tether-sdk 0.2.12__tar.gz → 0.2.13__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/AGENTS.md +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/PKG-INFO +8 -9
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/README.md +7 -8
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/benchmarks.md +7 -9
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/index.md +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/context-and-store.md +2 -2
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/release-notes.md +12 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/troubleshooting.md +2 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/built-in-tools.md +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_quirks.py +10 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_benchmarks_docs.py +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_harness.py +13 -4
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_repeat_reads.py +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_tether.py +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/__init__.py +1 -1
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/harness.py +24 -18
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru/active/.gitkeep +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru/active/north-star.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru/archive/.gitkeep +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru/archive/v0-2-bugs.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru/archive/work-on-v0-2-4.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.claude/gru.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.github/workflows/ci.yml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.github/workflows/delivery.yml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.github/workflows/docs.yml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/.gitignore +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/CONTRIBUTING.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/LICENSE +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/README.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/canary.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/canary_baseline.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/compare.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/grade.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/harness-bench/README.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/harness-bench/tether-adapter.patch +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/history.jsonl +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/base.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/baseline/harness.jsonl +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/baseline/ladder.jsonl +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/charts.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/probe.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/refs.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/report.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/run.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/snapshot.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/tasks.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/ladder/validate.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/naive_agent.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/report.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/run.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/hidden/test_hidden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ref/calc/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ref/calc/errors.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ref/calc/evaluator.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ref/calc/parser.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ref/calc/tokenizer.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/calc/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/calc/errors.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/calc/evaluator.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/calc/parser.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/calc/tokenizer.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/tests/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-hard/ws/tests/test_calc.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/hidden/test_hidden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/cli.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/ledger/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/ledger/parser.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/ledger/report.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/sample.csv +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/tests/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-medium/ws/tests/test_report.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-simple/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-simple/hidden/test_hidden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-simple/ws/tests/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-simple/ws/tests/test_textutils.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/code-simple/ws/textutils.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/codereview/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/codereview/ws/inventory.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/content/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/content/ws/.keep +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/hidden/test_hidden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/AGENTS.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/README.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/docs/architecture.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/docs/style.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/mini/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/mini/cli.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/mini/fmt.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/mini/stats.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/mini/textio.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/tests/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/focus/ws/tests/test_stats.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/plan/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/plan/ws/README.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/plan/ws/svc/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/plan/ws/svc/client.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/plan/ws/svc/config.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/qa/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/qa/ws/.keep +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/review/goal.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/bench/tasks/review/ws/paper.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/architecture.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/extending.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/glossary.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/guides/a-pipeline-with-checks.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/guides/add-your-own-tool.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/guides/fix-a-failing-test.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/guides/use-tether-from-your-agent.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/javascripts/mathjax.js +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/img/harness-cost.svg +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/img/harness-pass.svg +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/img/task-grid.svg +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/img/tiers.svg +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/index.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/ladder/results.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/llms.txt +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/agent-interface.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/cli.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/harness.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/index.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/multiplexer-and-decorator.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/providers.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/sandbox-and-skills.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/reference/tools.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/agent-interface.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/cli-and-headless.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/index.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/multi-agent.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/sdk-and-custom-tools.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/security-and-sandbox.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/skills.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/docs/tutorial/workflows-and-telemetry.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/conftest.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/proxy.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_agents.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_defaults.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_faults.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_lifecycle.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_limits.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_sandbox.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/e2e/test_surface.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/README.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/checks.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/course.yaml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/ruff.toml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/skills/planner/SKILL.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/skills/searcher/SKILL.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/skills/searcher/tools.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/skills/writer/SKILL.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/examples/course/skills/writer/tools.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/mkdocs.yml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/pyproject.toml +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/scripts/record_golden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/skills/researcher/SKILL.md +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/conftest.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/fixtures/dialects.jsonl +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/coding_edit_failure_modes.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/coding_edit_failure_modes.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/done_nudge.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/done_nudge.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/read_range_pingpong.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/read_range_pingpong.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/research_web.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/research_web.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/shell_wrappers.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/shell_wrappers.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/waiting_input_resume.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/golden/waiting_input_resume.snap.json +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_anthropic.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_base.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_declarative.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_dialects.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_ollama.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_openai.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/providers/test_retry.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_agent_interface.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_cli.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_cli_docs.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_context.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_context_config.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_decorator.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_docs_structure.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_edit_tool.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_golden.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_guides_docs.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_integration.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_llms_txt.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_misses.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_multiplexer.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_params.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_pipeline.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_providers_docs.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_quiet_and_fold.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_readme.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_release_hardening.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_retro.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_sandbox.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_security_hardening.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_skills.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_store.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_structured_history.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/test_troubleshooting_docs.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_agent.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_bash.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_done.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_edit.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_mem.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tests/tools/test_web.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/api.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/cli.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/context.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/decorator.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/mcp.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/multiplexer.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/pipeline.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/anthropic.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/base.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/declarative.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/ollama.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/providers/openai.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/retro.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/sandbox.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/schema.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/skills.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/store.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/__init__.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/agent.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/bash.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/done.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/edit.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/mem.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/tether/tools/web.py +0 -0
- {tether_sdk-0.2.12 → tether_sdk-0.2.13}/uv.lock +0 -0
|
@@ -54,7 +54,7 @@ Reviewers and the Team Leader cite the failing line when they reject a change.
|
|
|
54
54
|
- Sessions do **not** terminate simply because the model emitted a text response without a tool call (small 9B models frequently "think out loud" before acting).
|
|
55
55
|
- A task only completes when the agent explicitly calls `act("done", stdin="<final deliverable>")` (or `act("done '<short answer>'")`), pauses for human/parent input via `act("done --ask", stdin="<question>")` (transitioning the session to `WAITING_INPUT` and releasing the worker process until resumed via `send`), or hits `max_steps`. A bare `done` (no answer, no stdin) completes with the result `Done.` and still runs the `verify` gates; `done --ask` with no question is an error (#169).
|
|
56
56
|
- If the model emits plain text without calling `act`, the harness appends a brief deterministic nudge reminding it to either execute a command or call `done`.
|
|
57
|
-
- After 6 consecutive read-only/verification steps following a change, the harness appends a `[NUDGE]` to that step's observation reminding the model to call `done` if the task is complete, repeated every 8 more quiet steps until the next change (0 static prompt tokens, #74, #132). Once per session, at 75% of `max_steps` (limits of 8 or more), a one-line `[NUDGE]` says how many steps are left (#133). Before any change, 12 read-only steps in a row (then every 8 more) append a one-line `[NUDGE]` to make the edit or call `done` (#127). Seven reads of one file in a row with no change are a `cycle_stall`, and 4 stall warnings without a change end the session `FAILED` as `Stalled` (#92), after one `[WRAP UP]` turn
|
|
57
|
+
- After 6 consecutive read-only/verification steps following a change, the harness appends a `[NUDGE]` to that step's observation reminding the model to call `done` if the task is complete, repeated every 8 more quiet steps until the next change (0 static prompt tokens, #74, #132). Once per session, at 75% of `max_steps` (limits of 8 or more), a one-line `[NUDGE]` says how many steps are left (#133). Before any change, 12 read-only steps in a row (then every 8 more) append a one-line `[NUDGE]` to make the edit or call `done` (#127). Seven reads of one file in a row with no change are a `cycle_stall`, and 4 stall warnings without a change end the session `FAILED` as `Stalled` (#92), after one `[WRAP UP]` turn (the 4th warning triggers it, one more stall ends the session): a read-only session is asked for its final answer, a session that already changed something is asked to call `done` with what is done and what is left (unless its `verify` gates or green test already finish it); `tether stop <id>` terminates a detached worker (#89).
|
|
58
58
|
- When `verify` gate commands are configured (via `Harness(verify=[...])`, CLI `--verify`, or Skill YAML `verify: [...]`), `Harness.execute_session` deterministically executes each command inside the session's `SandboxPolicy` before allowing `done` to transition to `COMPLETED` (consuming 0 static prompt tokens until a gate fails). A gate that errors instead of failing (sandbox violation, exit 124/126/127) ends the session `FAILED` at once as `gate_error` ("gate misconfigured"), without counting toward the consecutive-failure abort. When the stall guard is about to end a session as `Stalled` and gates are configured, it runs them first: a change was made and every gate passes, so the session ends `COMPLETED` ("Finished by the harness"), since the deterministic gate is the completion criterion the caller asked for (#163). With no gates, the model's own last passing test run (runner-recognised, nothing changed since, still green when re-run) is the implicit gate, at a stall and at the step cap; a passing run after a change also appends a one-line `[NUDGE]` to call `done` (#177).
|
|
59
59
|
|
|
60
60
|
### Rule 4: Progressive `--help` (JIT Schema Disclosure)
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: tether-sdk
|
|
3
|
-
Version: 0.2.
|
|
3
|
+
Version: 0.2.13
|
|
4
4
|
Summary: General-purpose, headless-first LLM-Native micro-harness for small language models
|
|
5
5
|
Project-URL: Homepage, https://github.com/igormcsouza/tether
|
|
6
6
|
Project-URL: Repository, https://github.com/igormcsouza/tether
|
|
@@ -40,21 +40,20 @@ Most agent frameworks treat the context window as a dumping ground. Before your
|
|
|
40
40
|
Tether is built the other way round:
|
|
41
41
|
|
|
42
42
|
- **One tool, not thirty.** The model sees a single `act(cmd, stdin)` schema. Everything else (`bash`, `edit`, `web`, `mem`, your own functions) is a command behind it, off by default. Fixed framing is under 150 tokens.
|
|
43
|
-
- **Old output folds away.** Tool results stay in full
|
|
43
|
+
- **Old output folds away.** Tool results stay in full while there is room; once the history passes half the context, results older than two turns become a one-line tombstone. Everything lives in SQLite and comes back by search.
|
|
44
44
|
- **The envelope is deterministic.** Sandboxing, state, step limits, verification gates and termination are plain code. The model only reasons; it is never trusted with safety.
|
|
45
45
|
|
|
46
46
|
## How well does it work
|
|
47
47
|
|
|
48
|
-
Tether answers to one number: how many real coding exercises a small local model finishes. On the Polyglot-Python set (Aider's 34 Python exercises, graded by each exercise's own tests), `qwen3.5:9b` through Ollama
|
|
48
|
+
Tether answers to one number: how many real coding exercises a small local model finishes. On the Polyglot-Python set (Aider's 34 Python exercises, graded by each exercise's own tests), `qwen3.5:9b` through Ollama at the default 8192-token context, 2 trials each (68 runs):
|
|
49
49
|
|
|
50
|
-
| | v0.2.
|
|
50
|
+
| | before v0.2.10 | v0.2.10 |
|
|
51
51
|
|---|---|---|
|
|
52
|
-
| Passes |
|
|
53
|
-
| Mean score | 0.
|
|
54
|
-
|
|
|
55
|
-
| Steps that did nothing | 19% | 2.2% |
|
|
52
|
+
| Passes | 35 of 68 (51.5%) | 35 of 68 (51.5%) |
|
|
53
|
+
| Mean score | 0.718 | 0.763 |
|
|
54
|
+
| Runs killed by the wall cap | 11 | 14 |
|
|
56
55
|
|
|
57
|
-
One run is noisy (about 0.1), so read the pass count and the failure breakdown, not one decimal. Method, how to reproduce
|
|
56
|
+
The target is 80% by v0.3.0, so this is the number every change is ranked by, and it is flat since v0.2.10. One run is noisy (about 0.1), so read the pass count and the failure breakdown, not one decimal. The second yardstick is the [Coding Ladder](https://igormcsouza.github.io/tether/ladder/): 24 tasks, 6 tiers of difficulty times write, bugfix, plan and review, which shows how far up a model climbs and what stops it. Method, how to reproduce both and fair-comparison rules: [Benchmarks](https://igormcsouza.github.io/tether/benchmarks/).
|
|
58
57
|
|
|
59
58
|
## How it works
|
|
60
59
|
|
|
@@ -19,21 +19,20 @@ Most agent frameworks treat the context window as a dumping ground. Before your
|
|
|
19
19
|
Tether is built the other way round:
|
|
20
20
|
|
|
21
21
|
- **One tool, not thirty.** The model sees a single `act(cmd, stdin)` schema. Everything else (`bash`, `edit`, `web`, `mem`, your own functions) is a command behind it, off by default. Fixed framing is under 150 tokens.
|
|
22
|
-
- **Old output folds away.** Tool results stay in full
|
|
22
|
+
- **Old output folds away.** Tool results stay in full while there is room; once the history passes half the context, results older than two turns become a one-line tombstone. Everything lives in SQLite and comes back by search.
|
|
23
23
|
- **The envelope is deterministic.** Sandboxing, state, step limits, verification gates and termination are plain code. The model only reasons; it is never trusted with safety.
|
|
24
24
|
|
|
25
25
|
## How well does it work
|
|
26
26
|
|
|
27
|
-
Tether answers to one number: how many real coding exercises a small local model finishes. On the Polyglot-Python set (Aider's 34 Python exercises, graded by each exercise's own tests), `qwen3.5:9b` through Ollama
|
|
27
|
+
Tether answers to one number: how many real coding exercises a small local model finishes. On the Polyglot-Python set (Aider's 34 Python exercises, graded by each exercise's own tests), `qwen3.5:9b` through Ollama at the default 8192-token context, 2 trials each (68 runs):
|
|
28
28
|
|
|
29
|
-
| | v0.2.
|
|
29
|
+
| | before v0.2.10 | v0.2.10 |
|
|
30
30
|
|---|---|---|
|
|
31
|
-
| Passes |
|
|
32
|
-
| Mean score | 0.
|
|
33
|
-
|
|
|
34
|
-
| Steps that did nothing | 19% | 2.2% |
|
|
31
|
+
| Passes | 35 of 68 (51.5%) | 35 of 68 (51.5%) |
|
|
32
|
+
| Mean score | 0.718 | 0.763 |
|
|
33
|
+
| Runs killed by the wall cap | 11 | 14 |
|
|
35
34
|
|
|
36
|
-
One run is noisy (about 0.1), so read the pass count and the failure breakdown, not one decimal. Method, how to reproduce
|
|
35
|
+
The target is 80% by v0.3.0, so this is the number every change is ranked by, and it is flat since v0.2.10. One run is noisy (about 0.1), so read the pass count and the failure breakdown, not one decimal. The second yardstick is the [Coding Ladder](https://igormcsouza.github.io/tether/ladder/): 24 tasks, 6 tiers of difficulty times write, bugfix, plan and review, which shows how far up a model climbs and what stops it. Method, how to reproduce both and fair-comparison rules: [Benchmarks](https://igormcsouza.github.io/tether/benchmarks/).
|
|
37
36
|
|
|
38
37
|
## How it works
|
|
39
38
|
|
|
@@ -10,21 +10,19 @@ The **Polyglot-Python** set: the 34 Python exercises of Aider's [polyglot benchm
|
|
|
10
10
|
|
|
11
11
|
- **Graded by the exercise's own tests.** The test file is restored before grading, so editing it cannot pass.
|
|
12
12
|
- **Score** is the fraction of tests that pass (so a partial solution counts); a *pass* is every test green.
|
|
13
|
-
- The **fast set** is 12 exercises that mix passes, partial results and failures, run twice each (24 runs): about an hour and a half
|
|
13
|
+
- The **full set** is all 34 exercises run twice (68 runs), about five hours on one local GPU. The **fast set** is 12 exercises that mix passes, partial results and failures, run twice each (24 runs): about an hour and a half, enough to see a large change.
|
|
14
14
|
|
|
15
15
|
## Latest result
|
|
16
16
|
|
|
17
|
-
`qwen3.5:9b` through Ollama (`num_ctx` 8192),
|
|
17
|
+
`qwen3.5:9b` through Ollama (`num_ctx` 8192), all 34 exercises, 2 trials per exercise (68 runs):
|
|
18
18
|
|
|
19
|
-
| | v0.2.
|
|
19
|
+
| | before v0.2.10 | v0.2.10 |
|
|
20
20
|
|---|---|---|
|
|
21
|
-
| Passes |
|
|
22
|
-
| Mean score | 0.
|
|
23
|
-
|
|
|
24
|
-
| "Tool call cut off" hints (runs that had one) | 29 (9) | 2 (1) |
|
|
25
|
-
| Steps that did nothing (no tool call) | 19% | 2.2% |
|
|
21
|
+
| Passes | 35 of 68 (51.5%) | 35 of 68 (51.5%) |
|
|
22
|
+
| Mean score | 0.718 | 0.763 |
|
|
23
|
+
| Runs killed by the wall cap | 11 | 14 |
|
|
26
24
|
|
|
27
|
-
|
|
25
|
+
The pass count did not move: the extra score comes from partial solutions, and the runs that still fail mostly lose their time to long empty turns of a few thousand tokens. The [Coding Ladder](ladder/index.md) is the second yardstick: it shows where a model stops climbing instead of one pass rate.
|
|
28
26
|
|
|
29
27
|
**Read these numbers with care.** One run of one exercise is noisy (about 0.1 of score), 24 runs are a small sample, and a local server can time out for reasons that have nothing to do with tether. A difference smaller than that noise is not a result. The *failure breakdown* in the report keeps provider outages and timeouts apart from harness outcomes (`max_steps`, `stalled`, `wrong_answer`, `worker_error`) so you can see which one moved.
|
|
30
28
|
|
|
@@ -9,7 +9,7 @@ Most agent frameworks drown small models in 3,000–8,000 tokens of nested JSON
|
|
|
9
9
|
- **One Action Primitive (`act`)**: Every tool call—built-in or custom—routes through a single POSIX-style command multiplexer `act(cmd: str, stdin: str | None = None)`.
|
|
10
10
|
- **Zero Default Tool Bloat**: By default, zero optional tools are loaded. Only the mandatory `done` termination command is active until you opt in via `--tools`, `Harness(tools=[...])`, or a Skill.
|
|
11
11
|
- **Progressive `--help` Disclosure**: The system prompt lists only a 1-line index per enabled tool (`<= 12 tokens`). Full usage signatures (`<= 60 tokens`) are revealed Just-In-Time (JIT) when the agent runs `<cmd> --help` or makes a syntax mistake.
|
|
12
|
-
- **Bounded L1 Context + SQLite FTS5 RAG**: Active observations are capped with Head+Tail elision and folded into 1-line tombstones
|
|
12
|
+
- **Bounded L1 Context + SQLite FTS5 RAG**: Active observations are capped with Head+Tail elision and folded into 1-line tombstones once they are older than $K=2$ turns and the history fills more than half the context, while every raw output is indexed into `.tether/tether.db` (SQLite WAL + FTS5 BM25) for sub-millisecond recall.
|
|
13
13
|
|
|
14
14
|
---
|
|
15
15
|
|
|
@@ -20,9 +20,9 @@ When the provider sets `structured_history = True` (Ollama, OpenAI-compatible, A
|
|
|
20
20
|
|
|
21
21
|
## Observation folding and small models
|
|
22
22
|
|
|
23
|
-
Raw tool outputs stay in the prompt
|
|
23
|
+
Raw tool outputs stay in the prompt while there is room. Once the history fills more than half of `max_context_tokens`, an output older than $K=2$ turns folds into a one-line tombstone (see [Built-in Tools](../tutorial/built-in-tools.md#mem-pinned-scratchpad-bm25-observation-recall)). For small models this has consequences ([#27](https://github.com/igormcsouza/tether/issues/27)), seen in the [course-builder example](https://github.com/igormcsouza/tether/tree/main/examples/course):
|
|
24
24
|
|
|
25
25
|
- **Fix (#69).** Folding exception: the latest read of each of the last 8 files (`edit view <p>`, `bash cat/head/tail/sed -n <p>`; paths normalized against the workspace) is not folded until `<p>` is edited (`edit write|replace`, or any other bash command naming it: `sed -i`, `tee`, redirects, `python -c`) or re-read. The head+tail elision still applies (an observation may use up to a third of `max_context_tokens`, at least 2,400 chars), and the hard context budget remains #81. The `mem recall` hint is shown only when `mem` is enabled (else "re-run the command"). (`ContextAssembler(mem_enabled=..., workspace_root=...)`, `MAX_LIVE_READS`).
|
|
26
|
-
- **Re-viewing files loops, then invents data.** The model runs `edit view`, the output folds two turns later, it views again, and eventually writes from memory.
|
|
26
|
+
- **Re-viewing files loops, then invents data.** The model runs `edit view`, the output folds (two turns later once the context is more than half full), it views again, and eventually writes from memory.
|
|
27
27
|
- **What worked: one compact joined view, used right before the write, that ends with the exact next call.** The `course-source <module>` tool returns a module's plan and chosen videos in one output and finishes with `NOW WRITE THIS MODULE: act("edit write part-N.md", ...)`, so the data is still in the prompt when the write happens.
|
|
28
28
|
- **Write long outputs in small pieces.** One big write dropped a whole section. The writer skill writes one `part-N.md` per module with `edit write`, then a `course-build` tool concatenates them into `course.md`.
|
|
@@ -1,5 +1,17 @@
|
|
|
1
1
|
# Release Notes
|
|
2
2
|
|
|
3
|
+
## v0.2.13 (2026-11-03)
|
|
4
|
+
|
|
5
|
+
The stall guard now gives a session that already wrote its file one more turn to call `done`, and the loop warnings tell the model to write the file instead of answering in chat. Found by running the Coding Ladder on `qwen3.5:9b` at 16k context (17 tasks reached, 14 passed; the two tier-4 stops, write-4 and bugfix-4, are the model's limit, #200). Not benchmarked on Polyglot-Python before the release; that run follows.
|
|
6
|
+
|
|
7
|
+
### Fixes
|
|
8
|
+
|
|
9
|
+
- A session that already changed something and then repeats one command (typically `cat review.md` after writing the review) now gets the same one-shot `[WRAP UP]` turn as a read-only session, asking it to call `done` with what is done and what is left. Before, it ended `FAILED` as `Stalled` with its work on disk and no answer. The read-only wrap-up now also says to write the file the task asked for, if any, before `done`, and the `[LOOP DETECTED]` warnings say "write the file the task asked for or make the next edit" instead of "answer", which the model took as an invitation to reply in chat text instead of writing the file. On `qwen3.5:9b` (ladder review-3, 16k context, 20 runs each) the review file was written in 20 of 20 runs against 9 of 12 before; the model still often repeats `cat review.md` through the warnings and ends `Stalled` with its file on disk. A session whose `verify` gates or green test already pass is still finished by the harness without the extra turn. Found in the Coding Ladder: 9 of 16 stalled `qwen3.5:9b` runs ended this way.
|
|
10
|
+
|
|
11
|
+
### Documentation
|
|
12
|
+
|
|
13
|
+
- README and the benchmarks page report the full 34-exercise Polyglot-Python result for v0.2.10 (35 of 68, mean 0.763) instead of the v0.2.8 fast set, and link the Coding Ladder. The fold description now says results fold only once the history passes half the context (v0.2.12 behaviour).
|
|
14
|
+
|
|
3
15
|
## v0.2.12 (2026-11-02)
|
|
4
16
|
|
|
5
17
|
Two causes of the read-forever loop found in the Coding Ladder on `mistral-large-4` (plan-5, bugfix-5 and bugfix-6 used all 30 steps without writing a file), plus a repeating completion nudge and a steps-left hint (#132, #133). On `qwen3.5:9b` (ladder subset, 3 repeats) scores are unchanged and prompt tokens rise about 15%; on `mistral-large-4` three of five formerly zero cells now score (n=1). Not benchmarked on Polyglot-Python.
|
|
@@ -10,6 +10,8 @@ Entries are grouped by what you see. For problems that are known and not fixed y
|
|
|
10
10
|
|
|
11
11
|
**Cause.** The model repeated a command, cycled through the same few, or kept reading one file over shifting line ranges without editing. The guard ended the session early instead of spending every step.
|
|
12
12
|
|
|
13
|
+
Before the guard ends the session it adds one `[WRAP UP]` turn: a session that only read is asked for its final answer, one that already changed something is asked to call `done` with what is done and what is left. If the model answers, the session completes normally; if it repeats the command again, the session ends `Stalled`.
|
|
14
|
+
|
|
13
15
|
With `--verify` gates configured, a stall after a change whose gates all pass ends `COMPLETED` instead ("Finished by the harness"), because the work is verified.
|
|
14
16
|
|
|
15
17
|
**Fix.** Make the goal name the file and the edit. Run `tether read <id> --steps` to see the loop. Try a stronger model. Raising `--max-steps` only repeats the loop.
|
|
@@ -61,7 +61,7 @@ act(cmd="web fetch https://www.sqlite.org/wal.html")
|
|
|
61
61
|
|
|
62
62
|
## `mem` — Pinned Scratchpad & BM25 Observation Recall
|
|
63
63
|
|
|
64
|
-
When raw tool outputs age past $K=2$ turns, Tether folds them into 1-line tombstones (`[obs_3 folded: ...]`). The `mem` tool lets the agent pin key facts into active context or search/recall folded observations from SQLite FTS5:
|
|
64
|
+
When raw tool outputs age past $K=2$ turns and the history fills more than half of the context budget, Tether folds them into 1-line tombstones (`[obs_3 folded: ...]`). The `mem` tool lets the agent pin key facts into active context or search/recall folded observations from SQLite FTS5:
|
|
65
65
|
|
|
66
66
|
```text
|
|
67
67
|
# Pin a key-value fact into active working memory
|
|
@@ -127,6 +127,16 @@ def test_read_only_stall_is_asked_for_its_answer_once(tether: Tether, chaos) ->
|
|
|
127
127
|
assert "[WRAP UP]" in "\n".join(tether.tool_outputs(r["id"]))
|
|
128
128
|
|
|
129
129
|
|
|
130
|
+
def test_stall_after_a_change_is_asked_to_call_done_once(tether: Tether, chaos) -> None:
|
|
131
|
+
# Ladder review-3: the review was written, then `cat review.md` forever, never `done`
|
|
132
|
+
chaos.script(
|
|
133
|
+
"bash echo review > review.md", *(["bash cat review.md"] * 5), ("done", "reviewed")
|
|
134
|
+
)
|
|
135
|
+
r = tether.run("review", "--tools", "bash")
|
|
136
|
+
assert r["status"] == "COMPLETED" and r["result"] == "reviewed"
|
|
137
|
+
assert "[WRAP UP] You keep repeating" in "\n".join(tether.tool_outputs(r["id"]))
|
|
138
|
+
|
|
139
|
+
|
|
130
140
|
def test_done_with_a_stdin_flag_keeps_the_answer_clean(tether: Tether, chaos) -> None:
|
|
131
141
|
chaos.script("done --stdin 'final answer'")
|
|
132
142
|
r = tether.run("answer")
|
|
@@ -9,7 +9,7 @@ PAGE = (Path(__file__).resolve().parents[1] / "docs" / "benchmarks.md").read_tex
|
|
|
9
9
|
|
|
10
10
|
def test_reports_the_polyglot_score_and_the_method() -> None:
|
|
11
11
|
assert "Polyglot-Python" in PAGE and "polyglot-fast" in PAGE
|
|
12
|
-
assert "v0.2.
|
|
12
|
+
assert "v0.2.10" in PAGE and "restored before grading" in PAGE
|
|
13
13
|
|
|
14
14
|
|
|
15
15
|
def test_shows_how_to_reproduce() -> None:
|
|
@@ -1008,7 +1008,7 @@ async def test_read_only_stall_gets_one_wrap_up_turn_and_can_finish(tmp_path: Pa
|
|
|
1008
1008
|
run = await _run_commands(tmp_path, ["ls"] * 5) # 4 loop warnings, then `done` with "ok"
|
|
1009
1009
|
assert run.session.status == "COMPLETED" and run.session.result == "ok"
|
|
1010
1010
|
sent = "\n".join(m["content"] for m in run.provider.recorded_calls[5])
|
|
1011
|
-
assert "[WRAP UP]" in sent
|
|
1011
|
+
assert "[WRAP UP]" in sent and "the file the task asked for" in sent
|
|
1012
1012
|
|
|
1013
1013
|
|
|
1014
1014
|
async def test_ignoring_the_wrap_up_still_ends_stalled(tmp_path: Path) -> None:
|
|
@@ -1018,7 +1018,16 @@ async def test_ignoring_the_wrap_up_still_ends_stalled(tmp_path: Path) -> None:
|
|
|
1018
1018
|
assert sent.count("[WRAP UP]") == 1 # once per session
|
|
1019
1019
|
|
|
1020
1020
|
|
|
1021
|
-
async def
|
|
1022
|
-
|
|
1021
|
+
async def test_changed_session_stall_gets_a_wrap_up_that_can_finish(tmp_path: Path) -> None:
|
|
1022
|
+
"""A session that wrote its file and keeps `cat`-ing it is asked to call done."""
|
|
1023
|
+
run = await _run_commands(tmp_path, ["ping a.com", *["ls"] * 5])
|
|
1024
|
+
assert run.session.status == "COMPLETED" and run.session.result == "ok"
|
|
1025
|
+
sent = "\n".join(m["content"] for m in run.provider.recorded_calls[6])
|
|
1026
|
+
assert "[WRAP UP]" in sent and "what is done and what is left" in sent
|
|
1027
|
+
|
|
1028
|
+
|
|
1029
|
+
async def test_ignoring_the_wrap_up_after_a_change_still_ends_stalled(tmp_path: Path) -> None:
|
|
1030
|
+
run = await _run_commands(tmp_path, ["ping a.com", *["ls"] * 12])
|
|
1031
|
+
assert run.session.status == "FAILED" and (run.session.result or "").startswith("Stalled")
|
|
1023
1032
|
sent = "\n".join(m["content"] for c in run.provider.recorded_calls for m in c)
|
|
1024
|
-
assert "[WRAP UP]"
|
|
1033
|
+
assert sent.count("[WRAP UP]") == 1
|
|
@@ -74,6 +74,6 @@ async def test_escalated_loop_message_from_the_second_repeat(tmp_path: Path) ->
|
|
|
74
74
|
provider = FakeScriptedProvider(_script(["ping x"] * 4))
|
|
75
75
|
harness = Harness(db_path=tmp_path / "tether.db", tools=[ping], provider=provider)
|
|
76
76
|
await harness.run("goal")
|
|
77
|
-
assert "Do not repeat
|
|
77
|
+
assert "Do not repeat it" in _prompt(provider, 2)
|
|
78
78
|
assert "already have this output above" not in _prompt(provider, 2)
|
|
79
79
|
assert "already have this output above" in _prompt(provider, 3)
|
|
@@ -7,6 +7,6 @@ import tether
|
|
|
7
7
|
|
|
8
8
|
def test_version_and_public_exports() -> None:
|
|
9
9
|
"""Verify package __version__ and public SDK symbols."""
|
|
10
|
-
assert tether.__version__ == "0.2.
|
|
10
|
+
assert tether.__version__ == "0.2.13"
|
|
11
11
|
for symbol in tether.__all__:
|
|
12
12
|
assert hasattr(tether, symbol)
|
|
@@ -780,7 +780,7 @@ class Harness:
|
|
|
780
780
|
changes: list[str] = []
|
|
781
781
|
green_cmd: str | None = None # last passing test run, cleared by any later change
|
|
782
782
|
green_nudged = False
|
|
783
|
-
wrapped_up = False # one last "
|
|
783
|
+
wrapped_up = False # one last "finish now" turn before a stall ends the session
|
|
784
784
|
quiet_run = 0
|
|
785
785
|
steps_hinted = False
|
|
786
786
|
|
|
@@ -914,26 +914,31 @@ class Harness:
|
|
|
914
914
|
|
|
915
915
|
if is_loop_stall or cycle or range_stall:
|
|
916
916
|
stalls += 1
|
|
917
|
-
if stalls >= STALL_ABORT and not changed and not wrapped_up:
|
|
918
|
-
# Read-only work (a review, a plan, an answer) that keeps re-reading: give
|
|
919
|
-
# the model one more turn to answer from what it has instead of ending empty
|
|
920
|
-
wrapped_up, stalls = True, STALL_ABORT - 1
|
|
921
|
-
self.store.add_turn(
|
|
922
|
-
session_id=session_id,
|
|
923
|
-
step=step,
|
|
924
|
-
role="tool",
|
|
925
|
-
content=(
|
|
926
|
-
"[WRAP UP] You keep re-reading. Stop and answer now from what you "
|
|
927
|
-
'have read: call `act("done", stdin="<your final answer>")`.'
|
|
928
|
-
),
|
|
929
|
-
)
|
|
930
|
-
continue
|
|
931
917
|
if stalls >= STALL_ABORT:
|
|
932
918
|
# A model that fixed the task and then only repeats its check has
|
|
933
919
|
# finished: if every verify gate passes after a change, say so (#163)
|
|
934
920
|
finished = await self._finished_by_harness(
|
|
935
921
|
sandbox, verify_commands, green_cmd, changed
|
|
936
922
|
)
|
|
923
|
+
if not finished and not wrapped_up:
|
|
924
|
+
# Work that keeps re-reading (a review, a plan, a file it already
|
|
925
|
+
# wrote) gets one more turn to deliver instead of ending empty
|
|
926
|
+
wrapped_up, stalls = True, STALL_ABORT - 1
|
|
927
|
+
self.store.add_turn(
|
|
928
|
+
session_id=session_id,
|
|
929
|
+
step=step,
|
|
930
|
+
role="tool",
|
|
931
|
+
content=(
|
|
932
|
+
"[WRAP UP] You keep repeating a command. If the task is "
|
|
933
|
+
'complete, call `act("done", stdin="<what is done and what '
|
|
934
|
+
'is left>")`.'
|
|
935
|
+
if changed
|
|
936
|
+
else "[WRAP UP] You keep re-reading. Stop and answer now from "
|
|
937
|
+
"what you have read: write the file the task asked for, if "
|
|
938
|
+
'any, then call `act("done", stdin="<your final answer>")`.'
|
|
939
|
+
),
|
|
940
|
+
)
|
|
941
|
+
continue
|
|
937
942
|
self.store.update_session(
|
|
938
943
|
session_id,
|
|
939
944
|
status="COMPLETED" if finished else "FAILED",
|
|
@@ -969,14 +974,15 @@ class Harness:
|
|
|
969
974
|
if loop_run == 1:
|
|
970
975
|
stall_msg = (
|
|
971
976
|
f"[LOOP DETECTED] You just ran `{cmd.strip()}` with identical input "
|
|
972
|
-
"in the previous step. Do not repeat the
|
|
973
|
-
"
|
|
977
|
+
"in the previous step. Do not repeat it: use the output above, then "
|
|
978
|
+
"do what the task asked (write or edit the file) or call "
|
|
974
979
|
'`act("done", stdin="...")` to finish.'
|
|
975
980
|
)
|
|
976
981
|
else:
|
|
977
982
|
stall_msg = (
|
|
978
983
|
"[LOOP DETECTED] You already have this output above. Use it now: "
|
|
979
|
-
|
|
984
|
+
"write the file the task asked for or make the next edit, then call "
|
|
985
|
+
f'`act("done", stdin="...")`. Repeating `{cmd.strip()}` changes '
|
|
980
986
|
"nothing."
|
|
981
987
|
)
|
|
982
988
|
self.store.add_turn(
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|