak/link checks, stdlib
│   ├── validate-codeblocks.py    # Optional parser checks for code fences, pyyaml
│   ├── validate-frontmatter.py   # Report-only by default; --fix writes safe fixes
│   └── check-prompts.py          # V10 fixture/contract tests
├── config.json                   # Local deployment config
├── LICENSE
└── README.md

Workflow

source material / visible session
    -> internal source brief
    -> single complete draft with frontmatter + H1 + body
    -> sensitive scan + structural validators
    -> new draft path or existing-article unified diff
    -> explicit user confirmation before applying existing-article changes

Default behavior changed in V10:

• New articles start as draft: true.
• Sensitive Content Scan is a blocking gate before any article is offered as publishable.
• uv is optional. Use it when available for PEP 723 dependencies; otherwise run stdlib checks and report skipped dependency-backed checks.
• validate-frontmatter.py is report-only by default. It writes only with --fix.
• --fix must not change draft: true to false.
• Subagent prompts are optional assistive tools, not the default path.

Validators

Script	Gate	Coverage	Deps
scripts/scan-sensitive-content.py	Blocking	Likely secrets, auth headers, cookies, private keys, emails, private IP/URL/host references, absolute user paths	stdlib
scripts/validate-post.py	Blocking + advisory	H1 presence, duplicate H1, fence language/closure, session/tool leaks, internal .md links, style advisories	stdlib
scripts/validate-frontmatter.py	Blocking + advisory	Required Fuwari fields, valid date, boolean draft, title/H1 consistency, tags/category shape	pyyaml optional; simple fallback parser exists
scripts/validate-codeblocks.py	Blocking when runnable	Parser-backed checks for Python, JSON, YAML, Bash-like fences	pyyaml + local bash when available

Recommended local verification:

python3 -m py_compile scripts/*.py
python3 scripts/check-prompts.py
git diff --check

Configuration

Deployed config location:

~/.claude/skills/blog-update/config.json

Example:

{
  "blogBasePath": "your/blog/posts/path",
  "fileExtension": ".md"
}

Do not overwrite a user's deployed config.json during installation; it contains their local blog path.

Deployment Note

Historically the repo used lowercase skill.md while some loaders expected uppercase SKILL.md. Keep both contents synchronized when editing this repo. In this worktree, SKILL.md may be a regular file instead of the historical symlink; do not change that type unless you explicitly intend to update the compatibility layer.

Version History

Version	Major Changes
V1	Basic blog update
V2	Subagent architecture separation
V3	Smart multi-granularity merge
V4	CSO optimization, prompts split, <200 lines
V5	Merge algorithm enhancement, Red Flags safety check
V6	Full English localization
V7	Main-session-only architecture; 6-rule Extraction Contract; review-agent v3
V8	Scripted validation for body/code/frontmatter via uv run
V9	Evidence-first workflow with optional subagents and Evidence Ledger
V10	Minimal safety-first workflow: complete draft first, default draft: true, sensitive-content gate, report-only frontmatter validator, optional uv, diff/confirmation for existing posts