Agent Skill Library, suite
The Content Engine: seven skills, one pipeline
One idea in. A finished LinkedIn post, X thread, newsletter section, and video script out. Each skill does one job, so when something is off you know exactly which one to fix. Built from the one agent per task workflow Mark Fershteyn described on the GTM AI Podcast with Coach K.
The archive holds all seven SKILL.md files, the 3 helper scripts, and the setup readme. The scripts are plain Python with no installs and no internet access.
- Runner
Content Engine
The orchestrator. Writes no content itself. Confirms the topic, creates a dated run folder, then hands each step the file the last one produced: interview.md, draft.md, edited.md, hooks.md, repurposed.md, and a retro that appends rules to learnings.md.
Reads your topic. Writes the run folder.
- Step 1
Content Interviewer
Narrows the topic to one takeaway, then asks at least ten questions, one at a time, across five layers: story, proof, mistake, opinion, and reader. Saves the raw answers to interview.md so the drafter never has to invent anything.
Reads topic from you. Writes interview.md.
- Step 2
Content Drafter
Builds a five-line skeleton (open, stakes, body, opinion, close) and checks each line against the interview before writing. Invents nothing, traces every claim to a source, and deliberately runs a little long so the editor has material to cut.
Reads interview.md. Writes draft.md.
- Step 3Includes a script
Content Editor
Three passes in order: structure, line edit, then voice match against your own writing samples. Targets a 20 to 40 percent cut while keeping every fact, then runs lint_copy.py to flag em dashes, AI tells, hedges, and long paragraphs.
Reads draft.md. Writes edited.md.
- Step 4Includes a script
Hook Writer
Produces five opening lines from the finished piece, scores each on specificity, tension, and clarity, validates length with check_hooks.py, and recommends one with a runner-up so the next channel can test the alternative.
Reads edited.md. Writes hooks.md.
- Step 5Includes a script
Content Repurposer
Reshapes one edited piece for LinkedIn, X, newsletter, and short video, each to its own length and structure spec, then runs check_lengths.py against the channel limits. Every version has to stand alone.
Reads edited.md, hooks.md. Writes repurposed.md.
- Step 6
Content Retro
Asks three questions after the piece is done and converts the answers into short dated rules in learnings.md. Every other skill in the suite reads that file and treats its rules as overrides.
Reads everything above plus your feedback. Writes learnings.md.
Set up your workspace
Install all seven skills in your agent, then create three things in the folder where you work.
- voice/ ten or more of your best posts, one per file, so the drafter and editor can match how you write.
- learnings.md starts empty. The retro fills it, and every other skill treats its rules as overrides.
- banned.txt optional. Phrases you never want to see, one per line.
Then say: run the content engine on your topic. Each run saves its files to a dated folder. The first run will be decent. The retro is where it gets good: give honest feedback for about four runs and the output locks into your voice.
Get the Report