Stop pasting links one at a time. Liner curates the sources for one specific job into a mixtape your AI can actually read.
Payoff
Take one real job. Watch what the AI has to work with before Liner, and after.
The job
"Buy into an HDFC income-restricted co-op in Harlem and pass its board."
Generic answer
Ask and it can answer. It might even run a quick search. The result usually follows whatever ranks first: generic home-buying advice that glosses over the HDFC income caps, the resale formula, the flip tax, and what this board actually wants. Ask again tomorrow and you'll get a different version.
Mixtape-backed answer
Now the AI answers from the rules, the board process, and the trade-offs that shape the decision.
Yes, really. Every word on this page was written from a Liner mixtape: 25 curated sources on landing-page copywriting, compiled into context and handed to the AI that drafted it. Liner is free and open source. Inspect the mixtape that wrote this page →
Plain files you own
Liner's TUI is just the authoring surface. The thing you keep is ordinary Markdown and YAML: sources, notes, synthesis, and one assembled MIXTAPE.md.
Inspect the real mixtape that wrote this pageharlem-hdfc-coop/
tape.yaml synthesis.md MIXTAPE.md sources/ 01-income-caps.md 02-board-package.md 03-resale-limits.md working/ evaluation.yaml
source notes
AI-ready01 / NYC HDFC rules
Primary source. Establishes income caps, resale formula, and the part most buyers miss.
02 / Board package prep
Practical checklist. Turns abstract eligibility into the actual packet and interview.
03 / Is it worth it?
Contested source. Keeps the AI grounded in trade-offs across more than one answer.
You paste articles into Claude. You drop transcripts into ChatGPT. You keep half a dozen tabs open as "context" for the thing you're working on. Then you rebuild the same stack of sources the next time, or in the next tool.
That work is real, and right now it's invisible and disposable. Liner turns the work into an artifact you keep, reuse, and hand to the next AI tool.
Wedge
Skill
A skill teaches an AI how you want work done: the steps, voice, rubric, constraints, and taste. You can write one yourself. It shines when the same method applies across many jobs.
"run my landing-page critique"
"evaluate candidates with this rubric"
"summarize meetings in our house style"
Mixtape
A mixtape gives any AI the material for one concrete job: current rules, primary docs, contested takes, examples, and local files. It shines when the source set changes the answer.
"buy into an HDFC
income-restricted co-op in
Harlem and pass its board"
How they stack
Use a skill when the method is the reusable asset. Use a mixtape when the source context is the reusable asset. Because a mixtape is just a folder, any skill or agent that can read files can use it.
Ask the skill to use the mixtape as its source layer. The skill brings the method. The mixtape brings the research, examples, and source notes. When one mixtape keeps powering repeated work, it can become the evidence base for a new skill.
Gallery
Dev-tool maker · real mixtape
"Write a landing page that gets hype-allergic AI power users to install my TUI, starting from a blank category."
compiled output
MIXTAPE.md 25 curated sources 4 positioning lenses 1 synthesis pass ready for Claude Code ready for Codex
App developer
"Get my mental-health app past Apple's App Review while staying inside the medical-claims rules."
YouTuber
"Grow a Eurorack modular-synth channel from 200 to 10k subscribers."
First-time buyer
"Buy into an HDFC income-restricted co-op in Harlem and pass its board."
Expat freelancer
"File my US taxes as a US citizen freelancing from Lisbon while avoiding duplicate tax work."
Coming soon
"Your job goes here. That's the whole idea."
The real question
Sometimes, yes. A prompt works when the model already has enough context. Liner helps when the job turns on sources outside the model: niche rules, contested calls, current docs, local files, or a source set you need to reuse.
A prompt
A prompt asks the model to work from what it already knows, plus whatever search or retrieval the tool happens to run. That can be enough for familiar work and quick questions.
A mixtape
A mixtape hands the model the specific sources you chose: primary, contested, current, and local. The model reasons over the material that actually governs the job. Same model. Better inputs. Better answer.
It compounds
A search is a one-shot. Results shift every time you run it. A mixtape is a folder you keep and reuse across chats, sessions, and tools. The same vetted sources go in every time, so your project holds a consistent voice and standard.
Full sources stay available
A mixtape keeps the real sources. Each time you use it, the AI can pull the exact source and passage the question needs, at full length. The detail stays available for the question you discover later.
Better together
A skill teaches an AI how to do a task. A mixtape gives it what to work from.
Run them together: "use my copywriting skill with this mixtape". The skill brings the method. The mixtape brings the trusted sources for this job.
Curation is the judgment, the framing, and the opinionated choice of which sources earn their place. Liner makes that work easier to keep, inspect, and reuse.
Liner • The Mixtape Methodology $npx linersh
Outcomes
01
Passing the board of an HDFC co-op in Harlem. Filing taxes as a US citizen freelancing from Lisbon. The narrower the job, the more a mixtape can improve the answer.
02
A mixtape is a folder. Upload it to a Claude project. Point an agent or coding assistant at it. Or drop the master MIXTAPE.md into any chat. Same artifact, every tool.
03
Every source carries a curator's note. Every mixtape opens with a synthesis: your distilled take on the domain. You always know what's in the context, and why it's there.
04
A mixtape is a project folder. Send it to a teammate. Publish it. Fork someone else's and bend it to your job. And when one keeps earning its place across projects, that's your signal to promote it to a skill.
Why this works
A mixtape is curated context with standards. Liner prefers primary sources, concrete examples, and source notes that explain why each item belongs. That judgment is what keeps context useful.
Method
How Liner actually builds a mixtape. Each phase has a clear input and a clear output. The methodology catches coverage gaps before it ships.
01 / Framing
You hand Liner a real job with a named outcome. It extracts the JTBD and sketches a knowledge map of the domain.
01 / Framing
You hand Liner a real job with a named outcome. It extracts the JTBD and sketches a knowledge map of the domain.
02 / Candidate discovery
Liner follows citations, mines expert reading lists, pulls from course syllabi. Recall first; precision later.
03 / Fetching
Liner pulls every candidate into the local cache: web pages, transcripts, PDFs, your own files. Evaluation starts after the source is actually read.
03 / Fetching
Liner pulls every candidate into the local cache: web pages, transcripts, PDFs, your own files. Evaluation starts after the source is actually read.
04 / Evaluation
Liner rates each source against your JTBD. Every kept source gets a curator note with why it belongs, where the value is, and what's dated.
05 / Quality checks
Redundancy, coverage, disagreement, framing-gap. Catches useful sources with coverage gaps before they ship.
05 / Quality checks
Redundancy, coverage, disagreement, framing-gap. Catches useful sources with coverage gaps before they ship.
06 / Synthesis
Liner drafts the distilled view of the domain. That framing is what your downstream AI inherits before it touches a source.
07 / Final assembly
Liner packages everything into a portable folder: MIXTAPE.md, sources, and working notes. Any AI that can read files can use it.
07 / Final assembly
Liner packages everything into a portable folder: MIXTAPE.md, sources, and working notes. Any AI that can read files can use it.
Honest counts. When a fetch fails, Liner tells you which source and why. Security-check pages and stale cache entries stay out of the total.
Honest scope
Liner shines when
Use a lighter path when
The honest test
Will a curated context bundle make an AI meaningfully better at a specific job? If the answer is yes, Liner has a job to do.
Compared
Hosted tools like NotebookLM and Claude Projects let you upload documents into a workspace. Useful. Liner gives you a context artifact you can inspect, version, and carry across tools.
/// 01
Your mixtape is a folder of YAML and Markdown. Share it, review it, fork it, version it in Git.
/// 02
Your context lives as YAML and Markdown in a folder you own.
/// 03
The framing you bring to the domain becomes part of the artifact.
/// 04
Re-fetch sources, recompile the mixtape, hand the same shape to a different AI tool tomorrow.
The honest trade-off: Liner is less polished than a hosted product. It asks more of you. What you get in return is control, portability, and inspectability.
Install
$npx linersh
+ macOS & Linux (Windows soon)
+ Bring your own agent: Claude Code or Codex
+ Free, MIT-licensed, open source
+ Runs locally, no telemetry
Questions
Q /// 01
Q /// 02
Q /// 03
Q /// 04
Q /// 05
Q /// 06
Q /// 07
Q /// 08
Q /// 09