| name | review-blog-post |
| description | Review a draft blog post for adam-young.co.uk from two independent angles — a Staff iOS engineer (senior-credibility) and a mid-level iOS engineer (does-this-teach) — and return one merged, deduplicated report. Use when Adam wants a substance/audience review of a post before publishing, after blog-post-writer has drafted it. Triggers: "review this post", "review the blog post", "get the reviewers on it", "/review-blog-post". |
review-blog-post
Run two independent reviews of a draft blog post and merge them into one report Adam can act on. The two reviewers judge substance and audience reception, not prose voice — voice is the adam-voice skill's job, and the reviewers are told to defer to it.
0. Find the draft
You'll be given a slug, a path, or nothing:
- Slug (
clean-architecture) → the file is Content/blog/<slug>.md.
- Path → use it directly.
- Nothing → if there's an obvious in-progress draft (the most recently modified file under
Content/blog/, or the one just written this session), use it and say which. Otherwise ask which post.
Confirm the file exists before spawning anyone.
1. Spawn both reviewers in parallel
Launch the two agents in a single message (two tool calls) so they run concurrently and — more importantly — in isolated contexts. They must not see each other's notes; the whole point is two independent reads, not one persona-switching pass.
staff-ios-reviewer — the senior-credibility lens (claim defensibility, earned opinions, glossed edge cases, the embarrassment test, code correctness).midlevel-ios-reviewer — the does-this-teach lens (assumed prerequisites, unexplained jargon, followable examples, where a learner gets lost).
Hand each the same thing: the draft's path (and slug). Each returns a structured markdown review with a verdict, strengths, and severity-grouped issues.
2. Merge the two reviews
Produce one report, not two stapled together:
- Two verdicts up top. Show both reviewers' one-line verdicts side by side (
Staff: Ship with fixes / Mid-level: Lands with gaps), so Adam sees both audiences at a glance.
- Strengths, briefly merged. If both praised the same thing, say it once.
- Issues, deduplicated and severity-ordered. Walk both lists:
- Where both reviewers raised the same point, merge into one entry and mark it
[both] — that's the strongest signal and goes first within its severity.
- Otherwise tag each entry with its source —
[staff] or [mid] — so Adam knows whose lens it came from.
- Keep each reviewer's severity. If they disagree on severity for the same merged point, take the higher and note the split.
- Preserve the section/quote anchor and the concrete fix from the original.
- Voice pointers, collected. Any "→ voice" notes either reviewer made get gathered into one line that says: run the
adam-voice lint. Don't expand them — that's the other skill's job.
Order issues: [both] High → [staff]/[mid] High → Medium → Low.
3. Hand back
Present the merged report to Adam. Do not edit the draft, commit, or push — these are notes, he decides what to act on. End with a one-line bottom line: given both lenses, is this ready, ready-with-fixes, or needs another pass.
Boundaries
- The reviewers are read-only and so are you here — produce notes, change nothing.
- Substance and audience only. If the merged report is light because the only problems are voice/style, say that and point at
adam-voice rather than padding.
- Don't average the two reviewers into mush. Their disagreements are signal — a thing the Staff reviewer loved and the mid-level reviewer found impenetrable is exactly what Adam needs to see, so keep both.
