Add multi-repository workspace support to SWE-AF#7
Draft
Conversation
Add optional target_repo: str field (default='') to PlannedIssue in
swe_af/reasoners/schemas.py. Enables sprint planner to tag issues with
a target repository for multi-repo builds while preserving full
backward compatibility with existing constructions that omit the field.
Add 4 unit tests in tests/test_planned_issue_target_repo.py covering:
- default value is empty string
- custom value ('api') stored correctly
- model_dump() includes target_repo key
- existing construction patterns without target_repo still work
…eManifest, RepoPRResult models and extend existing schemas - Add _derive_repo_name(url) helper function for extracting repo name from git URLs - Add RepoSpec model with role/url/branch/sparse_paths/create_pr fields and URL + role validators - Add WorkspaceRepo model (frozen=False) with git_init_result field for post-clone mutation - Add WorkspaceManifest model with primary_repo property and full JSON round-trip support - Add RepoPRResult model with repo_name/success/pr_url/pr_number/error_message fields - Extend BuildConfig with repos: list[RepoSpec] field and _normalize_repos validator enforcing single-repo shorthand (repo_url → repos synthesis), exactly-one-primary, no-dup-url invariants, and mutual exclusion of repo_url + repos; add primary_repo computed property - Extend BuildResult: replace pr_url field with pr_results list, add backward-compat pr_url property returning first successful PR URL, override model_dump() to inject computed pr_url - Extend DAGState with workspace_manifest: dict | None = None for JSON-safe checkpoint storage - Extend CoderResult, GitInitResult, MergeResult, IssueResult with repo_name: str = "" - Add tests/test_multi_repo_schemas.py covering all new models, validators, properties, backward compatibility, and all AC-01 through AC-24 acceptance criteria
…Result
In the approval path of run_coding_loop(), populate IssueResult.repo_name
from coder_result.get("repo_name", "") so that multi-repo builds can track
which repository each issue was coded in. Empty string is passed through
unchanged; backfill happens downstream in dag_executor.
Add tests/test_coding_loop_repo_name.py covering the propagation, empty-string
passthrough, and absent-key fallback cases.
…ld path - Add async _clone_repos(cfg: BuildConfig, artifacts_dir: str) -> WorkspaceManifest that clones all repos from cfg.repos concurrently using asyncio.gather + asyncio.to_thread, resolves branches via git rev-parse, and handles partial clone cleanup on failure via shutil.rmtree. - Update build() to detect multi-repo config (len(cfg.repos) > 1) and invoke _clone_repos, setting repo_path from manifest.primary_repo.absolute_path. Pass workspace_manifest=manifest.model_dump() to execute() for multi-repo builds. - Add per-repo PR creation loop for multi-repo builds: iterates manifest.repos, respects ws_repo.create_pr flag, accumulates pr_results: list[RepoPRResult]. - Update BuildResult construction to use pr_results=[...] instead of pr_url=... (pr_url is now a computed property on BuildResult for backward compat). - Add workspace_manifest: dict | None = None parameter to execute() reasoner for future dag_executor multi-repo support. - Replace local _repo_name_from_url with import of _derive_repo_name from schemas. - Add tests/test_clone_repos.py with 17 unit + inspect tests covering AC-23.
…k utility Create swe_af/prompts/_utils.py with workspace_context_block(manifest) function that returns an empty string for None/single-repo manifests and a formatted multi-repo context block for workspaces with 2+ repositories. Fix pre-existing circular import between swe_af.prompts and swe_af.reasoners by making prompt imports in pipeline.py lazy (inside function bodies). This allows `from swe_af.prompts._utils import workspace_context_block` to work. Add tests/test_workspace_context_block.py with 9 unit tests covering None, single-repo, zero-repo, and multi-repo cases including AC-14 and AC-15.
…ge dispatch - Add async _init_all_repos() to dag_executor.py: no-op when dag_state.workspace_manifest is None; concurrently calls run_git_init once per repo via asyncio.gather and stores results back into WorkspaceRepo.git_init_result via the mutable WorkspaceManifest. - Augment _merge_level_branches() with multi-repo dispatch path: when workspace_manifest is set, groups completed IssueResults by repo_name and dispatches one run_merger call per repo concurrently; single-repo path (manifest is None) is preserved verbatim. - Update run_dag() to accept workspace_manifest: dict | None = None parameter, assign it to dag_state.workspace_manifest after _init_dag_state(), and invoke _init_all_repos() when set. - Add IssueResult.repo_name backfill in _execute_level(): when result.repo_name is empty, backfills from issue['target_repo']. - Add workspace_manifest: dict | None = None parameter to execute() reasoner in app.py; threads it through to run_dag(). - Create tests/test_dag_executor_multi_repo.py: 20 unit tests covering all acceptance criteria with AsyncMock call_fn.
…lone-repos and dag-executor-multi-repo
…om CoderResult to IssueResult
…AC-15 Create tests/test_multi_repo_smoke.py with 15 pytest functions (test_smoke_ac01 through test_smoke_ac15) that directly verify each acceptance criterion from the Multi-Repo PRD. Tests cover schema model instantiation, field defaults, validation error conditions, JSON serialization, and workspace_context_block behavior.
…ns and all_pr_results to github_pr_task_prompt - Added pm_task_prompt() to product_manager.py with workspace_manifest parameter - Added architect_task_prompt() to architect.py with workspace_manifest parameter - Added sprint_planner_task_prompt() to sprint_planner.py with workspace_manifest parameter; multi-repo manifest injects target_repo mandate into output - Updated coder_task_prompt() with workspace_manifest, target_repo, and architecture parameters; target_repo resolves to absolute_path for the named repo in the manifest - Added workspace_manifest parameter to verifier_task_prompt() - Added workspace_manifest parameter to workspace_setup_task_prompt() - Added all_pr_results parameter to github_pr_task_prompt() - All new parameters default to None/empty for full backward compatibility - Created tests/test_multi_repo_prompts.py with 30 tests covering AC-16 through AC-22
…th workspace_manifest and multi-repo parameters
…schema contracts AC-01 through AC-15
Add previously untracked integration test files left by coder agents: - test_clone_repos_to_dag_executor_pipeline.py - test_conftest_malformed_planner_execute_nodeids_integration.py - test_execute_workspace_manifest_dag_pipeline.py - test_execute_workspace_manifest_passthrough.py - test_mock_fixture_cross_feature_integration.py
…D acceptance criteria Creates tests/test_multi_repo_integration.py with 25 test functions (test_ac_01 through test_ac_25) that directly translate each acceptance criterion from the multi-repo workspace PRD into inline Python assertions. - AC-01 to AC-13: schema validation (RepoSpec, BuildConfig, WorkspaceManifest, RepoPRResult, BuildResult, DAGState, PlannedIssue, CoderResult, GitInitResult, MergeResult) - AC-14 to AC-15: workspace_context_block empty-string and table output - AC-16 to AC-22: prompt function signature checks (pm, architect, sprint_planner, coder, verifier, workspace_setup, github_pr) - AC-23: _clone_repos async signature - AC-24: duplicate repo name rejection - AC-25: regression check — all multi-repo feature tests pass via subprocess
…suite for all 25 PRD acceptance criteria
- Remove tracked pipeline artifact: examples/pyrust/.claude/plans/ (matches .claude/ gitignore rule; was committed inadvertently by the autonomous pipeline) - Root .artifacts/, .pytest_cache/, .worktrees/ were gitignored but physically present; removed them so working tree is clean - .gitignore already covers all standard Python, Rust, env, and pipeline-specific patterns Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Member
|
@AbirAbbas i think we also need to ensure the issue loop (coder and tester etc..) or even maybe other loops need to have need to have interrepo changes right? Like horizontally touching issues and tests ? 👀 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
RepoSpecobjects (with roles, branches, and sparse-checkout paths) alongside or instead of a singlerepo_url.RepoSpec,WorkspaceRepo,WorkspaceManifest,RepoPRResult) and extendsBuildConfig,DAGState,PlannedIssue,CoderResult,GitInitResult, andMergeResultwith repo-scoped fields._clone_repos), a shared git cache layout, and workspace-context prompt utilities so planners and coders are fully aware of cross-repo file locations.repo_urlstring input.Changes
swe_af/execution/schemas.py,swe_af/reasoners/schemas.pyswe_af/app.py(_clone_repos, updatedbuild())swe_af/execution/dag_executor.py(_init_all_repos,_merge_level_branches,execute())swe_af/prompts/(pm, architect, sprint_planner, coder, verifier, workspace, github_pr)swe_af/prompts/_utils.py(workspace_context_block)swe_af/execution/coding_loop.py(repo_name propagation)tests/test_multi_repo_integration.py,tests/test_clone_repos.py,tests/test_smoke_*.py,tests/test_dag_executor_multi_repo.py, and othersTest plan
python -m pytest tests/test_multi_repo_integration.py -v— all 25 acceptance-criteria tests must pass.python -m pytest tests/ --ignore=tests/fast -q— 283 tests pass, 0 failures (pre-existingtests/fast/failures unrelated to this feature are excluded).repo_url="https://github.com/..."completes without error and returns aBuildResultwith a validpr_urlstring.repos=[RepoSpec(url="...", role="primary"), RepoSpec(url="...", role="dependency")]; confirmworkspace_manifestis populated onDAGStateand per-repo PRs are generated._clone_reposis importable:from swe_af.app import _clone_repos.BuildConfigrejects dual primaries, simultaneousrepo_url+repos, and duplicate URLs viaValidationError.Notes
test_multi_repo_integration.pyruns a curated 11-file subset (not the fulltests/tree) because a pre-existing unrelated failure intests/fast/(AttributeErroronswe_af.fast.app) predates this branch. This deviation is documented in the test docstring.architectureparameter added tocoder_task_promptis intentionally unused for now (documented API surface for future use).146b100have broken fixture dependencies and are not real feature tests — they can be removed in a follow-up chore.🤖 Built with AgentField SWE-AF
🔌 Powered by AgentField