# Docs - **Get started** - [Introduction](/introduction): What Repo2RLEnv builds, why it exists, and where it fits next to Harbor. - [Quickstart](/quickstart): Generate, validate and run your first Harbor environment from a real GitHub repository in about five minutes. - [Installation](/installation): Install Repo2RLEnv, add the extras your route needs, and set up the credentials each command uses. - [Choose a pipeline](/choose-a-pipeline): Match your source material, reward and infrastructure to the right generator. - Tutorials - [Tutorials](/tutorials): Practical walkthroughs for building coding-agent evaluations from your codebase with Repo2RLEnv, Tasksmith and Harbor. - [Build an evaluation suite for your codebase with Tasksmith](/tutorials/evaluate-your-codebase): Turn your repository's merged pull requests into repeatable coding-agent evaluations with Tasksmith, Harbor and deterministic tests. - **Tasksmith** - [Tasksmith](/pipelines/tasksmith): Repo2RLEnv's flagship generator: an agent turns a merged pull request into a Harbor environment with a private verifier, then checks and repairs its own work. - [Run Tasksmith on many PRs](/pipelines/tasksmith_parallel_campaigns): Run a bounded, parallel Tasksmith campaign over many merged PRs until you reach a target number of verified tasks. - [HF_ML_Tasksmith dataset](https://huggingface.co/datasets/FineEnvs/HF_ML_Tasksmith) - **Concepts** - [How it works](/concepts/how-it-works): The path from a source repository to a scored, published Harbor environment, and which machine runs each step. - [Anatomy of a task](/concepts/tasks): What each file in a Harbor task does, who can see it, and what Repo2RLEnv records in task.toml. - [Rewards](/concepts/rewards): How each task family is scored, where the score is written, and how to use it for evaluation and RL training. - [Quality and verification](/concepts/quality): How a generated task becomes a verified one: controls, the review and repair loop, and evaluation labels. - [Glossary](/concepts/glossary): Short definitions of the terms used across the Repo2RLEnv docs, each linked to the page that covers it. - Pipelines - [Pipelines](/pipelines): Every way Repo2RLEnv turns repositories, pull requests and task seeds into Harbor tasks, grouped by the kind of task you get. - **Repository repair** - [pr_runtime](/pipelines/pr_runtime): Turn merged pull requests that ship tests into SWE-bench-style tasks graded by the tests the fix makes pass. - [commit_runtime](/pipelines/commit_runtime): Mine bug-fix commits, not pull requests, into tasks graded by the tests each commit makes pass. - [cve_patches](/pipelines/cve_patches): Turn published security advisories into tasks where the agent patches a real vulnerability and a regression test decides the reward. - [swe_smith](/pipelines/repo_mutate) - [swe_next](/pipelines/swe_next) - [r2e_gym](/pipelines/r2e_gym) - **Implementation and reconstruction** - [code_instruct](/pipelines/code_instruct): Have an LLM write coding tasks anchored in a repository's own code, and keep only those its test and solution prove out. - [equivalence_tests](/pipelines/equivalence_tests): Extract real functions from a repository and have an LLM write tests that check a reimplementation against the original. - [r2e](/pipelines/r2e) - [swe_flow](/pipelines/repo_reconstruct) - [swe_gen](/pipelines/pr_to_env) - [codemidas](/pipelines/codemidas) - **Patch similarity** - [pr_diff](/pipelines/pr_diff): Turn merged pull requests into tasks scored by how closely the agent's patch matches the real fix. - **Terminal tasks** - [seta_seed2synth](/pipelines/terminal_synth) - [seta_evol](/pipelines/task_evolve) - [dataarc](/pipelines/dataarc) - [tmax](/pipelines/tmax) - [endless_terminals](/pipelines/endless_terminals) - [terminalworld](/pipelines/terminalworld) - [cli_gym](/pipelines/env_repair) - **Reasoning and optimization** - [scaler](/pipelines/scaler) - [frontiersmith](/pipelines/frontiersmith) - **Guides** - [Run tasks with Harbor](/guides/run-with-harbor): Download a Repo2RLEnv dataset, check it with the oracle and nop controls, score real agents locally or in cloud sandboxes, and read the rewards. - [Environment Bootstrap](/reference/BOOTSTRAP) - [Remote execution](/guides/remote-execution): Set up Modal or Daytona credentials, a campaign budget, a runtime wheel and a worker, so research recipes, Tasksmith and the quality loop can run target code remotely. - [Owned generation recipes](/pipelines/owned_recipes) - [Review and repair a Harbor task](/pipelines/quality_loop) - [Retain every generated task, with evidence labels](/pipelines/task_evaluation_labels) - [Publish to the Hub](/guides/publishing): Share a dataset directory quickly with repo2rlenv push, or cut an immutable, hash-checked release with repo2rlenv release, then pull either one back. - [Releasing Harbor task collections](/pipelines/dataset_release) - [Authentication](/reference/AUTH) - [Container registry authentication](/reference/REGISTRY_AUTH) - [Troubleshooting](/guides/troubleshooting): Symptoms, causes and fixes for the errors you're most likely to hit when generating, validating, running and publishing tasks. - **Reference** - [CLI reference](/reference/cli): Every repo2rlenv command, subcommand and flag with its default, read from the argparse definitions. - [Input / Output Spec](/reference/SPEC) - [Reward Schema Reference](/reference/REWARD_SCHEMA) - [Environment variables](/reference/ENV) - [Python API reference](/reference/API) - [Agent harnesses + how RL traces leave the sandbox](/reference/AGENTS) - [Follow a task through its prompts](/pipelines/prompt_reference) - Full prompts (generated) - [Shared terminal prompts and output schemas](/pipelines/prompts/shared_terminal) - [Harbor review and repair: complete prompt reference](/pipelines/prompts/quality_loop) - [Tasksmith: complete prompt reference](/pipelines/prompts/tasksmith) - [CodeMidas: complete prompt reference](/pipelines/prompts/codemidas) - [SWE-smith: complete prompt reference](/pipelines/prompts/swe_smith) - [R2E: complete prompt reference](/pipelines/prompts/r2e) - [SWE-gen: complete prompt reference](/pipelines/prompts/swe_gen) - [SWE-Next: complete prompt reference](/pipelines/prompts/swe_next) - [R2E-Gym / SWEGEN: complete prompt reference](/pipelines/prompts/r2e_gym) - [SCALER: instruction construction](/pipelines/prompts/scaler) - [Endless Terminals: complete prompt reference](/pipelines/prompts/endless_terminals) - [CLI-Gym: complete prompt reference](/pipelines/prompts/cli_gym) - [SWE-Flow: complete prompt reference](/pipelines/prompts/swe_flow) - [SETA Seed2Synth: complete prompt reference](/pipelines/prompts/seta_seed2synth) - [SETA Evol: complete prompt reference](/pipelines/prompts/seta_evol) - [TMax: complete prompt reference](/pipelines/prompts/tmax) - [TerminalWorld (EuniAI): complete prompt reference](/pipelines/prompts/terminalworld) - [DataArc terminal synthesis (Envs-FORGE-linked code): complete prompt reference](/pipelines/prompts/dataarc) - [FrontierSmith optimization synthesis: complete prompt reference](/pipelines/prompts/frontiersmith) - **Results** - [Published Harbor datasets](/pipelines/releases) - [Yield and cost per task](/pipelines/economics) - [Experiment accounting: models, compute and stage costs](/pipelines/experiment_accounting) - [Native pipeline results](/pipelines/native_results) - **Project** - RFCs - [Pipeline RFCs](/rfcs) - [RFC 0001: pr_diff](/rfcs/0001-pr-diff) - [RFC 0002: pr_runtime](/rfcs/0002-pr-runtime) - [RFC 0003: commit_runtime](/rfcs/0003-commit-runtime) - [RFC 0004: code_instruct](/rfcs/0004-code-instruct) - [RFC 0005: equivalence_tests](/rfcs/0005-equivalence-tests) - [RFC 0006: cve_patches](/rfcs/0006-cve-patches) - [RFC 0007: pr_to_env](/rfcs/0007-pr-to-env) - [RFC 0008: env_setup](/rfcs/0008-env-setup) - [RFC 0009: test_synthesis](/rfcs/0009-test-synthesis) - [RFC 0010: issue_runtime](/rfcs/0010-issue-runtime) - [RFC 0011: repository-owned generation recipes](/rfcs/0011-owned-recipes) - [RFC 0012: swe_smith recipe for repo_mutate](/rfcs/0012-swe-smith-recipe) - [RFC 0013: seta_seed2synth recipe for terminal_synth](/rfcs/0013-seta-seed2synth-recipe) - [RFC 0014: seta_evol recipe for task_evolve](/rfcs/0014-seta-evol-recipe) - [RFC 0015: swe_gen recipe for pr_to_env](/rfcs/0015-swe-gen-recipe) - [RFC 0016: swe_flow recipe for repo_reconstruct](/rfcs/0016-swe-flow-recipe) - [RFC 0017: r2e recipe for equivalence_tests](/rfcs/0017-r2e-recipe) - [RFC 0018: tmax recipe for terminal_synth](/rfcs/0018-tmax-recipe) - [RFC 0019: terminalworld recipe for terminal_reconstruct](/rfcs/0019-terminalworld-recipe) - [RFC 0020: endless_terminals recipe for terminal_synth](/rfcs/0020-endless-terminals-recipe) - [RFC 0021: cli_gym recipe for env_repair](/rfcs/0021-cli-gym-recipe) - [RFC 0022: dataarc recipe for terminal_synth](/rfcs/0022-dataarc-terminal-recipe) - [RFC 0023: swe_next recipe for pr_runtime](/rfcs/0023-swe-next-recipe) - [RFC 0024: r2e_gym recipe for commit_runtime](/rfcs/0024-r2e-gym-recipe) - [RFC 0025: sec_bench recipe for cve_patches](/rfcs/0025-sec-bench-recipe) - [RFC 0026: scaler recipe for reasoning_synth](/rfcs/0026-scaler-recipe) - [RFC 0027: Harbor task review and repair](/rfcs/0027-harbor-quality-loop) - [RFC 0028: Tasksmith PR pilot](/rfcs/0028-tasksmith-pr-pilot) - [RFC 0029: Tasksmith repository bootstrap and resource profiles](/rfcs/0029-tasksmith-hf-scale) - [RFC 0030: Bounded expansion and immutable Harbor releases](/rfcs/0030-campaign-expansion-and-release) - [RFC 0031: CodeMidas source-to-environment recipe](/rfcs/0031-codemidas-recipe) - [RFC 0032: FrontierSmith optimization synthesis](/rfcs/0032-frontiersmith-recipe) - [RFC NNNN: ](/rfcs/TEMPLATE) - Release notes - [CodeMidas: v0.9.2 release notes and dataset audit](/release_notes/codemidas) - [Version history](/release_notes/HISTORY) - [v0.8.2.post3: Image distribution & reproducible push](/release_notes/v0.8.2.post3) - V0.8.3 - [commit_runtime: filter / leak / artifact fixes + 52-env reference dataset (Arc 3)](/release_notes/v0.8.3/findings-commit_runtime) - [pr_diff: Harbor-runnable env + 6-component reward](/release_notes/v0.8.3/findings-pr_diff) - [pr_runtime: graded F2P/P2P reward + scale audit (Arc 2)](/release_notes/v0.8.3/findings-pr_runtime) - Contributing - [Cookbook: adding a new pipeline](/contributing/ADDING_A_PIPELINE) - [Maintain the documentation](/contributing/DOCUMENTATION): How the docs site is built, where content lives, and the rules for changing it. - [Learner Python runtime](/pipelines/learner_runtime) - [Related Work & Provenance](/reference/RELATED_WORK)