Terminus · Onboarding
High-quality tasks, built by people who know the craft.
A plain-English guide for Project Terminus — Harbor, Claude skills, and the Snorkel loop — even if words like Docker and GitHub are new.
02 / What you’ll learn
Eight guides for shipping accepted tasks.
Each page is authored for new hires — graded against the same craft rules reviewers use.
Words you’ll hear
Plain English for Docker, Harbor, oracle, Snorkel — built for people who aren’t deep in tech yet.
Install the stack
macOS or Windows/WSL — VS Code, Docker, uv, Harbor — then smoke-test until oracle scores 1.
Work from the terminal
Install STB, authenticate, create and test tasks, submit revisions, review assignments, and fix common CLI errors.
Every field explained
instruction.md, task.toml, Docker, oracle, tests, rubrics — what each piece means in Edition 2.
Slash-command recipes
/discover → /research → /convert, plus local-checks, human-review, snorkel-zip, and task-debug.
Read Snorkel results
HARD / MEDIUM / Solvable, pass rates, NEEDS_REVISION — what to do when the report card lands.
Assessment principles
Fair traps, no leakage, always write reward — prepare without dumping answer keys.
Ask AI with context
Copy focused prompts for planning, oracle and nop debugging, difficulty rework, quality checks, and reviewer revisions.
03 / How it runs
From idea to accepted.
Pipeline
How a task moves from idea → accepted
Read in order. Loops mean go back and fix. Step 08 is done.
Pick a track
Your lead assigns the kind of reasoning (optmodel, datalake, vulns…).
Discover ideas
Type /<track>-discover — get 3 ideas. Pick the one with the best trap.
Research
/<track>-research turns one idea into a Go / No-Go plan.
Convert
/<track>-convert builds the real task folder (Docker, oracle, tests).
Validate locally
oracle = 1 · nop = 0 · ruff clean · /local-checks · /human-review
Upload to Snorkel
/snorkel-zip → upload ZIP in the web UI → wait for AI eval.
Read & revise
Too easy? Harden. Unsolvable? Add fair disclosure. Re-upload same ID.
Accepted
HARD (or MEDIUM), Solvable, ≤5/10 passes, reviewer OK → stop.
05 // Ready when you are
Join the loop powering frontier CLI benchmarks.
Read Craft before assessment. Sign in to track your PRs and link Snorkel submission IDs.