§ Earned Media Skills · by 701am
Twenty-eight AI agent skills for earned media, journalist outreach, and the long work of being remembered. Free. Open source. No framework, no runtime, no setup beyond the clone.
9:47 a.m. · Wednesday · Some newsroom
The reporter is on her second coffee. She has eighty-one unread emails, a deadline at three, and an editor who wants the draft by lunch. She is scanning subject lines. Re: circling back. Delete. Quick question. Delete. Hope this finds you well — delete before she even reads the rest. By the time she stops scrolling she has dispatched seventeen pitches in under a minute, and she has not opened one of them.
She did not delete them because she is callous. She deleted them because every single subject line told her, in advance, that the body would not be worth her time. The shape of the pitch betrayed it.
This is the problem. Not that pitches don't get sent. That they don't get read.
Earned media is the coverage you cannot buy — the feature, the quote, the citation, the moment a third party decided your work was worth their reader's attention. It carries more weight than any ad you will ever run. Getting it consistently is a craft, and the craft begins with the subject line a working reporter does not delete at 9:47 a.m. on a Wednesday.
You have tried it. You ran a prompt. The model produced something plausible-sounding. You read it again and felt — what's the word — embarrassed. Not because the grammar was wrong. Because every working reporter on earth has seen this exact email five hundred times.
The tells, in the order they get spotted:
These are not stylistic preferences. They are signals. Each one tells a reader, in two seconds or less, that the next ten minutes of their life is not what is in the body of this email. They learn to delete by shape.
Each skill is a folder with one file: SKILL.md. Your
agent reads it and follows it. No framework. No runtime. No
configuration. Clone the repo. Point Claude at the folder. It works.
One skill — client-context — holds the client's voice,
proof points, angle library, journalist history, and competitive
landscape. Every other skill reads it first. That is why the output
sounds like your client wrote it, not a template did.
Pitches sent, replies received, placements logged — all in plain markdown, in an Obsidian vault you can read with your own eyes. The system gets smarter each week because it remembers what it already tried.
Before any outreach output reaches you, it runs through a twenty-four-pattern check. The leverage, the seamless, the I hope this finds you, the buried hook — caught and removed. What you see is what a journalist sees.
№ 01
You know your story better than any agency could in three months of onboarding. You don't need a strategist. You need the system to run the parts that are craft — the right journalist for this angle, the pitch under a hundred and fifty words, the follow-up that adds a fact instead of asking again. The skills handle the craft. You handle the relationships.
№ 02
You can run a structured program — weekly monitor, thirty-day editorial calendar, disciplined outreach, monthly report your CEO will actually read — without adding headcount. The portfolio manager gives you visibility across product lines or business units simultaneously. One operator. Full coverage.
№ 03
One vault per client. The portfolio manager scans every client simultaneously and tells you which reporter is being pitched by two different accounts this week, which angle type is landing across the book, which client has gone quiet. The kind of cross-client awareness you cannot have from inside any single engagement.
№ 04
You ran a prompt. The output was generic. You closed the tab. Fair. The client-context layer and the humanizer were built for exactly this problem. The output is specific because it is built from the client's actual voice and tested angles. Then every word runs through the twenty-four-pattern check. Try it once. If it still sounds like a robot wrote it, walk.
Each one is a single markdown file. Read them. Fork them. Rewrite them for the way your desk works.
★ Read aloud at the desk, every morning ★
Every output must pass one test.
Would a journalist feel respected receiving this?
If the pitch inflates the client's importance, if the press release reads as promotion, if the follow-up adds nothing new — the skill says so before producing the work. You fix it or you do not send it.
This is not a constraint placed upon the system. It is the system. Reporters who learn — over months, over years — that the pitches arriving from a sender are worth their time will open them on sight. That trust is the asset. The trust is what earned media actually is.
— The desk at 701am
git clone https://github.com/701am/earned-media-skills.git
"Run the intake skill for [client name]"
Intake asks one question at a time — business, voice, founding story, product proof, angle history, competitors. Answer honestly. At the end you have a client context file every other skill reads.
Paste the system prompt from AGENTS.md at the start
of your session. Tell Claude where the skills folder is. Run
intake. Same result. Same vault.
MIT licensed · Use in client work, products, forks · Pull requests welcome
★ Earned media that earns it ★
We give this away because the alternative — every PR firm in the world running its own brittle, hand-coded version of the same workflows — produces worse pitches for every reporter on earth. The work is hard enough without that.
Use it. Fork it. Tell us what we missed. Build a better follow-up sequence and open a pull request. Build a smarter newsjack and the next operator who clones this repo will inherit your work.
— D.K. · 701am · Filed at sunrise