Harmix/open-stanley / open-stanley/skills/stanley
stanley
VerifiedYour head of content. Use this whenever the user wants help with LinkedIn or X (Twitter) — writing or scheduling a post, finding things to comment on, turning a meeting, voice note, or Slack thread into a post, reviewing why a post did or didn't work, planning the week's content, or asking "what should I post". Also use it for anything that mentions Stanley, ghostwriting, personal brand, founder-led content, building in public, or growing followers, even if the user doesn't say "post". Start here; it routes to the other open-stanley skills and holds the operating rules they all share.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 14.2 KB |
| references/capabilities.md | markdown | 3.3 KB |
| references/platform-access.md | markdown | 4.0 KB |
| references/recommended-mcps.md | markdown | 5.5 KB |
| references/recommended.mcp.json | config | 2.0 KB |
| references/research-linkedin-2026.md | markdown | 6.7 KB |
| references/scheduling.md | markdown | 8.6 KB |
| references/vault-backends.md | markdown | 6.8 KB |
| references/verify-live.md | markdown | 2.2 KB |
| references/why-these-rules.md | markdown | 2.1 KB |
| scripts/stanley-auditexecutable | script | 24.6 KB |
| scripts/stanley-briefexecutable | script | 6.4 KB |
| scripts/stanley-chartexecutable | script | 14.5 KB |
| scripts/stanley-previewexecutable | script | 5.8 KB |
| scripts/stanley-statsexecutable | script | 14.9 KB |
| scripts/stanley-trace | other | 6.3 KB |
| scripts/stanley-vaultexecutable | script | 35.1 KB |
references/research-linkedin-2026.md
View on GitHub# External evidence on what works on LinkedIn (2026)
Priors for the experiment loop, not rules. The monthly strategy check-in reads this when it proposes the next experiment; `write` and `stanley-audit` carry the few findings that are large, consistent, and cheap to check. Everything here is correlational and from other people's accounts; the user's own ledger decides.
## Virio, "What 226,000 LinkedIn Posts Revealed" (March 2026)
25 research agents over 226K B2B posts, validated on 236 held-out posts, then compared with 20 ghostwriting clients (1,138 posts before, 527 after). Public engagement only; no pipeline metrics. Shared by a customer, October 2026.
Higher engagement, by effect size:
- **Career milestone posts: +56–108%**, the strongest type in every industry. Milestone + gratitude ("Grateful to [Name] for…") adds ~30% on top.
- **Opening with "I…": +8–25%** over "You should…" framing. Across their 20 clients fewer than 10% of posts started with "I", so it is under-used even by paid writers.
- **Ending with a question: +4.4% comments**, 2.2x stronger for B2B executive accounts; asking for comments outright: +6.2% comments. These drive comments, not reactions.
- Specific numbers ("$2M ARR" over "strong revenue growth"): +5%. Naming real people and companies: +5%.
Lower engagement:
- **Opening with a question: −5%**, worse at 3+ posts/week.
- **Two text-only posts in a row: −8 to −11%**, the biggest avoidable penalty at 3+/week.
- **URLs in the body: −4 to −8%**; put the link in the first comment.
- "You should" advice framing: −2%. "Unpopular opinion"/hot takes: the worst risk-reward of any pattern (0.5% viral, 1.3% flop). Tension→resolution arcs underperform a steady positive tone.
Images depend on account size: **under 5K followers +12%**, 5–50K −1 to −2%, 50–100K neutral, over 100K slightly positive. (Their own clients, mostly 5–50K, saw images win anyway; they say so and call it unresolved.)
Reactions are not interchangeable. **Insight (💡) is the single best predictor of reposts** (top-10% insight-share posts get 8x more reposts); Celebrate drives volume but not resharing; Love builds depth, not reach; a high **Like share means people scrolled past** and correlates with lower performance. Insight and Celebrate suppress each other, so a post optimises for one. Insight/Celebrate peak Tuesday–Wednesday, Love on Sunday, humour on Friday.
Cadence and cycles: each extra weekly post costs ~2% per-post engagement but total weekly reach still grows; each additional *format* in a week (text, image, carousel) adds ~3.6%. Creators cycle through slumps (~13 posts), cold plateaus (~19 posts, the hardest to leave), warm growth (~16) and hot streaks (~8, short). This is normal cycling, not a content problem. Posts off a creator's usual topics get ~22% less engagement, while creators who widen topics slowly over months improve: consistency week to week, evolution quarter to quarter.
Their clients after ghostwriting: median post length 522 → 1,670 chars; comments 2 → 4; insight share 2.2% → 4.4%; 14 of 20 improved on engagement, 6 got worse (strong organic baselines or tiny samples).
## Philip Sørensen (Brew), "I finally cracked the LinkedIn algo" (September 2026; 208 reactions, 53 comments)
Founder's own account, not a study. Five claims: (1) write for the followers you already have — LinkedIn's first test batch is largely them, so check Settings → Audience for the top country and write for it; (2) use less AI, a few minutes of your own thinking per post is now the bar; (3) be positive and shout people out after events and meetings, "everything you do is content"; (4) be yourself, the only advantage left; (5) posting time does not matter. The comments push back on (5) and note the trade-off between impressions and reaching the actual ICP. Shared by the user, October 2026.
## Field notes from the first real account (Sept–Nov 2026, under 5K followers)
Two posts reached roughly ten times the account's median. Both opened with a well-known institution's name in the first line (Google Scholar; Y Combinator), both told a first-hand conversation or observation ("I asked a YC partner…", "I noticed Scholar…"), both carried a real image (a screenshot; a photo), both were under 1,300 characters, and both ended on a question. The posts that did worst were feature explanations with perfect receipts and no name a stranger would know. The lesson the plugin took: rank stories by a famous name plus first-hand access plus a real image before ranking by receipt quality, and never let a small implementation detail take a weekly slot.
The best post of the account's first ten weeks needed none of its mining: an observation about a public feed (Google Scholar) with a screenshot. A mining-data creator with 4.7K followers gets 200–1,300 reactions per post from public rankings of listed companies rendered as treemaps, four details from filings, and "let me know what data you would like displayed next". That is the `atlas` skill: a lane any account can post in from day one.
Comments: four scout rounds in a row were rejected on voice until the shape became one plain sentence under 20 words that answers that post's specific claim. Two tidy sentences with a contrast read as generated every time.
Images: rendered cards were called "AI slop" by the user after three weeks; real screenshots and the user's own photos replaced them. Every PNG rendered in a cloud sandbox also carried a C2PA manifest that LinkedIn labels "Content credentials" (`stanley-vault image-clean`).
## What this plugin does with it
| Finding | Where it lives |
|---|---|
| First-person opener; no "you should"; no question opener; no hot takes | `write` §2 and `stanley-audit --platform linkedin` (`question-opener` WARN, `advice-framing` INFO, `hot-take` INFO) |
| Two text-only posts in a row | `stanley-audit` reads the previous post's `image` flag from the ledger; `post-add --image` records it |
| Images for accounts under 5K followers | `my-human.md` `followers:` line; audit says which side of 5K the account is on |
| Ending with a question | the `ending` experiment, classified from the text; an arm with no posts is reported as starving |
| Milestones, gratitude, shout-outs after events | `mine` §2 story kinds `milestone` and `people` |
| Insight share, like share | `metrics-add --insight --celebrate --love --like`; `stanley-stats` reports insight and like share; recap reads the reaction breakdown |
| Performance cycles; on-brand drift | monthly check-in: no lane change on fewer than ~20 posts; slumps are cycles |
| Format mix per week | monthly format-gap step; the weekly drafts vary format when the ledger shows a text-only run |
| Posting time | not a rule; a candidate experiment at most |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |