This Free Terminal Agent Made Me Delete Claude Code (Oh-My-Pi Full Test)
A full hands-on test of Oh-My-Pi (the om CLI), an MIT-licensed terminal coding agent forked from Mario Zechner's minimal Pi, covering its role-based model routing, its in-process Rust core that skips fork-and-exec for every tool call, and the hashline edit format that lifted Grok Code Fast from a 6.7% to a 68.3% pass rate with no retraining.
AI Stack Engineer10 minTranscript found
Quick learning frame
Read this before watching.
Creative automation uses agents to accelerate production while keeping human taste in story, pacing, selection, and critique.
New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Skill you build: The ability to judge a coding agent by its harness mechanics, edit format, tool execution model, and language server and debugger integration, rather than by which model it happens to be running.
Watch for the shift from claim to mechanism. The learning value is the point where the transcript reveals a repeatable action, tool boundary, context move, review habit, or artifact.
Concept diagram
Where this video fits.
01Brief
02Source
03Generation
04Selection
05Edit
06Taste Review
Deep lesson
Turn this video into working knowledge.
1,644 cleaned transcript words reviewed across 488 timed caption segments.
Thesis
This Free Terminal Agent Made Me Delete Claude Code (Oh-My-Pi Full Test) teaches a practical creative automation move: A full hands-on test of Oh-My-Pi (the om CLI), an MIT-licensed terminal coding agent forked from Mario Zechner's minimal Pi, covering its role-based model routing, its in-process Rust core that skips fork-and-exec for every tool call, and the hashline edit format that lifted Grok Code Fast from a 6.7% to a 68.3% pass rate with no retraining.
The goal is not to remember the video. The goal is to extract the operating principle, tie it to timestamped evidence, test how far the claim transfers, and make something reusable.
0:00
Harness beats weights
“There's a benchmark going around where one coding model jumped from a 6.7% pass rate to 68.3%. Same model, same weights, same prompt. Nobody retrained anything. The only thing that changed was the tool it was running inside.”
The same model with the same weights and the same prompt went from a 6.7% to a 68.3% pass rate purely because of the tool it ran inside, and that tool is om, a fork of Mario Zechner's deliberately tiny Pi agent, MIT licensed and free, installed by a one-line curl script, Homebrew, a native PowerShell script on Windows with no WSL, or bun 1.3.14 or later. Install om, run the completions command for your shell and drop it into your shell config, then confirm it picked up your existing Claude Code, Cursor or Copilot rules and MCP definitions without conversion.
2:46
Roles, not one model
“Local Works 2, Alma, LM Studio, Llama.cp. CPP and VLLM all connect without any key. But the clever bit is the RO system. You're not picking one model. You assign a default model for normal work, a cheap...”
om reads rules, skills and MCP servers from eight config formats in their original shape, which removes the switching cost, and then splits model choice into roles: a default for normal work, a cheap model for the small role that handles subagent fan-out, a heavy reasoning model for slow, and a separate plan-mode model, with OAuth sign-in reusing coding subscriptions and an automatic fallback chain when a provider starts returning 429s. Write your own four-row role table assigning a specific model and its cost to default, small, slow and plan, then swap between them mid-session to feel the difference.
8:38
IDE-grade tooling
“just watch read only from a browser with frames encrypted client side and there's a memory layer called hindsight where the agent saves facts about your repo as it works and loads a compressed mental model of the...”
Renaming a function imported in five places including a barrel file went through the language server rather than grep and replace, so every import, alias and re-export updated as one atomic workspace operation across 14 wired language server operations, and 28 debug adapter protocol operations let om attach a live debugger to a hung process, pause it, walk the stack and read in-memory variables to identify both sides of a planted deadlock, using Delve for Go, debugpy for Python and LLDB for native code. Plant a deliberate deadlock or a rename that crosses a re-export, then compare how a grep-based agent and a language-server-plus-debugger agent handle it.
01
Brief
Start with this video's job: A full hands-on test of Oh-My-Pi (the om CLI), an MIT-licensed terminal coding agent forked from Mario Zechner's minimal Pi, covering its role-based model routing, its in-process Rust core that skips fork-and-exec for every tool call, and the hashline edit format that lifted Grok Code Fast from a 6.7% to a 68.3% pass rate with no retraining. Treat "Brief" as the outcome you are trying to make visible, not a topic label. Anchor it to 0:00, where the video says: “There's a benchmark going around where one coding model jumped from a 6.7% pass rate to 68.3%. Same model, same weights, same prompt. Nobody retrained anything. The only thing that changed was the tool it was running inside.”
02
Source
Use "Source" to locate the part of the creative automation workflow the video is demonstrating. Ask what changes in your real setup if this claim is true. Anchor it to 2:46, where the video says: “Local Works 2, Alma, LM Studio, Llama.cp. CPP and VLLM all connect without any key. But the clever bit is the RO system. You're not picking one model. You assign a default model for normal work, a cheap...”
03
Generation
Turn "Generation" into the reusable artifact for this lesson: A creative workflow board with critique criteria and review checkpoints. This is where watching becomes something you can inspect and reuse.
04
Selection
Use "Selection" as the application surface. Decide whether the idea touches a browser flow, a local file, a model choice, a source document, a UI, or a review step.
05
Edit
Use "Edit" to prove the lesson. The evidence should connect back to the video title, transcript anchors, and a concrete output, not a generic best-practice claim.
06
Taste Review
Use "Taste Review" to carry the idea forward: save the prompt, checklist, diagram, or operating rule that would make the next agent run better.
Example
Source-backed work packet
Convert the video into a scoped task that includes the transcript claim, target workflow, acceptance criteria, and proof. The output should be a creative workflow board with critique criteria and review checkpoints..
Example
Claim vs. demo brief
Separate what the speaker claims, what the demo actually proves, and what still needs outside verification before you adopt the workflow.
Example
Teach-back module
Transform the lesson into a definition, a mechanism diagram, one misconception, one practice exercise, and a check-for-understanding question.
Do not learn it wrong
Treating the title as the lesson without checking what the transcript actually says.
Letting the prompt drift into generic advice that could apply to any video in the playlist.
Copying the tool setup without identifying the operating principle that transfers to your own stack.
Skipping the artifact, which means the learning never becomes operational or inspectable.
Do not count this as learned until these are true.
01
State the transcript-backed claim in your own words: A full hands-on test of Oh-My-Pi (the om CLI), an MIT-licensed terminal coding agent forked from Mario Zechner's minimal Pi, covering its role-based model routing, its in-process Rust core that skips fork-and-exec for every tool call, and the hashline edit format that lifted Grok Code Fast from a 6.7% to a 68.3% pass rate with no retraining.
02
Explain the practical stakes without hype: New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
03
Map the idea onto the Brief -> Source -> Generation -> Selection -> Edit -> Taste Review sequence and name the weakest link.
04
Produce the artifact and include the evidence that proves it: A creative workflow board with critique criteria and review checkpoints.
Put it into practice
Give this grounded prompt to Codex or Claude after watching.
You are helping me turn one specific YouTube video into real, durable learning.
Source video:
- Title: This Free Terminal Agent Made Me Delete Claude Code (Oh-My-Pi Full Test)
- URL: https://www.youtube.com/watch?v=wNw9fKErhdg
- Topic: Creative Automation
- My current learning frame: Install om, assign a cheap model to small and a reasoning model to slow, build a single-file canvas Snake game through a dozen small follow-up edits to see whether any edit fails, then plant a deadlock and make the agent debug it with the attached debugger instead of print statements.
- Why this matters: New playlist item from AI Stack Engineer; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Transcript anchors from this exact video:
- 0:00 / Evidence 1: "There's a benchmark going around where one coding model jumped from a 6.7% pass rate to 68.3%. Same model, same weights, same prompt. Nobody retrained anything. The only thing that changed was the tool it was running inside."
- 2:46 / Evidence 2: "Local Works 2, Alma, LM Studio, Llama.cp. CPP and VLLM all connect without any key. But the clever bit is the RO system. You're not picking one model. You assign a default model for normal work, a cheap..."
- 4:32 / Evidence 3: "like every other agent does, the real implementations are compiled straight into the process. RIP Grep in process, file walking in process. Even Bash itself, a vendored shell called brush lives inside the binary with sessions that persist..."
- 6:52 / Evidence 4: "language server, the same rename protocol your IDE uses. So every import, alias, and reexport updated as one atomic workspace operation. It has 14 language server operations wired in covering diagnostics, references, and symbols. So the agent sees..."
- 8:38 / Evidence 5: "just watch read only from a browser with frames encrypted client side and there's a memory layer called hindsight where the agent saves facts about your repo as it works and loads a compressed mental model of the..."
Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable claims from the video. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A creative workflow board with critique criteria and review checkpoints.
5. Include:
- a plain-English definition of the core idea
- a diagram or structured model using this sequence: Brief -> Source -> Generation -> Selection -> Edit -> Taste Review
- 3 concrete examples that apply the video idea to real agentic work
- 2 failure modes the video helps prevent
- a checklist I can use the next time I run Codex or Claude
- one practical exercise with a clear done signal
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.
Quality bar:
- Make this specific to "This Free Terminal Agent Made Me Delete Claude Code (Oh-My-Pi Full Test)", not a generic Creative Automation essay.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- If evidence is weak, say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.
Misconceptions
What to stop believing.
Creative AI removes the need for taste.
It increases the need for taste because output volume explodes.
The best prompt is enough.
References, critique, iteration, and post-production matter just as much.
Practice studio
Learning only counts when you make something.
01
Transcript evidence map
Separate what the video actually says from what you already believe about the topic.
3 source-backed takeaways with timestamps, confidence, and a transfer note.02
One useful artifact
Apply the video to a real workflow and produce a creative workflow board with critique criteria and review checkpoints..
A reusable artifact with a done signal and one verification step.03
Teach-back card
Explain the lesson to someone who has not watched the video yet.
A 90-second explanation, one diagram, one example, and one misconception to avoid.
Recall check
Answer first, then reveal — without rewatching.
What single change took a coding model from a 6.7% to a 68.3% pass rate without retraining?
What are om's model roles and what is each one for?
How did om find the planted deadlock instead of guessing from source?
Source shelf
Use the video as a doorway, then verify with primary sources.
