terminal_synth — owned terminal task generation¶

The first implemented recipe is seta_seed2synth, adapted from SETA. It preserves the released seed-to-idea and test-first datapoint-building stages. The controller writes structured author outputs as Harbor tasks, executes a fresh baseline and reference, and returns failures to the builder for bounded repair.

Pipeline, step by step¶

flowchart TD
  S["JSON / JSONL question and optional answer"] --> D["Deduplicate and order seeds"]
  D --> P1["P1 · Extract capabilities and draft task design"]
  P1 --> P2["P2 · Author tests, fixtures, instruction and reference"]
  P2 --> V["Validate TerminalDraft"]
  V --> Q["Q1 · Optional draft consistency review"]
  Q --> E["Write temporary Harbor bundle"]
  E --> R["Remote fresh nop + oracle trials"]
  Q -->|"Blocking issue"| F
  V -->|"Schema error"| F["Previous draft + concrete failure feedback"]
  R -->|"Wrong reward or execution failure"| F
  F -->|"At most max_repairs"| P2
  R -->|"nop 0; oracle 1"| H["Export Harbor task and input lineage"]

P1, P2, … identify actual model calls. Unlabelled stages are code or remote execution.

Choose the skill. A seed is evidence about a real workflow. One design call extracts core_capabilities and a draft_spec; capability extraction and task design are not separate API calls.

Materialize the task. The builder returns all files through TerminalDraft. It is told to reason about tests first; the controller does not run a separate test-author call for this recipe.

Execute and repair. The controller owns Dockerfile scaffolding, test contracts and reward code. It rebuilds the task remotely and returns actual schema, baseline or oracle failures to the same builder.

Every prompt and its data¶

Two authoring calls on the first successful attempt: design, then builder. With review_drafts: true, Q1 adds a consistency review before execution. Each repair repeats the builder and, when enabled, Q1. See the shared Q1 walkthrough.

Read the complete seta_seed2synth prompt reference for every retained template, appended instruction, substitution, example and output schema. The shared prompt guide explains how to inspect the fully resolved request from a real run.

Follow one task¶

Illustration: a question about filenames containing spaces becomes a directory-processing task with awkward filenames, a precise output contract, a shell reference and tests that inspect the resulting files. The seed answer informs design but is not automatically copied into the learner instruction.

What repeats, what is checked¶

A completed pair must return exactly 0 for the unsolved task and 1 for the reference. A model self_review is a consistency note, not proof that either execution passed. Parent/source metadata and model requests remain separate from learner-visible files.

An exported bundle is a generation result. Independent leakage review, shortcut probes and blind solver traces belong to the later quality campaign.

Implementation map¶

• seta_seed2synth/recipe.py
• terminal/runner.py
• terminal/draft.py
• terminal/grade.py

Input and CLI¶

Supply a JSON array or JSONL of seed objects. Each requires source, title and question_text; include answer_text, tags, url, attribution and license metadata when available. Exact duplicate records are deduplicated before model calls. Seed files are input data, not an upstream package dependency.

{"source":"unix_linux_se","title":"Preserve filenames with spaces","question_text":"A batch script splits paths containing spaces. How can its traversal preserve complete filenames?","answer_text":"Use a delimiter that cannot occur in filenames and keep expansions quoted."}

Create a campaign and remote worker as described in owned recipes, build the owned runtime with uv build, then run:

repo2rlenv generate --config examples/owned-seta.yaml
repo2rlenv generate --config examples/owned-seta.yaml --resume --json

Run receipts retain source identity, model requests, task designs, each materialized attempt, author self-review and actual Harbor trial results. A completed export is reused only when its content matches the recorded identity. Ambiguous remote outcomes stop dispatch; they are not silently retried.

Supported profile and adaptations¶

The current profile is a Linux CPU container based on Python 3.12, with bash, jq, sqlite3, git, curl, tmux, uv and pytest. Tasks may install additional dependencies during image build. Execution is offline. Systemd, privileged networking, GPUs and external services are outside this profile. Verifiers inspect the final container state; reference and test files are excluded from the learner image build context.

The upstream idea and datapoint prompt files are retained with their Apache-2.0 license. Owned runtime additions replace folder/tool output with strict JSON, legacy Harbor metadata with schema 1.3, and network-installing test scripts with preinstalled dependencies and deterministic JUnit parsing. Five to ten weighted tests and author self-review remain part of the method; reward is binary. This is a documented workflow adaptation, not byte-identical upstream execution.

The released collection contains 100 tasks: 23 retained tasks and 77 new exports. All 77 new exports have matching baseline reward 0 and reference reward 1 receipts. The expansion recorded $54.64, comprising $41.74 of model usage across 617 calls and \(12.90 of estimated worker/build costs. That is **\)0.71 per new export**, including unsuccessful attempts, with no outstanding reservations. The retained tasks and later independent validation are outside this cost scope.

The new exports map to 77 distinct source questions. Common source tags are jq (18), bash (15), find (15), awk (13), sed (13) and tar (10); tags overlap. Source URLs and question/answer attribution remain attached to the task lineage. Eight source records lack a recorded content license; the report preserves that gap instead of assigning an inferred license. See the dataset manifest, generation economics and published Harbor dataset.

Specification audits, adversarial verifier checks and blind solver evaluations follow generation. A passing author self-review is not independent quality acceptance. The release inventory records labels across methods.

See RFC 0013 and the packaged pipelines/recipes/seta_seed2synth/provenance.md for source mapping and notices.