dev-audit
Post-mortem on what the user has already published on dev.to - which posts actually worked, why, and what to stop doing. Uses the user's own analytics plus ~/.claude/devto/log.md history.
naheel0 · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add naheel0/dev-skills --skill dev-audit技能说明
name: dev-audit description: >- Post-mortem on what the user has already published on dev.to - which posts actually worked, why, and what to stop doing. Uses the user's own analytics plus ~/.claude/devto/log.md history.
dev-audit
The only honest source of what works for a dev.to account is that account. Every rule in every dev.to guide is a prior. The user's own last 30 posts are the evidence.
Input
Ask for whichever the user has:
- The dev.to analytics export (Dashboard -> article stats). Or API via
GET https://dev.to/api/analytics/...withapi-key+acceptheaders (totals,historical,past_day,referrers,dashboard). - Or per-post screenshots with views, reactions, comments, reading time.
- Or just the posts and their reaction counts, which is enough for a first pass.
- Also read
~/.claude/devto/log.mdif it exists, since it records which title formula and tags each post used.
What to actually measure
Raw impressions/views are the least useful number, because they are mostly a function of how many followers the user already has. Compute these instead, and show the working:
| metric | how | what it tells you |
|---|---|---|
| Engagement rate | (reactions + comments + bookmarks) / views | whether the post earned its reach |
| Comment ratio | comments / reactions | whether it started something or just got a nod |
| Reach multiple | views / follower count | whether it travelled past the existing audience |
| Save/send rate | bookmarks / views | strongest single predictor of future reach |
Rank by engagement rate and reach multiple, not views. A post with 900 views and 40 comments beats the one with 12,000 views and 6.
Analytics caveats (warn honestly)
- Average read time is precomputed server-side and internally inconsistent (totals vs daily rows can disagree 1.7–6.1x). Treat as directional only.
- Historical data degrades for old posts (~15% coverage for 2017, 95% for 2019, 100% for 2026). Gate computations on coverage — track posts with ≥80% of lifetime views covered.
- Follower count is not readership. Reciprocal-follow farming inflates it; ~37% of usernames look auto-generated. Don't weight reach multiple too heavily if follower growth looks suspicious.
- Median post half-life is ~4 days. Most views land in the first week.
Then find the pattern
With the top 5 and bottom 5 side by side, look for what actually separates them, and be willing to conclude something the user will not like:
- Title formula. Which numbers from
title-formulas.jsonare in the top 5? - Format. Tutorial, listicle, discussion, opinion, announcement, series.
- Length. 1–3 min vs 4–7 min vs 8–13 min reads.
- Tags. Which tag stacks appear in the top 5? Which are absent?
- Cover image. Do top posts have covers? Do bottom posts lack them?
- Canonical URL. Do syndicated posts underperform native ones?
- Day and time — check this last, and only if the other four show nothing. It is almost never the cause.
- First-hour comments. Posts the user replied to inside an hour vs not.
State the finding as a claim with the evidence attached, and say how confident it is. With 30 posts you can see a pattern; with 6 you cannot, and you should say that instead of inventing one.
Output
AUDIT · 31 posts · Jun 12 - Sep 5
TOP 5 BY ENGAGEMENT RATE
8.1% #12 Comparison "I cut proposal time from 5h to 20m" 1,940 views
6.4% #4 Story "I Passed The Google Interviews" 2,210 views
...
BOTTOM 5
0.4% #1 Listicle "7 tools every dev needs" 11,400 views
...
WHAT THE DATA SAYS
1. Posts where you were the one who looked bad: mean 6.2% vs 1.1% for
everything else. n=6. Strongest signal, not close.
2. Tool listicles get views and nothing else. High reach, no comments.
Three of your bottom five.
3. Cover image present: mean 4.1% vs 1.8% without. n=14.
4. Day of week shows nothing. Your Tuesday mean and your Friday mean are
inside the noise. Stop optimising it.
STOP: listicles about tools.
DO MORE: the ones with a cost you paid, and a number.
Then hand the conclusions to a planning step so next week's posts are built on the user's own evidence rather than on defaults.
Self-improvement: learn()
After every audit, run python3 ~/.config/opencode/skills/dev-learn.py audit --apply.
This reads ~/.claude/devto/log.md, computes per-post engagement metrics, and:
- dev-post: promotes title formulas and tags in the top-5 engagement rate (weight += 0.1), demotes bottom-5 (weight -= 0.1), archives formulas/tags whose weight drops to 0. Requires ≥10 audited posts before any change.
- dev-profile: flags rubric items for manual re-weighting based on which fixes users actually apply.
- dev-human: no automatic slop changes (voice-specific); user reviews
recurring slop words from
humanize.py --reportoutput.
Never deletes formulas/tags/items — demotes weight to 0 and moves to archive.
Re-run dev-post after learning to see updated suggestions.
Skipped (ponytail)
- No auto-fetch of analytics API (requires DEVTO_API_KEY + manual setup).
- No unfollow/follow automation (Forem API exposes follow-add only).
- No schedule optimization beyond "Tue/Wed/Thu ~9–10am ET" prior.