aidemo-blog
Skill32 starsChanged 7 days agoMITView on GitHub
Run a blog-generation wave for aidemo using the centralized blog engine. Use when the user says "run a blog wave", "generate blog articles", "write the next N articles", "blog batch", or "continue the blog generation" for aidemo. This is a THIN adapter — the engine's `blog-wave` skill + its served guides carry the actual doctrine (research-first citations, SDXL hero taxonomy, wave runbook, model tiering). It points at aidemo's local config, style overrides, topic map, and repo-specific gotchas, and mandates filing engine feedback at the end of every wave. Do NOT use for: editing one existing article (edit the JSON + re-bake), deploying the blog (aidemo's deploy target), or changing the engine itself (file feedback instead).
- Commits and pushes
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- docs/llms.txtllms.txt
- blog-waveSkill
- landing-pageSkill
- record-demoSkill
- releaseSkill
- verifySkill
The file
Files
aidemo-blog — blog waves for aidemo (thin adapter)
The engine holds the logic; this file holds only what is specific to aidemo. Do not duplicate doctrine here — fetch it live so it can never go stale.
Run the guide and follow it
blog-engine guide wave-runbook # the 6 phases, disk-is-ledger, packing, launchd autoresume
blog-engine guide authoring # writer contract + non-hallucination doctrine
blog-engine guide sdxl-taxonomy # hero image failure taxonomy + judge rubric
blog-engine guide model-tiering # opus-first; sonnet token-save; codex serial-only
blog-engine guide source-policy # Tier 1-4 citation whitelist
blog-engine guide cadence # plan release cadence before bulk-generatingThe engine-owned blog-wave skill is the orchestrator runbook — it drives the
six phases and defers to the guides above. Follow it. Everything below is
aidemo-local.
aidemo's local files (the engine reads these; you point at them)
blog/blog.config.json— machine-enforced values: site/brand, word & link budgets, banned-diction additions, cluster→style routing, citation policy + host tiers, model tiers + wave size, cadence. The engine gates read this; never restate a config value in this skill.blog/style/writer-overrides.md— aidemo's voice, tone, and claims policy. Appended verbatim to every rendered lane prompt under aSTRICT — these win on conflictheader. This is where editorial voice lives, not here.blog/data/topics.json— the hub-and-spoke topic map (slugs, clusters, pillars, crossLinks, science/evidence hooks).- your generation log — the lab notebook + hero-debt table; append one entry
per wave. This repo keeps it under
blog/data/ordocs/blog/(generation-log.md); if yours is elsewhere, record the exact path in the gotchas region below so future waves andblog-engine feedbackfind it. blog/data/articles/— the corpus (disk is the ledger).
Self-update (do this at a wave boundary, never mid-wave)
A SessionStart hook prints an "update available" notice when the pinned engine version is behind. To take it:
blog-engine skill check # is a newer engine version out?
blog-engine skill update # re-pull the thin skill + re-pin installed.json (also repins launchd loops)skill update is the single deliberate bump point. Guide-only fixes propagate
instantly (served live); a validator/renderer/driver fix needs a version bump.
THE ROUTING RULE — apply at the end of every wave
For each learning this wave, ask: "would another consumer's next wave benefit?"
- Yes → file an engine issue with
blog-engine feedback(categories: guide / lane-template / validator / renderer / judge / driver / bake). - No → make a local edit (voice, topics, model budget, style routing, cadence dates) and record it in your generation log (see the local-files list above).
- Both → local hotfix now and an engine issue for the durable fix, noting the workaround you applied locally.
Filing zero issues is a valid outcome — but it must be stated in the wave report with a reason. The end-of-wave report therefore always carries this line:
Feedback: <N> engine issues filed (<links>) / 0 with reasonaidemo repo-local gotchas
<!-- Everything between the two markers below is CONSUMER-OWNED: `blog-engine skill
update` preserves it verbatim and rewrites everything else. Edit your local
gotchas only between the markers, and never delete the markers themselves. --><!-- consumer:start --><!-- Repo-specific footguns ONLY (deploy quirks, a stale LAN IP, a load-bearing
theme classname, a shared quota with another job, a local preview env var).
Doctrine that any consumer would hit belongs in an engine feedback issue, not
here. Start empty; append as aidemo accumulates scars. -->- Migration COMPLETED 2026-07-20 — waves now run entirely on the engine
(engine v0.4.0) via THIS skill. The old local
blog/scripts/*+ops/lane-prompt.md+ the pre-engineblog-waveskill/workflows were deleted (git history preserves them).validateis 102/102 and the engine bake is byte-identical todocs/blog(422 files). Seeblog/README.md§ "Blog-engine migration (M3)". - Hero lane blocked by blog-engine#15 —
blog-engine heroes …crashes withKeyError: 'lifestyle'because the engine ignoresimages.defaultStyle(aidemo =matte-clay) and hardcodes alifestylefallback. Non-blocking now (all 102 heroes are committed), but the NEXT wave that needs new heroes is blocked until the fix lands — check the issue before generating heroes. - Deploy = git push.
docs/blog/is the committed GitHub Pages output at https://aidemo.top/blog/ — never hand-edit it, and only ever commit a bake whose diff you have reviewed. Live in ~30-60s after push. - robots.txt: the site root
docs/robots.txtowns crawler policy; the blog must NOT ship its own (publish.robots: false— keep it that way). - Parity knobs are load-bearing:
site.codeBlocks: true,site.indexHeading: "The aidemo blog",publish.robots: false, and the engine's shiki pin (exactly 4.3.1) are what make the engine bake byte-identical to the live site. Changing any of them churns committed pages. - Hero lane semantics differ from the engine's: subjects live in
data/hero-subjects.json(one object metaphor per slug), style is the fixed matte-clay wrapper (mirrored inblog/style/styles.json), seed = fnv1a(slug)+seedOffset, sizes 800/400 only, rendered 1216x832. Re-roll by bumping seedOffset, never by prompt-editing. - One SDXL job machine-wide — calorie-slo shares the GPU/backend
(
ps aux | grep sdxl_backendbefore rendering). git commit -s(DCO) is required in this repo; blog commits are pathspec-only, straight to main.- Lifecycle / GSC (2026-09-29).
lifecycle.enabled: true,gsc.siteUrl: sc-domain:aidemo.top; the service account has Restricted read. The key is NEVER in git:export GSC_CREDENTIALS=$HOME/dropshipping-irondust/secrets/google-service-account.jsonbeforeblog-engine gsc sync. Raw pulls (blog/data/gsc/) are gitignored; proposals/reports inblog/data/lifecycle/are committed. PROPOSE-ONLY:gsc sync -> score -> disposition --dry-run -> human review; neverdisposition applywithout the owner or during a Google update rollout. Note: a lifecyclenoindex/301/410changes baked output, anddocs/blog/is the live deploy, so any apply must be followed by a reviewed re-bake. Seeblog/LIFECYCLE-READINESS.md.
<!-- consumer:end -->Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

