| name | blog |
| argument-hint | <scenario> [--source <path|url>] [--target <repo|path>] [--slug <slug>] [--promo linkedin,x,channels] [--dry-run] |
| disable-model-invocation | true |
| allowed-tools | Read, Write, Edit, Glob, Grep, Bash, Agent |
| description | Turn a source scenario into a publishable blog post. First supported playbook:
Loom/demo-to-blog, where a raw notes file plus Loom screenshots become a
polished site post after Advisor briefing, photo/text alignment audit, site
convention audit, narrative/fact-check audit, source-faithful image handling
(exact URLs or local files downloaded from them), optional social/channel-promotion
artifact generation, and build verification. Manual-invoke only because it writes content and may spawn
delegates.
TRIGGER when: /blog <scenario> invoked; asked to create a blog post from a
Loom/video/demo/transcript/screenshots; asked to turn .claude/specs/<name>/
demo.md into an openharness-web blog post.
|
Blog
Create a publishable blog post from $ARGUMENTS. This is a task skill: it may write files, localize assets, spawn delegates, and run target-site validation. Do not use it for casual copy edits.
Arguments received: $ARGUMENTS
Dispatch
- If
$ARGUMENTS is empty, ask for a one-line scenario and stop.
- Parse any explicit flags:
--source <path|url> — raw notes, transcript, Loom URL, or source folder.--target <repo|path> — blog repo or site root. Treat /worktrees ... as repo-relative .oh/worktrees/..., never filesystem root.--slug <slug> — target post slug, lowercase kebab-case.--promo linkedin,x,channels — also generate a reviewable promotion artifact for social posts and channel blasts; never publish.--dry-run — perform audits and propose filename/outline only; write nothing.
- Route to the first playbook,
references/loom-to-blog.md, when the scenario mentions Loom, video, demo, transcript, screenshots, demo.md, or a source under .claude/specs/.
- If no playbook fits, say the only implemented reference is
loom-to-blog and ask whether to adapt it.
Required first read
For Loom/demo work, read and follow:
${CLAUDE_SKILL_DIR}/references/loom-to-blog.md
That reference is authoritative for the workflow details. This SKILL.md only defines dispatch, boundaries, and shared completion rules.
Shared rules
- Use Advisor before drafting. For non-trivial source material, read
.oh/agents/advisor.md and obtain a tight implementation briefing/delegate plan before spawning workers or writing the post. If the custom advisor agent is unavailable, use a planning-capable agent and require the same 5-field Advisor Briefing contract.
- Audit all media before publishing. Every source image/video screenshot must be accounted for as keep/drop/reference-only/missing. For Loom/demo.md sources, use source-faithful image handling: either embed the exact source image URLs or save local copies downloaded from those exact URLs when PR/site rendering would otherwise break. Do not publish unredacted screenshots that expose tokens, OAuth callback URLs, private accounts, hostnames, sensitive usage panels, or unrelated personal UI.
- Target conventions win. Read the target repo's local
AGENTS.md/CLAUDE.md if present, then inspect existing posts and build scripts before writing.
- No raw transcript dump. A blog post needs a hook, narrative structure, concise steps, useful captions, and verified links. Timestamps can inform the outline; they are not the post.
- Verify before done. Run the target site's lightweight checks (
typecheck, build, or documented equivalents) unless --dry-run was passed. Report exact blockers.
- Respect existing work. Check
git status --short in the target repo before editing. Do not overwrite unrelated dirty files.
Output contract
At completion, report:
Post: <path>
Assets: <path or none>
Promo: <path or none>
Audits: <summary of delegates/media coverage>
Verify: <commands and pass/fail>
Result: CREATED | DRY-RUN | BLOCKED
Memory Protocol
Append a log entry to .oh/memory/<UTC-date>/log.md for every invocation:
## blog -- HH:MM UTC
- **Result**: CREATED | DRY-RUN | BLOCKED | FAIL
- **Scenario**: <one-line scenario>
- **Source**: <source path/url or inferred>
- **Target**: <target repo/path or inferred>
- **Post**: <post path or none>
- **Promo**: <promo artifact path or none>
- **Observation**: <one sentence about source quality, media risk, promotion readiness, or verification>
Then run the qualify/improve loop from .oh/skills/retro/references/memory-protocol.md. Prefer improving this skill or its references over adding duplicate MEMORY.md notes when the lesson is procedural.
Example
/blog Create blog from @.claude/specs/openharness-demo --target /worktrees openharness-web
Expected behavior: read demo.md, audit all Loom images, use Advisor plus specialized delegates, write a Docusaurus post in the target blog repo, embed only selected safe screenshots using exact demo.md URLs or local files downloaded from those URLs, run validation, and report changed paths.
