feat: parallelize the task, on by default — and close every open issue - #156
Merged
Merged
Conversation
**The unit of parallelism is one task, not several.** v0.1.83 shipped hive mode batching a PR group's remaining tasks into one lead session. That was the wrong unit: the ask was "parallelize each task". The batching, the `TASKS COMPLETE:` manifest, the multi-checkoff and the batch index math are DELETED — `core/loop_working_hive.py` and `core/task_runner_hive.py` are gone, `core/hive.py` is 275 → 52 LOC, and `loop_working_stage.py` is byte-identical to its pre-#155 self. The loop is what it always was: one session per task, one task checked off, `TASK COMPLETE`. What replaces it is smaller. The agent running a task session is a **lead**: it may cut *its own single task* into pieces with disjoint write sets and hand each to a `hive-worker` subagent, doing everything that overlaps itself. The orchestrator cannot tell a fanned-out session from a solo one except by what lands in the tree — so there is no new bookkeeping. **On by default** (`TaskOptions.parallel`, `--parallel`/`--no-parallel`, replacing `--parallel-tasks`; deleted, not aliased). `CLAUDETM_HIVE_MAX_PARALLEL` is a ceiling, not a target — the lead sizes its own team, and zero or one worker are both legitimate answers. One shared checkout, never worktrees; the file set is the only lock and the lead alone runs git. Subagent output is now coloured per worker instance, keyed on its `tool_use_id` so a worker keeps its colour for life and two concurrent `hive-worker`s are separable at a glance. `NO_COLOR`/non-TTY suppress it; log files never receive escapes. Closes every open issue: - **#153** `merge-pr` deleted the *checked-out* branch, not the merged PR's — it destroyed an unrelated open PR's branch in the wild. Both call sites fixed (CLI and orchestrator) against one shared policy in `core/git_branch.py`: the PR's head branch only, only after a merge confirmed against GitHub, never the base, `-d` first with `-D` only when nothing is unpushed. `_GitOps._delete_local_branch` (unconditional `-D`) deleted. - **#152** `merge-pr` printed "merged successfully" on a failed merge and exited 0, then ran the destructive cleanup anyway. The exit code is now authoritative, the merge is verified against GitHub before anything is deleted, and a policy refusal names `--admin`. - **#147** the review-stage timeout warned and proceeded toward merge while the CI-stage timeout blocked — same timer, opposite posture. One shared policy now: block by default, `--admin` force-advances. Tolerated failures can no longer cause the wait or the block. - **#146** every description claimed PRs merge "when CI passes and approved"; no approval concept existed. Prose corrected everywhere, and `reviewDecision` is now read: `CHANGES_REQUESTED` blocks an auto-merge (deliberately even under `--admin`, since `--admin` is passed on every run here — a human clears it on GitHub, so it cannot deadlock). - **#116** a 404 on a never-started job's logs surfaced as `GitHubError` every cycle. It is now a classified signal, and a CI failure GitHub has declared permanent (billing lock, Actions disabled) trips a breaker that posts one explanatory PR comment and stops. The issue was partly stale: on current main that case already spent zero sessions. - **#100** the container mounted the operator's `~/.claude`, so the agent ran as and billed whoever owned that token — a real cross-developer leak. Every mount is gone from `Dockerfile`, `docker-compose.yml`, all five compose examples, the Swarm stack and the Kubernetes manifest; auth is a scoped, rotatable api-key profile in a named volume. Also fixes a layering inversion introduced along the way (`core/` importing from `cli_commands/`) and quotes four compose `${VAR:?msg}` lines that made YAML parse a list item as a map — those examples never parsed. Closes #153, closes #152, closes #147, closes #146, closes #116, closes #100 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
Warning Review limit reachedYou’ve reached a temporary PR review limit under our Fair Usage Limits Policy. Next review available in: 36 minutes Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available. How can I continue?After more reviews become available, a review can be triggered using the To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews. How do review limits work?CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability. For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window. Please refer docs for additional details. Review details⚙️ Run configurationConfiguration used: Path: .coderabbit.yml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (14)
📝 WalkthroughWalkthroughThis PR replaces PR-group hive batching with default-on per-task worker fan-out. It adds merge verification, safe branch cleanup, review and CI timeout policies, permanent external CI detection, scoped Docker API-key profiles, and corresponding API, CLI, MCP, documentation, and test updates. ChangesParallel execution and public contracts
Merge and CI controls
Deployment and documentation
Estimated code review effort: 5 (Critical) | ~120 minutes Possibly related issues
Possibly related PRs
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 13
🧹 Nitpick comments (9)
tests/core/test_prompts_working.py (1)
1279-1293: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚖️ Poor tradeoffConsider moving the fan-out tests into their own module.
The new classes push this file to roughly 1530 lines. The house limit is 500 lines per file. A focused module such as
tests/core/test_prompts_working_hive.pywould holdTestParallelOffIsByteIdentical,TestParallelIsPurelyAdditive,TestFanOutRules, andTestMaxParallelPlumbingwithout changing any assertion. The digest baselines are a nice touch, so keeping them grouped in one place helps.Based on learnings: "In the Python durability test suite (tests/core/*.py), keep each test module under the 500-line limit... split scenarios into focused modules rather than growing a single file". As per coding guidelines: "Keep each file at or below 500 lines of code; split larger modules according to single responsibility and SOLID principles."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/core/test_prompts_working.py` around lines 1279 - 1293, Move the fan-out test section from test_prompts_working.py into a focused module such as test_prompts_working_hive.py, including TestParallelOffIsByteIdentical, TestParallelIsPurelyAdditive, TestFanOutRules, and TestMaxParallelPlumbing. Preserve all assertions, digest baselines, imports, and shared helpers required by those tests, and keep each test module under the 500-line limit.Sources: Coding guidelines, Learnings
src/claude_task_master/cli_commands/merge_finalize.py (2)
191-201: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueSmall polish for
checkout_and_pull.Two optional points:
- The docstring has no
Args/Returnssections. Other public functions in this module use the full Google style, so this one stands out.run_git("pull", timeout=120)depends on the base branch having an upstream. If it does not, the pull fails, the function returnsFalse, and cleanup is skipped. That is a safe outcome, so this is only a readability note.♻️ Optional docstring expansion
def checkout_and_pull(branch: str) -> bool: - """Checkout a branch and pull latest changes. Returns True on success.""" + """Checkout a branch and pull the latest changes. + + Args: + branch: Branch to check out. + + Returns: + True when both the checkout and the pull succeed. + """As per coding guidelines: "Use Google-style docstrings for public APIs".
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/claude_task_master/cli_commands/merge_finalize.py` around lines 191 - 201, Expand the checkout_and_pull docstring to Google style by documenting the branch argument and the boolean return value, while preserving the existing checkout and pull behavior.Source: Coding guidelines
46-62: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueConsider whether
"not authorized"belongs inPOLICY_REFUSAL_MARKERS.A "not authorized" error usually means the token lacks permission.
--admincannot help in that case, because--adminalso requires admin rights. The hint text does mention "requires admin rights on the repo", so the advice is not wrong, only sometimes unhelpful. Keep it if you prefer a hint over silence.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/claude_task_master/cli_commands/merge_finalize.py` around lines 46 - 62, Review POLICY_REFUSAL_MARKERS and remove "not authorized" unless the merge error handling can reliably distinguish a base-branch policy refusal from a token-permission failure. Ensure ADMIN_HINT is shown only when --admin could plausibly override the refusal, while preserving the existing markers and case-insensitive matching.tests/cli_commands/test_pr_resolution.py (1)
22-36: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winOptional: add a case for the
#52input form.
parse_pr_inputaccepts a leading#, but no test exercises that branch. A one-line case would cover it.🧪 Optional test addition
def test_url_is_parsed(self) -> None: client = MagicMock() assert resolve_pr_number(client, "https://github.com/owner/repo/pull/52") == 52 + + def test_hash_prefixed_number_is_parsed(self) -> None: + client = MagicMock() + assert resolve_pr_number(client, "`#52`") == 52As per path instructions: "Review test coverage and edge cases."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/cli_commands/test_pr_resolution.py` around lines 22 - 36, Add a test case in TestExplicitInput covering resolve_pr_number with the “#52” input form, and assert it returns 52 without requiring Git or branch lookup.Source: Path instructions
tests/cli_commands/test_merge_finalize.py (1)
113-122: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winOptional:
checkout_and_pullhas no direct test.It is the last gate before
delete_merged_branchruns, so its failure paths matter. A short test for the "checkout fails" and "pull fails" cases would lock in theFalsereturn.🧪 Optional test sketch
class TestCheckoutAndPull: def test_checkout_failure_returns_false(self) -> None: with patch.object(merge_finalize, "run_git", return_value=_proc(returncode=1)): assert merge_finalize.checkout_and_pull("main") is False def test_pull_failure_returns_false(self) -> None: with patch.object( merge_finalize, "run_git", side_effect=[_proc(), _proc(returncode=1)] ): assert merge_finalize.checkout_and_pull("main") is FalseAs per path instructions: "Review test coverage and edge cases."
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/cli_commands/test_merge_finalize.py` around lines 113 - 122, Add direct coverage for merge_finalize.checkout_and_pull: test that a failed checkout returns False, and that a successful checkout followed by a failed pull also returns False. Patch merge_finalize.run_git with the appropriate return values, using the existing process-result helper, while leaving successful behavior unchanged.Source: Path instructions
src/claude_task_master/cli_commands/pr_resolution.py (1)
156-166: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueOptional: the fallback message can confuse a user who already passed
--create-pr.If
create_pris true and the current branch is a default branch, the code falls through to Line 165 and suggestsclaudetm merge-pr --create-pr, which the user just used. A branch-specific message would explain the real reason. The CLI normally blocks this earlier throughvalidate_not_default_branch(), so this only affects direct callers ofresolve_pr_number.♻️ Optional message split
branch = get_current_branch() - if create_pr and branch and branch not in DEFAULT_BRANCHES: - created = open_pr_for_current_branch(branch) - if created is not None: - return created - raise typer.Exit(1) + if create_pr: + if branch and branch not in DEFAULT_BRANCHES: + created = open_pr_for_current_branch(branch) + if created is not None: + return created + raise typer.Exit(1) + console.error(f"Cannot open a PR from '{branch or 'an unknown branch'}'.") + console.info("Checkout a feature branch first: git checkout -b <branch>") + raise typer.Exit(1)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/claude_task_master/cli_commands/pr_resolution.py` around lines 156 - 166, Update the fallback messaging in resolve_pr_number so create_pr=true on a default branch reports that PR creation is unavailable for the current branch instead of suggesting --create-pr again. Preserve the existing generic no-PR guidance for other cases and retain the current exit behavior.src/claude_task_master/cli_commands/fix_pr.py (1)
113-139: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueOptional: the fallback normalization here duplicates
verify_merged.
verify_mergedalready normalizeshead_branchandbase_branchthrough_text, so both arestr | Noneby construction. Theisinstanceguards at Lines 135-139 only protect against aPRStatusmock supplying a non-stringbase_branch. Also,merge_failure_hintreturnsNonewheneveradminis true, so the extranot adminon Line 119 is already implied. Neither is harmful; simplifying would keep the branch logic easier to read.♻️ Optional simplification
- if hint is None and not admin and verification.state == "OPEN": + if hint is None and verification.state == "OPEN":🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/claude_task_master/cli_commands/fix_pr.py` around lines 113 - 139, Optionally simplify the post-verification branch in the fix_pr command: rely on verify_merged’s normalized str | None base_branch and head_branch results instead of repeating isinstance guards and fallback normalization, and remove the redundant not admin condition when checking a missing merge_failure_hint. Preserve the existing fallback behavior and admin-specific messaging.tests/github/test_client_pr_review_decision.py (1)
39-79: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winUse the configured test-name pattern.
Rename added test methods to
test_<function>_<scenario>_<expected_result>. This makes failures identify the tested operation and expected result.
tests/github/test_client_pr_review_decision.py#L39-L79: include_parse_pr_status_responseand the expected parsed decision in each test name.tests/core/test_stages_review_timeout.py#L110-L204: include the reviewed stage method, timeout condition, and expected transition in each test name.tests/core/test_pr_context.py#L352-L404: rename the test to identifysave_ci_failures, the never-started job condition, and the warning result.tests/core/test_git_branch_cleanup.py#L30-L137: includedelete_merged_branchorcurrent_branch, the branch condition, and the expected safe action.Based on coding guidelines, “One test file per module with descriptive test names following pattern: test_function_scenario_expected_result.”
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/github/test_client_pr_review_decision.py` around lines 39 - 79, Rename the added tests to follow test_<function>_<scenario>_<expected_result>. In tests/github/test_client_pr_review_decision.py lines 39-79, include _parse_pr_status_response and the expected parsed decision; in tests/core/test_stages_review_timeout.py lines 110-204, include the reviewed stage method, timeout condition, and expected transition; in tests/core/test_pr_context.py lines 352-404, identify save_ci_failures, the never-started job condition, and warning result; and in tests/core/test_git_branch_cleanup.py lines 30-137, identify delete_merged_branch or current_branch, the branch condition, and expected safe action.Source: Coding guidelines
src/claude_task_master/github/ci_infra.py (1)
36-51: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick winConsider narrowing the broadest phrases.
"quota"and"payment"are short substrings. An annotation such as "disk quota exceeded" or "payment gateway unreachable" would match and trip a terminal block.The zero-steps precondition in
external_block_reasonalready keeps most ordinary failures away from this list, so the risk is small. Still, the consequence of a false positive here is a permanent stop rather than a bounded retry, so a slightly more specific phrase is cheap insurance.♻️ Suggested narrowing
"actions is disabled", "actions are disabled", "workflows are disabled", "has been suspended", - "payment", - "quota", + "payment method", + "spending quota", + "usage quota",🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/claude_task_master/github/ci_infra.py` around lines 36 - 51, Narrow the broad matches in _PERMANENT_PHRASES by replacing the generic "quota" and "payment" entries with more specific platform-side phrases that identify account quota or billing/payment problems. Preserve the existing terminal-block matching behavior and all other phrases unchanged.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@CHANGELOG.md`:
- Line 20: Update the changelog entry for `#146` to accurately state that the code
handles reviewDecision by blocking merges when changes are requested, while an
approving review is not required and does not gate merging.
In `@docs/docker.md`:
- Around line 1127-1130: Update the “Agent Credential Not Found” troubleshooting
section to state that a missing CLAUDETM_PROFILE profile causes
ProfileManager.resolve_active to raise ProfileError, not an OAuth fallback;
revise the recovery steps to address correcting or removing the invalid profile,
and reserve OAuth fallback guidance for runs with no selected profile.
- Around line 753-760: Update the standalone Compose example containing the
claudetm-profiles mount to add a top-level volumes declaration for
claudetm-profiles with external enabled, so Docker Compose recognizes it as an
existing external volume.
- Around line 879-892: Update the claudetm service startup configuration around
CLAUDETM_PASSWORD and the claudetm_password secret so the mounted secret at
/run/secrets/claudetm_password populates and exports CLAUDETM_PASSWORD when no
deployer-provided password is set, before claudetm-server starts. Preserve an
explicitly configured CLAUDETM_PASSWORD and ensure the existing secret remains
available to the service.
- Around line 997-1000: Update the pod securityContext for the agent-profile
emptyDir volume to grant writable access to claudetm (UID 1000), preferably by
adding fsGroup: 1000; otherwise prepare the volume ownership in a root init
container while keeping application containers non-root.
In `@examples/docker-compose/basic-production.yml`:
- Around line 38-42: Replace the hard-coded external profile volume with the
required CLAUDETM_PROFILES_VOLUME variable in
examples/docker-compose/basic-production.yml:38-42,
production-with-nginx.yml:42-46, production-with-caddy.yml:42-46, and
production-monitoring.yml:36-40; use a distinct development value in
development.yml:39-43. Update all setup and verification commands in
examples/docker-compose/README.md:23-35 to require a unique per-deployment
volume value and reflect the development value.
In `@README.md`:
- Line 728: Update the Docker deployment command’s volume mount to map
~/workspace to /app/project instead of /home/claudetm/workspace, matching the
Dockerfile WORKDIR and ensuring the server operates on the intended checkout.
- Line 727: Replace every host GitHub CLI configuration mount with a dedicated
least-privilege agent credential supplied through the deployment platform’s
secret mechanism. Apply this consistently at README.md:727-727,
docker-compose.yml:58-61, docs/docker.md:49-50, docs/docker.md:74-75,
docs/docker.md:753-760, and all five files under examples/docker-compose/;
remove any instructions requiring ~/.config/gh mounts.
In `@src/claude_task_master/api/models_task.py`:
- Around line 171-176: Update the TaskInitRequest model to configure Pydantic
with ConfigDict(extra="forbid"), ensuring unknown fields such as parallel_tasks
are rejected with a 422 response instead of being discarded. Add a regression
test for the request route that submits parallel_tasks and verifies the 422
validation response.
In `@src/claude_task_master/core/state_models.py`:
- Around line 60-66: Update src/claude_task_master/core/state_models.py:60-66 to
accept legacy options.parallel_tasks during state loading, while keeping
model_dump() output canonical under options.parallel and preserving the default
for states without either key. Add a regression test in
tests/core/test_state.py:77-100 through StateManager.load_state() that loads
parallel_tasks: false and verifies the resulting options.parallel is false.
In `@src/claude_task_master/mcp/tool_handlers_control.py`:
- Line 196: Update the state-loading/schema migration flow associated with the
control tool’s parallel setting to recognize legacy parallel_tasks, map its
value to parallel before unknown fields are discarded, and preserve the existing
default for states without the legacy field. Add a regression test covering an
older state.json containing parallel_tasks and verify resumed behavior uses the
migrated value.
In `@tests/core/test_agent_message.py`:
- Around line 875-878: Update the logger assertion loop in the test to inspect
the raw arguments from log_tool_use and log_tool_result instead of applying
ANSI_RE to repr(call). Assert directly that the expected arguments contain no
ANSI escape sequences, preserving coverage for both call_args_list collections.
- Around line 705-953: Split the oversized test files while preserving all
existing scenarios and assertions: move the worker-coloring helpers and
TestSubagentColoring/TestSubagentColoringDisabled from
tests/core/test_agent_message.py:705-953 into a focused module; move the
color-gating and SubagentPalette scenarios from
tests/core/test_console.py:1195-1406 into focused module(s); and move the
configuration-option scenarios from tests/mcp/test_tools_control.py:401-432 into
a focused module. Ensure each resulting Python file stays within the 500-line
repository limit, or document an approved exception if the
one-test-file-per-module rule prevents the split.
---
Nitpick comments:
In `@src/claude_task_master/cli_commands/fix_pr.py`:
- Around line 113-139: Optionally simplify the post-verification branch in the
fix_pr command: rely on verify_merged’s normalized str | None base_branch and
head_branch results instead of repeating isinstance guards and fallback
normalization, and remove the redundant not admin condition when checking a
missing merge_failure_hint. Preserve the existing fallback behavior and
admin-specific messaging.
In `@src/claude_task_master/cli_commands/merge_finalize.py`:
- Around line 191-201: Expand the checkout_and_pull docstring to Google style by
documenting the branch argument and the boolean return value, while preserving
the existing checkout and pull behavior.
- Around line 46-62: Review POLICY_REFUSAL_MARKERS and remove "not authorized"
unless the merge error handling can reliably distinguish a base-branch policy
refusal from a token-permission failure. Ensure ADMIN_HINT is shown only when
--admin could plausibly override the refusal, while preserving the existing
markers and case-insensitive matching.
In `@src/claude_task_master/cli_commands/pr_resolution.py`:
- Around line 156-166: Update the fallback messaging in resolve_pr_number so
create_pr=true on a default branch reports that PR creation is unavailable for
the current branch instead of suggesting --create-pr again. Preserve the
existing generic no-PR guidance for other cases and retain the current exit
behavior.
In `@src/claude_task_master/github/ci_infra.py`:
- Around line 36-51: Narrow the broad matches in _PERMANENT_PHRASES by replacing
the generic "quota" and "payment" entries with more specific platform-side
phrases that identify account quota or billing/payment problems. Preserve the
existing terminal-block matching behavior and all other phrases unchanged.
In `@tests/cli_commands/test_merge_finalize.py`:
- Around line 113-122: Add direct coverage for merge_finalize.checkout_and_pull:
test that a failed checkout returns False, and that a successful checkout
followed by a failed pull also returns False. Patch merge_finalize.run_git with
the appropriate return values, using the existing process-result helper, while
leaving successful behavior unchanged.
In `@tests/cli_commands/test_pr_resolution.py`:
- Around line 22-36: Add a test case in TestExplicitInput covering
resolve_pr_number with the “#52” input form, and assert it returns 52 without
requiring Git or branch lookup.
In `@tests/core/test_prompts_working.py`:
- Around line 1279-1293: Move the fan-out test section from
test_prompts_working.py into a focused module such as
test_prompts_working_hive.py, including TestParallelOffIsByteIdentical,
TestParallelIsPurelyAdditive, TestFanOutRules, and TestMaxParallelPlumbing.
Preserve all assertions, digest baselines, imports, and shared helpers required
by those tests, and keep each test module under the 500-line limit.
In `@tests/github/test_client_pr_review_decision.py`:
- Around line 39-79: Rename the added tests to follow
test_<function>_<scenario>_<expected_result>. In
tests/github/test_client_pr_review_decision.py lines 39-79, include
_parse_pr_status_response and the expected parsed decision; in
tests/core/test_stages_review_timeout.py lines 110-204, include the reviewed
stage method, timeout condition, and expected transition; in
tests/core/test_pr_context.py lines 352-404, identify save_ci_failures, the
never-started job condition, and warning result; and in
tests/core/test_git_branch_cleanup.py lines 30-137, identify
delete_merged_branch or current_branch, the branch condition, and expected safe
action.
🪄 Autofix
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yml
Review profile: CHILL
Plan: Pro
Run ID: 2ddd5784-25b9-4419-b035-b3c4224c8f09
📒 Files selected for processing (81)
CHANGELOG.mdCLAUDE.mdDockerfileREADME.mddocker-compose.ymldocs/api-reference.mddocs/docker.mddocs/mcp-tools.mdexamples/docker-compose/.env.exampleexamples/docker-compose/README.mdexamples/docker-compose/basic-production.ymlexamples/docker-compose/development.ymlexamples/docker-compose/production-monitoring.ymlexamples/docker-compose/production-with-caddy.ymlexamples/docker-compose/production-with-nginx.ymlsrc/claude_task_master/api/models_task.pysrc/claude_task_master/api/routes_info.pysrc/claude_task_master/api/routes_task.pysrc/claude_task_master/cli_commands/control.pysrc/claude_task_master/cli_commands/fix_pr.pysrc/claude_task_master/cli_commands/merge_finalize.pysrc/claude_task_master/cli_commands/pr_resolution.pysrc/claude_task_master/cli_commands/workflow_start.pysrc/claude_task_master/core/agent.pysrc/claude_task_master/core/agent_message.pysrc/claude_task_master/core/agent_phases.pysrc/claude_task_master/core/console.pysrc/claude_task_master/core/git_branch.pysrc/claude_task_master/core/hive.pysrc/claude_task_master/core/loop_working_hive.pysrc/claude_task_master/core/loop_working_stage.pysrc/claude_task_master/core/pr_context_ci.pysrc/claude_task_master/core/prompts_working.pysrc/claude_task_master/core/prompts_working_hive.pysrc/claude_task_master/core/stages/base.pysrc/claude_task_master/core/stages/ci_stage.pysrc/claude_task_master/core/stages/git_ops.pysrc/claude_task_master/core/stages/merge_stage.pysrc/claude_task_master/core/stages/pr_fix_stage.pysrc/claude_task_master/core/stages/review_stage.pysrc/claude_task_master/core/state_models.pysrc/claude_task_master/core/subagents.pysrc/claude_task_master/core/task_runner.pysrc/claude_task_master/core/task_runner_hive.pysrc/claude_task_master/core/task_runner_session.pysrc/claude_task_master/github/ci_infra.pysrc/claude_task_master/github/ci_logs.pysrc/claude_task_master/github/client_pr_helpers.pysrc/claude_task_master/github/client_pr_models.pysrc/claude_task_master/mcp/server_specs.pysrc/claude_task_master/mcp/tool_handlers_control.pysrc/claude_task_master/mcp/tool_handlers_task.pytests/api/test_routes_task.pytests/cli/test_start_command.pytests/cli_commands/test_control.pytests/cli_commands/test_fix_pr.pytests/cli_commands/test_merge_finalize.pytests/cli_commands/test_merge_pr_cleanup.pytests/cli_commands/test_merge_pr_loop.pytests/cli_commands/test_pr_resolution.pytests/core/test_agent_message.pytests/core/test_console.pytests/core/test_git_branch_cleanup.pytests/core/test_hive.pytests/core/test_hive_loop.pytests/core/test_pr_context.pytests/core/test_prompts_working.pytests/core/test_stages_approval.pytests/core/test_stages_ci_breaker.pytests/core/test_stages_ci_rerun.pytests/core/test_stages_merged_branch_cleanup.pytests/core/test_stages_review_timeout.pytests/core/test_state.pytests/core/test_subagents.pytests/core/test_task_runner.pytests/core/test_workflow_stages.pytests/github/test_ci_infra.pytests/github/test_ci_logs.pytests/github/test_client_pr_review_decision.pytests/mcp/test_tools_control.pytests/mcp/test_tools_task_management.py
💤 Files with no reviewable changes (4)
- src/claude_task_master/core/task_runner_hive.py
- src/claude_task_master/core/stages/git_ops.py
- tests/core/test_hive_loop.py
- src/claude_task_master/core/loop_working_hive.py
Comment on lines
+997
to
+1000
| # Per-pod scratch: the init container recreates the profile on every start, | ||
| # so replicas never contend for one writable volume. | ||
| - name: agent-profile | ||
| emptyDir: {} |
There was a problem hiding this comment.
🩺 Stability & Availability | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- docs/docker.md: deployment example ---'
sed -n '940,1025p' docs/docker.md
printf '%s\n' '--- Dockerfile references ---'
fd -i 'Dockerfile' . --type f | while IFS= read -r file; do
printf '\n### %s\n' "$file"
rg -n -C 4 'claudetm|\.claudetm|USER|profiles\.json' "$file" || true
done
printf '%s\n' '--- Kubernetes-related context in the document ---'
rg -n -C 5 'initContainers|securityContext|fsGroup|emptyDir|agent-profile|profiles\.json|claudetm' docs/docker.mdRepository: developerz-ai/claude-task-master
Length of output: 31316
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- profile implementation ---'
rg -n -C 8 'profiles\.json|profile add|class .*Profile|def .*profile|mkdir|write_text|open\(' \
--glob '*.py' --glob '*.ts' --glob '*.js' --glob '*.tsx' --glob '*.jsx' .
printf '%s\n' '--- image user and Kubernetes manifest structure ---'
python3 - <<'PY'
from pathlib import Path
import re
dockerfile = Path("Dockerfile").read_text()
for pattern in (r"USER\s+\S+", r"useradd[^\n]+", r"chown[^\n]+"):
print(pattern, re.findall(pattern, dockerfile))
text = Path("docs/docker.md").read_text()
start = text.index("### Kubernetes Deployment")
end = text.index("Create the two Secrets", start)
block = text[start:end]
print("\nKubernetes block has securityContext:", "securityContext:" in block)
print("Kubernetes block has runAsUser:", "runAsUser:" in block)
print("Kubernetes block has fsGroup:", "fsGroup:" in block)
print("Kubernetes block has emptyDir:", "emptyDir:" in block)
PYRepository: developerz-ai/claude-task-master
Length of output: 50389
🌐 Web query:
Kubernetes emptyDir default ownership permissions non-root container fsGroup official documentation
💡 Result:
For Kubernetes pods, an emptyDir volume is initially created by the kubelet [1]. By default, this directory is owned by the root user (UID 0) [2]. Regarding ownership and permissions for non-root containers: 1. fsGroup behavior: While the fsGroup field in a Pod's securityContext is designed to help manage volume permissions for non-root users, official documentation states that fsGroupChangePolicy—which controls how Kubernetes manages these permissions—has no effect on ephemeral volume types such as emptyDir [3][4]. Although some earlier or specific environment-dependent tests may suggest fsGroup can influence emptyDir in certain contexts [5][6], the official Kubernetes documentation excludes emptyDir from the fsGroupChangePolicy mechanism [3]. 2. Default permissions: Without additional configuration, the directory is typically created with permissions that may prevent a non-root container from writing to it if the container user does not belong to the root group [2]. 3. Managing emptyDir permissions: To explicitly control the permissions of an emptyDir volume, Kubernetes introduced an optional mode field within the EmptyDirVolumeSource specification [7]. This field allows users to set the permission bits (e.g., 0777) for the directory. If this field is not specified, it defaults to 0777, though this may interact with other security settings [7]. In summary, for non-root containers, relying solely on fsGroup to manage emptyDir ownership is not supported by the official fsGroupChangePolicy mechanism [3][4]. Users requiring specific access for non-root containers should consider the mode field if available in their cluster version, or configure the container environment/entrypoint to handle the directory permissions appropriately [7][2].
Citations:
- 1: https://kubernetes.io/docs/concepts/storage/volumes/
- 2: Volumes are created in container with root ownership and strict permissions kubernetes/kubernetes#2630
- 3: https://kubernetes.io/docs/tasks/configure-pod-container/security-context/
- 4: https://safeguard.sh/resources/blog/k8s-fsgroup
- 5: https://github.com/kubernetes/kubernetes/blob/e9b95827/test/e2e/common/storage/empty_dir.go
- 6: kubernetes: securityContext's fsGroup is not honored with emptyDir volume kata-containers/kata-containers#1580
- 7: https://www.kubernetes.dev/resources/keps/5502/
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- candidate profile files ---'
fd -i 'profile|credential|config' . --type f \
| grep -vE '(^|/)(tests|node_modules|\.git)/' \
| head -200
printf '%s\n' '--- exact profiles.json references outside tests ---'
rg -n -C 6 'profiles\.json|profile add|api-key profile' . \
-g '!tests/**' -g '!node_modules/**' -g '!*lock*' \
| head -300
printf '%s\n' '--- relevant package layout ---'
fd -t d -d 3 . | head -150Repository: developerz-ai/claude-task-master
Length of output: 25229
🏁 Script executed:
#!/bin/bash
set -eu
python3 - <<'PY'
from pathlib import Path
import re
dockerfile = Path("Dockerfile").read_text()
doc = Path("docs/docker.md").read_text()
assert re.search(r"USER\s+claudetm\b", dockerfile), "image does not switch to claudetm"
assert re.search(r"useradd\s+--uid\s+1000\s+--gid\s+1000\s+.*claudetm", dockerfile), \
"claudetm UID/GID 1000 not found"
assert "chown -R claudetm:claudetm /home/claudetm /app" in dockerfile, \
"image ownership setup not found"
start = doc.index("### Kubernetes Deployment")
end = doc.index("Create the two Secrets", start)
manifest = doc[start:end]
checks = {
"init container mounts profile path": (
"initContainers:" in manifest
and "mountPath: /home/claudetm/.claudetm" in manifest
),
"application container mounts profile path": (
manifest.count("mountPath: /home/claudetm/.claudetm") == 2
),
"profile volume is emptyDir": "emptyDir: {}" in manifest,
"pod has no fsGroup": "fsGroup:" not in manifest,
"pod has no pod-level runAsUser": "runAsUser:" not in manifest,
}
for name, result in checks.items():
print(f"{name}: {'PASS' if result else 'FAIL'}")
if not all(checks.values()):
raise SystemExit("manifest assumptions did not hold")
print("\nEffective image user: claudetm (UID/GID 1000)")
print("Mounted emptyDir replaces the image directory at /home/claudetm/.claudetm")
print("The init container and application container both use that mounted path")
print("No manifest-level group ownership or writable-volume override is configured")
PYRepository: developerz-ai/claude-task-master
Length of output: 623
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- src/claude_task_master/core/profiles.py outline ---'
ast-grep outline src/claude_task_master/core/profiles.py
printf '%s\n' '--- profile registry writes ---'
rg -n -C 12 'profiles\.json|write_text|mkdir|_save|save' src/claude_task_master/core/profiles.py src/claude_task_master/cli_commands/profile.py
printf '%s\n' '--- relevant profile command flow ---'
rg -n -C 10 'profile_add|add_profile|ProfileStore|Profiles|\.add\(' \
src/claude_task_master/cli_commands/profile.py src/claude_task_master/core/profiles.pyRepository: developerz-ai/claude-task-master
Length of output: 24804
🌐 Web query:
site:kubernetes.io/docs emptyDir fsGroup ownership permissions non-root writable emptyDir
💡 Result:
In Kubernetes, managing ownership and permissions for an emptyDir volume to ensure non-root writability is primarily handled through the Pod's securityContext [1][2]. Key mechanisms and behaviors: 1. fsGroup and Volume Ownership: By default, when you specify an fsGroup in the Pod's securityContext, Kubernetes recursively changes the ownership and permissions of the contents of the volume to match that fsGroup [1]. This ensures that the volume is readable and writable by containers running with that group ID [1]. 2. Limitations with emptyDir: While fsGroup works for many volume types, it is important to note that the fsGroupChangePolicy field, which allows you to control how ownership is changed (e.g., to improve performance), does not take effect on ephemeral volume types like emptyDir [1]. 3. Default Permissions: If you do not rely on fsGroup, an emptyDir volume is created by the Kubelet. The resulting permissions and ownership depend on the Kubelet's operation. If you need specific permissions, you can use the defaultMode field within the EmptyDirVolumeSource to set the octal mode bits for files created in the directory (default is 0644) [3]. However, this field does not affect directory permissions [3]. 4. Non-root Writable Strategy: To ensure a non-root container can write to an emptyDir: - Use the fsGroup field in the Pod's securityContext to set the volume's group ownership [1]. - Ensure the container process runs with a group ID that matches the fsGroup (or is a member of it) [1]. - If using User Namespaces, the UIDs and GIDs specified in the pod spec (like fsGroup or runAsUser) refer to the IDs inside the container, and Kubernetes ensures the volume is mapped correctly for those IDs [4][5]. For scenarios where you need granular control over volume permissions that fsGroup does not provide, you may need to use an init container to change the permissions (e.g., chmod/chown) of the volume path before the main application container starts, as the main container may lack the privileges to modify root-owned directories [1].
Citations:
- 1: https://kubernetes.io/docs/tasks/configure-pod-container/security-context/
- 2: https://kubernetes.io/docs/concepts/workloads/pods/advanced-pod-config/
- 3: https://kubernetes.io/docs/reference/kubernetes-api/core/pod-v1/
- 4: https://kubernetes.io/docs/concepts/workloads/pods/user-namespaces/
- 5: https://kubernetes.io/docs/tasks/configure-pod-container/user-namespaces/
🏁 Script executed:
#!/bin/bash
set -eu
printf '%s\n' '--- profile registry implementation ---'
sed -n '1,220p' src/claude_task_master/core/profiles.py
printf '%s\n' '--- profile registry write paths ---'
rg -n -C 10 'profiles\.json|write_text|mkdir|save|_save' \
src/claude_task_master/core/profiles.py \
src/claude_task_master/cli_commands/profile.pyRepository: developerz-ai/claude-task-master
Length of output: 23762
Make the profile volume writable by claudetm.
emptyDir hides the image-owned directory. ProfileManager.save() must create profiles.json.tmp, but both containers run as non-root claudetm (UID 1000). Add fsGroup: 1000 to the pod securityContext, or prepare the volume in a root init container while keeping the application non-root.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/docker.md` around lines 997 - 1000, Update the pod securityContext for
the agent-profile emptyDir volume to grant writable access to claudetm (UID
1000), preferably by adding fsGroup: 1000; otherwise prepare the volume
ownership in a root init container while keeping application containers
non-root.
Source: Path instructions
Comment on lines
+171
to
176
| parallel: bool | None = Field( | ||
| default=None, | ||
| description=( | ||
| "Whether to run a PR group's remaining tasks as one 'hive' session that fans " | ||
| "disjoint-write-set tasks out to subagents (off by default)" | ||
| "Whether a work session may split its one task across 'hive-worker' subagents " | ||
| "with disjoint write sets (on by default)" | ||
| ), |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
rg -n -C 3 'model_config|ConfigDict|extra\s*=|parallel_tasks|parallel' \
src/claude_task_master/api tests/apiRepository: developerz-ai/claude-task-master
Length of output: 7609
🌐 Web query:
According to the official Pydantic 2.12 documentation, what is the default behavior for unknown fields on BaseModel, and how can a model reject them with validation errors?
💡 Result:
According to the official Pydantic 2.12 documentation, the default behavior for unknown fields—data provided during model initialization that is not explicitly defined in the BaseModel—is to ignore them [1][2]. These extra fields are not stored in the model's instance dictionary [3]. To reject unknown fields and raise a validation error, you must configure the model to forbid them using the extra parameter within model_config [1][2]. This is achieved by setting extra='forbid' in a ConfigDict [2][4]: from pydantic import BaseModel, ConfigDict class Model(BaseModel): x: int model_config = ConfigDict(extra='forbid') When a model is configured this way, any attempt to initialize or validate it with unknown fields will raise a ValidationError [2][4]. Additionally, you can override this behavior for a single validation call without changing the model's permanent configuration by passing the extra argument directly to validation methods, such as model_validate() [1][2]: Model.model_validate({"x": 1, "y": 2}, extra="forbid")
Citations:
- 1: https://pydantic.dev/docs/validation/2.12/concepts/models/
- 2: https://pydantic.dev/docs/validation/2.12/api/pydantic/config/
- 3: https://pydantic.dev/docs/validation/latest/concepts/models/
- 4: https://pydantic.dev/docs/validation/latest/api/pydantic/config/
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- models_task.py structure ---'
ast-grep outline src/claude_task_master/api/models_task.py
printf '%s\n' '--- model definitions and configuration ---'
rg -n -C 5 'class .*BaseModel|model_config|ConfigDict|extra|parallel_tasks|class .*Request' \
src/claude_task_master/api/models_task.py src/claude_task_master/api
printf '%s\n' '--- route request parsing ---'
sed -n '1,145p' src/claude_task_master/api/routes_task.py
printf '%s\n' '--- relevant model sections ---'
sed -n '1,190p' src/claude_task_master/api/models_task.py
sed -n '240,325p' src/claude_task_master/api/models_task.py
printf '%s\n' '--- tests for unknown fields and legacy names ---'
rg -n -C 4 'unknown|extra|422|parallel_tasks|parallel' tests/api
printf '%s\n' '--- standalone Pydantic behavior probe ---'
python3 - <<'PY'
try:
import pydantic
from pydantic import BaseModel
print("pydantic_version=", pydantic.__version__)
class TaskInit(BaseModel):
goal: str
parallel: bool | None = None
for payload in (
{"goal": "x", "parallel_tasks": False},
{"goal": "x", "parallel": False, "parallel_tasks": True},
):
model = TaskInit.model_validate(payload)
print("payload=", payload, "model_dump=", model.model_dump(),
"fields_set=", sorted(model.model_fields_set))
except Exception as exc:
print(type(exc).__name__, str(exc))
PYRepository: developerz-ai/claude-task-master
Length of output: 50388
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- request model and route ---'
sed -n '1,45p' src/claude_task_master/api/models_task.py
sed -n '202,280p' src/claude_task_master/api/models_task.py
sed -n '80,130p' src/claude_task_master/api/routes_task.py
printf '%s\n' '--- default and propagation paths ---'
rg -n -C 5 'TaskInitRequest|TaskOptions\(|parallel\s*=' src/claude_task_master tests/api
printf '%s\n' '--- focused Pydantic probe ---'
python3 - <<'PY'
from pydantic import BaseModel
class TaskInitRequest(BaseModel):
goal: str
parallel: bool | None = None
payload = {"goal": "x", "parallel_tasks": False}
model = TaskInitRequest.model_validate(payload)
print("model_dump:", model.model_dump())
print("parallel:", model.parallel)
print("extra retained:", getattr(model, "parallel_tasks", "<absent>"))
print("default extra policy:", TaskInitRequest.model_config.get("extra"))
PYRepository: developerz-ai/claude-task-master
Length of output: 29398
Reject unknown parallel_tasks request fields.
TaskInitRequest defaults parallel to True, so Pydantic discards "parallel_tasks": false and the route starts parallel execution. Set ConfigDict(extra="forbid") and add a 422 regression test.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/claude_task_master/api/models_task.py` around lines 171 - 176, Update the
TaskInitRequest model to configure Pydantic with ConfigDict(extra="forbid"),
ensuring unknown fields such as parallel_tasks are rejected with a 422 response
instead of being discarded. Add a regression test for the request route that
submits parallel_tasks and verifies the 422 validation response.
Comment on lines
+60
to
+66
| # Let a work session's agent split its ONE task across `hive-worker` | ||
| # subagents when the pieces have disjoint write sets. The task stays the | ||
| # unit of work — one session, one task checked off — so this only ever adds | ||
| # a fan-out brief to the work prompt. On by default: fanning out is the | ||
| # lead's judgement call and zero workers is a legitimate answer, so the flag | ||
| # grants permission rather than mandating parallelism. | ||
| parallel: bool = True |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Find persisted-state loading and any existing schema migrations.
rg -n -C 4 --glob='*.py' \
'parallel_tasks|model_validate|model_validate_json|TaskOptions\(' \
src tests
# Find checked-in state fixtures or documentation that still use the legacy key.
rg -n -C 3 'parallel_tasks' .Repository: developerz-ai/claude-task-master
Length of output: 50389
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- state model ---'
sed -n '1,140p' src/claude_task_master/core/state_models.py
printf '%s\n' '--- state loading methods ---'
rg -n -C 10 --glob='*.py' \
'def (load_state|save_state|initialize)|state_path|json\.load|model_validate|TaskState\(' \
src/claude_task_master/core
printf '%s\n' '--- exact legacy-key references ---'
rg -n --glob='*.py' 'parallel_tasks' src tests || trueRepository: developerz-ai/claude-task-master
Length of output: 50388
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- migration implementation ---'
sed -n '330,410p' src/claude_task_master/core/state_save_load.py
printf '%s\n' '--- state manager composition and exports ---'
sed -n '1,120p' src/claude_task_master/core/state.py
rg -n --glob='*.py' 'class StateManager|StateSaveLoad|TaskState\(\*\*|TaskOptions' \
src/claude_task_master/core/state*.py tests/core/test_state*.py
printf '%s\n' '--- focused legacy-key search ---'
rg -n --glob='*.py' --glob='*.json' --glob='*.md' 'parallel_tasks' . || true
printf '%s\n' '--- read-only source invariant probe ---'
python3 - <<'PY'
from pathlib import Path
import re
models = Path("src/claude_task_master/core/state_models.py").read_text()
loader = Path("src/claude_task_master/core/state_save_load.py").read_text()
migration_empty = bool(re.search(
r"_STATE_MIGRATIONS:\s*dict\[int,\s*Callable\[\[dict\[str,\s*Any\]\],\s*dict\[str,\s*Any\]\]\]\s*=\s*\{\}",
models,
))
migration_before_validation = loader.find("data = self._migrate_state(data)") < loader.find(
"return TaskState(**data)"
)
legacy_key_in_model = "parallel_tasks" in models
print({
"current_migrations_empty": migration_empty,
"migration_before_task_state_validation": migration_before_validation,
"legacy_key_in_task_options_model": legacy_key_in_model,
})
PYRepository: developerz-ai/claude-task-master
Length of output: 17181
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- changelog context ---'
sed -n '1,30p' CHANGELOG.md
printf '%s\n' '--- commit introducing the current field ---'
git log --all --oneline -S'parallel_tasks' -- src tests CHANGELOG.md | head -20
printf '%s\n' '--- recent state-model history ---'
git log --all --oneline -- src/claude_task_master/core/state_models.py | head -12
printf '%s\n' '--- current commit summary and parent diff summary ---'
git show --no-patch --oneline HEAD
git diff --stat HEAD^ HEAD -- src/claude_task_master/core/state_models.py tests/core/test_state.py CHANGELOG.mdRepository: developerz-ai/claude-task-master
Length of output: 8487
🏁 Script executed:
#!/bin/bash
set -euo pipefail
printf '%s\n' '--- previous persisted model ---'
git show HEAD^:src/claude_task_master/core/state_models.py | sed -n '25,85p'
printf '%s\n' '--- previous serialization test ---'
git show HEAD^:tests/core/test_state.py | sed -n '70,110p'
printf '%s\n' '--- current model configuration ---'
rg -n -C 3 'model_config|extra\s*=|parallel\s*:|parallel_tasks' \
src/claude_task_master/core/state_models.py \
tests/core/test_state.py
printf '%s\n' '--- standalone Pydantic behavior probe ---'
python3 - <<'PY'
try:
from pydantic import BaseModel
except ImportError as exc:
print(f"pydantic unavailable: {exc}")
raise SystemExit(0)
class Options(BaseModel):
parallel: bool = True
legacy = Options.model_validate({"parallel_tasks": False})
print({
"parsed_parallel": legacy.parallel,
"serialized": legacy.model_dump(),
"legacy_key_retained": "parallel_tasks" in legacy.model_dump(),
})
PYRepository: developerz-ai/claude-task-master
Length of output: 5088
Migrate legacy parallel_tasks when loading state.
Pre-change state.json files persist options.parallel_tasks. The current migration table does not map this key, so Pydantic ignores it and options.parallel defaults to True. Add a migration or validation-only alias, keep model_dump() canonical as parallel, and add a StateManager.load_state() regression test with parallel_tasks: false.
📍 Affects 2 files
src/claude_task_master/core/state_models.py#L60-L66(this comment)tests/core/test_state.py#L77-L100
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/claude_task_master/core/state_models.py` around lines 60 - 66, Update
src/claude_task_master/core/state_models.py:60-66 to accept legacy
options.parallel_tasks during state loading, while keeping model_dump() output
canonical under options.parallel and preserving the default for states without
either key. Add a regression test in tests/core/test_state.py:77-100 through
StateManager.load_state() that loads parallel_tasks: false and verifies the
resulting options.parallel is false.
| enable_release: bool | None = None, | ||
| enable_verification: bool | None = None, | ||
| parallel_tasks: bool | None = None, | ||
| parallel: bool | None = None, |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Find residual references to the old parallel_tasks option name.
rg -n -C2 'parallel_tasks|--parallel-tasks|no-parallel-tasks'Repository: developerz-ai/claude-task-master
Length of output: 170
🏁 Script executed:
#!/bin/bash
set -e
printf '%s\n' '--- tracked files related to schemas, state, migrations, and documentation ---'
git ls-files | rg -i '(schema|state|migration|docs?|readme|tool|config)' | head -200
printf '%s\n' '--- all tracked references to parallel and related option forms ---'
rg -n -i -C2 'parallel[_-]?tasks|parallel' --glob '!*.lock' --glob '!*.pyc' . | head -300
printf '%s\n' '--- target handler structure ---'
ast-grep outline src/claude_task_master/mcp/tool_handlers_control.py
printf '%s\n' '--- target handler context ---'
sed -n '160,270p' src/claude_task_master/mcp/tool_handlers_control.pyRepository: developerz-ai/claude-task-master
Length of output: 46084
🏁 Script executed:
#!/bin/bash
set -e
printf '%s\n' '--- exact old-name references ---'
git grep -n -i -E 'parallel_tasks|--parallel-tasks|no-parallel-tasks' -- . ':!examples/orchestrator-workers.ipynb' || true
printf '%s\n' '--- live parallel declarations and call sites ---'
git grep -n -E '(^|[^[:alnum:]_])parallel([^[:alnum:]_]|$)' -- 'src/**' 'tests/**' 'docs/**' 'README.md' 'CLAUDE.md' | head -400
printf '%s\n' '--- state and model definitions ---'
rg -n -C3 'class TaskOptions|TaskOptions\(|options|model_dump|model_validate|extra=' src/claude_task_master/core src/claude_task_master/mcp src/claude_task_master/api
printf '%s\n' '--- relevant MCP model and tool registration ---'
sed -n '1,260p' src/claude_task_master/mcp/tool_models.py
rg -n -C5 'update_config|initialize_task|parallel' src/claude_task_master/mcp src/claude_task_master/api src/claude_task_master/cli_commandsRepository: developerz-ai/claude-task-master
Length of output: 50389
🏁 Script executed:
#!/bin/bash
set -e
printf '%s\n' '--- state model ---'
sed -n '1,130p' src/claude_task_master/core/state_models.py
printf '%s\n' '--- state load/save paths ---'
rg -n -C4 'json\.loads|model_validate|State\(|TaskState|load_state|save_state' \
src/claude_task_master/core/state*.py src/claude_task_master/core/state.py
printf '%s\n' '--- API option models ---'
sed -n '140,330p' src/claude_task_master/api/models_task.py
printf '%s\n' '--- MCP schemas and registration ---'
sed -n '100,225p' src/claude_task_master/mcp/server_specs.py
sed -n '270,340p' src/claude_task_master/mcp/tool_handlers_task.py
printf '%s\n' '--- focused tests for old-name behavior or state compatibility ---'
git grep -n -i -E 'legacy|migration|backward|compat|extra|parallel_tasks|parallel-tasks' -- \
'tests/**' 'src/claude_task_master/core/**' 'src/claude_task_master/mcp/**' \
'src/claude_task_master/api/**' | head -250Repository: developerz-ai/claude-task-master
Length of output: 50389
🏁 Script executed:
#!/bin/bash
set -e
printf '%s\n' '--- migration and load implementation ---'
sed -n '300,430p' src/claude_task_master/core/state_save_load.py
printf '%s\n' '--- state option update implementation ---'
sed -n '430,490p' src/claude_task_master/core/state.py
printf '%s\n' '--- focused migration tests ---'
rg -n -C5 '_migrate_state|schema_version|unknown|extra|legacy|migration' \
tests/core tests/cli tests/mcp | head -300
printf '%s\n' '--- compact exact-name inventory by tracked file ---'
git grep -l -i -E 'parallel_tasks|--parallel-tasks|no-parallel-tasks' -- . \
':!examples/orchestrator-workers.ipynb' | sortRepository: developerz-ai/claude-task-master
Length of output: 31296
🏁 Script executed:
#!/bin/bash
set -e
python3 - <<'PY'
from pydantic import BaseModel
class TaskOptions(BaseModel):
parallel: bool = True
legacy = TaskOptions.model_validate({"parallel_tasks": False})
print("serialized:", legacy.model_dump())
print("parallel:", legacy.parallel)
print("old_name_retained:", "parallel_tasks" in legacy.model_dump())
PY
printf '%s\n' '--- old-name references outside historical changelog entries ---'
git grep -n -i -E 'parallel_tasks|--parallel-tasks|no-parallel-tasks' -- . \
':!CHANGELOG.md' ':!examples/orchestrator-workers.ipynb' || true
printf '%s\n' '--- migration registry and schema version ---'
git grep -n -C2 -E 'CURRENT_SCHEMA_VERSION|_STATE_MIGRATIONS|parallel_tasks' -- \
src/claude_task_master/core/state_models.py \
src/claude_task_master/core/state_save_load.pyRepository: developerz-ai/claude-task-master
Length of output: 291
Add a migration for legacy parallel_tasks state. parallel_tasks is absent from live code and appears only in historical CHANGELOG.md, but existing state.json files can still contain it. With no schema migration, the field is discarded and parallel defaults to True, which can change behavior when resuming an older run. Handle the legacy field explicitly and add a regression test.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@src/claude_task_master/mcp/tool_handlers_control.py` at line 196, Update the
state-loading/schema migration flow associated with the control tool’s parallel
setting to recognize legacy parallel_tasks, map its value to parallel before
unknown fields are discarded, and preserve the existing default for states
without the legacy field. Add a regression test covering an older state.json
containing parallel_tasks and verify resumed behavior uses the migrated value.
Comment on lines
+705
to
+953
| # ============================================================================= | ||
| # Per-worker coloring | ||
| # ============================================================================= | ||
|
|
||
|
|
||
| class _TtyStdout: | ||
| """stdout stand-in that claims to be a terminal, to force color on.""" | ||
|
|
||
| def isatty(self) -> bool: | ||
| return True | ||
|
|
||
|
|
||
| class _PipedStdout: | ||
| """stdout stand-in for a pipe/redirect — the log-file case.""" | ||
|
|
||
| def isatty(self) -> bool: | ||
| return False | ||
|
|
||
|
|
||
| def _enable_color(monkeypatch) -> None: | ||
| """Make color_enabled() report True for the rest of this test. | ||
|
|
||
| Must be called from inside the test *body*, never from a fixture: pytest's | ||
| capture manager re-installs ``sys.stdout`` between the setup and call | ||
| phases, so a stdout patched during setup is silently replaced and the test | ||
| would quietly measure the uncolored path instead. | ||
| """ | ||
| monkeypatch.delenv("NO_COLOR", raising=False) | ||
| monkeypatch.setattr("sys.stdout", _TtyStdout()) | ||
|
|
||
|
|
||
| def _spawn(block_id: str, subagent_type: str = "hive-worker") -> MagicMock: | ||
| """Build the lead's own Agent tool call that spawns a worker.""" | ||
| msg = _make_assistant_message( | ||
| [_make_tool_use_block("Agent", {"subagent_type": subagent_type}, block_id=block_id)] | ||
| ) | ||
| msg.parent_tool_use_id = None | ||
| return msg | ||
|
|
||
|
|
||
| class TestSubagentColoring: | ||
| """Concurrent workers are separable from each other, not just from the lead. | ||
|
|
||
| Color is keyed to the worker's ``tool_use_id`` — stable for its whole life — | ||
| because the several ``hive-worker``s a lead runs share one agent name, so | ||
| coloring by name would repaint them all identically. | ||
| """ | ||
|
|
||
| def test_two_concurrent_workers_are_colored_differently(self, monkeypatch): | ||
| """Two live hive-workers must not render as one undifferentiated stream.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_spawn("toolu_b"), "") | ||
| proc.process_message( | ||
| _make_subagent_message([_make_text_block("A works")], "toolu_a"), "" | ||
| ) | ||
| first = mock_console.claude_text.call_args[0][0] | ||
| proc.process_message( | ||
| _make_subagent_message([_make_text_block("B works")], "toolu_b"), "" | ||
| ) | ||
| second = mock_console.claude_text.call_args[0][0] | ||
|
|
||
| assert ANSI_RE.findall(first) != ANSI_RE.findall(second) | ||
|
|
||
| def test_two_concurrent_workers_are_labelled_differently(self): | ||
| """Same agent name, different instance → a visible discriminator.""" | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_spawn("toolu_b"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("a")], "toolu_a"), "") | ||
| first = mock_console.claude_text.call_args[0][0] | ||
| proc.process_message(_make_subagent_message([_make_text_block("b")], "toolu_b"), "") | ||
| second = mock_console.claude_text.call_args[0][0] | ||
|
|
||
| assert "hive-worker#1" in first | ||
| assert "hive-worker#2" in second | ||
|
|
||
| def test_ordinals_follow_spawn_order_not_speaking_order(self, monkeypatch): | ||
| """#n counts workers as the reader watched them spawn.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_first"), "") | ||
| proc.process_message(_spawn("toolu_second"), "") | ||
| # The second-spawned worker happens to speak first. | ||
| proc.process_message( | ||
| _make_subagent_message([_make_text_block("x")], "toolu_second"), "" | ||
| ) | ||
|
|
||
| assert "hive-worker#2" in mock_console.claude_text.call_args[0][0] | ||
|
|
||
| def test_same_worker_keeps_its_color_across_messages(self, monkeypatch): | ||
| """A color that changes mid-run is worse than no color at all.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_spawn("toolu_b"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("one")], "toolu_a"), "") | ||
| first = mock_console.claude_text.call_args[0][0] | ||
| proc.process_message(_make_subagent_message([_make_text_block("two")], "toolu_b"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("three")], "toolu_a"), "") | ||
| third = mock_console.claude_text.call_args[0][0] | ||
|
|
||
| assert ANSI_RE.findall(first) == ANSI_RE.findall(third) | ||
| assert "hive-worker#1" in third | ||
|
|
||
| def test_tool_lines_share_the_workers_color(self, monkeypatch): | ||
| """Tool activity is colored like the worker's text, not separately.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("hi")], "toolu_a"), "") | ||
| text_line = mock_console.claude_text.call_args[0][0] | ||
| proc.process_message( | ||
| _make_subagent_message( | ||
| [_make_tool_use_block("Read", {"file_path": "/x"})], "toolu_a" | ||
| ), | ||
| "", | ||
| ) | ||
| tool_line = mock_console.tool.call_args[0][0] | ||
|
|
||
| assert ANSI_RE.findall(text_line) == ANSI_RE.findall(tool_line) | ||
|
|
||
| def test_lead_output_stays_plain(self, monkeypatch): | ||
| """The lead's own line is byte-identical to before: no marker, no color.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| msg = _make_assistant_message([_make_text_block("mine")]) | ||
| msg.parent_tool_use_id = None | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(msg, "") | ||
|
|
||
| assert mock_console.claude_text.call_args[0][0] == "mine" | ||
|
|
||
| def test_subagent_text_still_not_accumulated_when_colored(self, monkeypatch): | ||
| """Coloring is cosmetic: the isolation of result_text is untouched.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console"): | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| out = proc.process_message( | ||
| _make_subagent_message([_make_text_block("TASK COMPLETE")], "toolu_a"), | ||
| "lead ", | ||
| ) | ||
|
|
||
| assert out == "lead " | ||
| assert "TASK COMPLETE" not in out | ||
|
|
||
| def test_logger_receives_no_escape_sequences(self, monkeypatch): | ||
| """Log files must stay free of ANSI — the marker is console-only.""" | ||
| _enable_color(monkeypatch) | ||
| mock_logger = MagicMock() | ||
| proc = MessageProcessor(logger=mock_logger) | ||
| with patch("claude_task_master.core.agent_message.console"): | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message( | ||
| _make_subagent_message( | ||
| [_make_tool_use_block("Read", {"file_path": "/foo.py"})], "toolu_a" | ||
| ), | ||
| "", | ||
| ) | ||
| proc.process_message( | ||
| _make_subagent_message([_make_tool_result_block("tu_1")], "toolu_a"), "" | ||
| ) | ||
|
|
||
| for call in ( | ||
| mock_logger.log_tool_use.call_args_list + mock_logger.log_tool_result.call_args_list | ||
| ): | ||
| assert ANSI_RE.search(repr(call)) is None | ||
|
|
||
| def test_logger_calls_are_unchanged(self, monkeypatch): | ||
| """The exact log_tool_use/log_tool_result payloads are as before.""" | ||
| _enable_color(monkeypatch) | ||
| mock_logger = MagicMock() | ||
| proc = MessageProcessor(logger=mock_logger) | ||
| with patch("claude_task_master.core.agent_message.console"): | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message( | ||
| _make_subagent_message( | ||
| [_make_tool_use_block("Read", {"file_path": "/foo.py"})], "toolu_a" | ||
| ), | ||
| "", | ||
| ) | ||
|
|
||
| mock_logger.log_tool_use.assert_called_with("Read", {"file_path": "/foo.py"}) | ||
|
|
||
| def test_reset_clears_color_assignments(self, monkeypatch): | ||
| """Ordinals restart per query instead of growing across a long run.""" | ||
| _enable_color(monkeypatch) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_spawn("toolu_b"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("b")], "toolu_b"), "") | ||
| assert "hive-worker#2" in mock_console.claude_text.call_args[0][0] | ||
|
|
||
| proc.reset_result_state() | ||
| proc.process_message(_spawn("toolu_c"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("c")], "toolu_c"), "") | ||
| assert "hive-worker#1" in mock_console.claude_text.call_args[0][0] | ||
|
|
||
|
|
||
| class TestSubagentColoringDisabled: | ||
| """With color disabled the marker is plain text — logs stay readable.""" | ||
|
|
||
| def test_no_color_env_suppresses_escapes(self, monkeypatch): | ||
| """NO_COLOR wins even on a terminal.""" | ||
| monkeypatch.setenv("NO_COLOR", "1") | ||
| monkeypatch.setattr("sys.stdout", _TtyStdout()) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("hi")], "toolu_a"), "") | ||
|
|
||
| printed = mock_console.claude_text.call_args[0][0] | ||
| assert ANSI_RE.search(printed) is None | ||
| assert printed == "↳ [hive-worker#1] hi" | ||
|
|
||
| def test_redirected_output_has_no_escapes(self, monkeypatch): | ||
| """A non-TTY (pipe, redirect into a log file) gets plain text.""" | ||
| monkeypatch.delenv("NO_COLOR", raising=False) | ||
| monkeypatch.setattr("sys.stdout", _PipedStdout()) | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("hi")], "toolu_a"), "") | ||
|
|
||
| assert ANSI_RE.search(mock_console.claude_text.call_args[0][0]) is None | ||
|
|
||
| def test_workers_still_labelled_distinctly_without_color(self, monkeypatch): | ||
| """Without color the discriminator alone keeps workers apart.""" | ||
| monkeypatch.setenv("NO_COLOR", "1") | ||
| proc = MessageProcessor() | ||
| with patch("claude_task_master.core.agent_message.console") as mock_console: | ||
| proc.process_message(_spawn("toolu_a"), "") | ||
| proc.process_message(_spawn("toolu_b"), "") | ||
| proc.process_message(_make_subagent_message([_make_text_block("a")], "toolu_a"), "") | ||
| first = mock_console.claude_text.call_args[0][0] | ||
| proc.process_message(_make_subagent_message([_make_text_block("b")], "toolu_b"), "") | ||
| second = mock_console.claude_text.call_args[0][0] | ||
|
|
||
| assert first != second | ||
| assert "hive-worker#1" in first | ||
| assert "hive-worker#2" in second |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟠 Major | 🏗️ Heavy lift
Split the oversized test modules.
The repository rule limits every Python file to 500 lines. These modules exceed that limit after this change. Move focused scenarios into smaller test modules, or document an approved exception for the conflicting one-test-file-per-module rule.
tests/core/test_agent_message.py#L705-L953: move the worker-coloring scenarios and helpers into a focused module.tests/core/test_console.py#L1195-L1406: move color-gating andSubagentPalettescenarios into focused modules.tests/mcp/test_tools_control.py#L401-L432: move configuration-option scenarios into a focused module.
📍 Affects 3 files
tests/core/test_agent_message.py#L705-L953(this comment)tests/core/test_console.py#L1195-L1406tests/mcp/test_tools_control.py#L401-L432
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@tests/core/test_agent_message.py` around lines 705 - 953, Split the oversized
test files while preserving all existing scenarios and assertions: move the
worker-coloring helpers and TestSubagentColoring/TestSubagentColoringDisabled
from tests/core/test_agent_message.py:705-953 into a focused module; move the
color-gating and SubagentPalette scenarios from
tests/core/test_console.py:1195-1406 into focused module(s); and move the
configuration-option scenarios from tests/mcp/test_tools_control.py:401-432 into
a focused module. Ensure each resulting Python file stays within the 500-line
repository limit, or document an approved exception if the
one-test-file-per-module rule prevents the split.
Source: Coding guidelines
Seven of thirteen findings fixed, six skipped with reasons (posted on the PR). Real bug found: pydantic silently dropped the removed `parallel_tasks` key, and `parallel` now defaults to True — so a REST client asking for no parallelism would silently get it. `_RemovedFieldGuard` now 422s with the replacement name instead. Targeted rather than `extra="forbid"`, because the README documents a `project_dir` key on /task/init that no model declares. Also: a vacuous ANSI-leak test (`repr()` renders ESC as the four characters `\\x1b`, so the regex could never match) now walks raw logger args; the missing-profile troubleshooting said OAuth fallback where `resolve_active` actually raises `ProfileNotFoundError`; the Swarm example populated `CLAUDETM_PASSWORD` from a secret it never read, leaving auth middleware uninstalled on a 0.0.0.0 bind; an undeclared external volume; a shared credential volume across all five compose examples; and the CHANGELOG's own claim that no `reviewDecision` handling exists. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Collaborator
Author
|
Thanks — all 13 findings verified against the tree. 7 fixed, 6 skipped, pushed in One was a real bug worth calling out: pydantic silently drops the removed Fixed
Skipped, with reasons
Gate after the fixes: 5961 passed, 3 skipped (was 5957 — 4 new tests, none weakened) · ruff ✓ · |
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
1 participant
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.
Two things at once: the hive redesign the previous PR got wrong, and every open issue in the tracker.
The unit of parallelism is one task, not several
v0.1.83 shipped hive mode as batching a PR group's remaining tasks into one lead session. That was the wrong unit — the ask was "parallelize each task". So the batching is deleted, not deprecated:
HiveBatch,plan_hive_batch, theTASKS COMPLETE:manifest, the multi-checkoff and the batch index math are gone;core/loop_working_hive.pyandcore/task_runner_hive.pyare removed;core/hive.pygoes 275 → 52 LOC; andloop_working_stage.pyis byte-identical to its pre-#155 self.What replaces it is smaller. The loop is what it always was — one session per task, one task checked off,
TASK COMPLETE. The agent running that session is now a lead: it may cut its own single task into pieces with disjoint write sets and hand each to ahive-workersubagent, doing everything that overlaps itself. The orchestrator cannot tell a fanned-out session from a solo one except by what lands in the tree, so there is no new bookkeeping to get wrong.Proof the restore is exact: the five SHA-256 prompt digests from before #155 were kept unchanged and pass against
parallel=False.On by default.
TaskOptions.parallel = True;--parallel/--no-parallelreplaces--parallel-tasks— deleted, not aliased (breaking, noted in the changelog).CLAUDETM_HIVE_MAX_PARALLELis a ceiling, not a target: the lead sizes its own team, and zero or one worker are both legitimate answers. One shared checkout, never worktrees — the file set is the only lock, and the lead alone runs git.Coloured subagent output. Each worker gets a colour keyed on its
tool_use_id, so it keeps that colour for life and two concurrenthive-workers are separable at a glance (↳ [hive-worker#1]). Palette is disjoint from the status colours so a name never reads as success/error.NO_COLORand non-TTY suppress it; log files never receive escapes.Every open issue
merge-prdeleted the checked-out branch instead of the merged PR's — in the wild it destroyed an unrelated open PR's branch, and ran even though the merge had failed. Both call sites (CLI and orchestrator) now share one policy incore/git_branch.py: the PR's head branch only, only after a merge confirmed against GitHub, never the base,-dfirst with-Donly when the remote ref exists and nothing is unpushed._GitOps._delete_local_branch(unconditional-D) deleted.merge-prprinted "merged successfully" and exited 0 when the merge did not happen — verified on two PRs leftOPENwithmergedAt=null. The exit code is now authoritative, the merge is verified against the repository before anything is deleted, and a branch-policy refusal names--admin.--adminforce-advances. A tolerated failure can no longer cause the wait or the block.src/. Prose corrected at 13 sites, andreviewDecisionis now actually read:CHANGES_REQUESTEDblocks an auto-merge. Deliberately even under--admin—--adminis passed on essentially every run here, so honouring it would delete the gate; it cannot deadlock, because a human clears it on GitHub and the next cycle proceeds.GitHubErrorevery cycle. It is now a classified signal (CILogsUnavailableError,CIJob.never_started), and a failure GitHub has declared permanent — billing lock, Actions disabled — trips a breaker that posts one explanatory PR comment (idempotent via a marker) and stops cleanly. Partly stale: on currentmainthat case already spent zero agent sessions; the real gaps were two wasted re-runs and a silent block.~/.claude, so the agent ran as — and billed — whoever owned that token. A real cross-developer leak. Every mount is gone fromDockerfile,docker-compose.yml, all five compose examples, the examples README/.env.example, the Swarm stack and the Kubernetes manifest. Auth is a scoped, rotatable api-key profile in a named volume, selected byCLAUDETM_PROFILE.Found on the way (not in any issue)
core/importing fromcli_commands/. The shared branch policy now lives incore/git_branch.pyand the CLI imports down into it, like every other entry layer.${CLAUDETM_PASSWORD:?Error: ...}unquoted makes YAML read the list item as a map. Fixed and verified withdocker compose config.ANTHROPIC_API_KEYfailsCredentialManager.get_valid_token(), which short-circuits only forprofile.type == "api-key". The manifest uses an initContainer + profile, and the doc says why so nobody "simplifies" it back.Deliberately not done
core/stages/review_stage.pystill importscli_commands.ci_helpers— the same inversion, but pre-existing and untouched by this work.Gate
pytest5957 passed, 3 skipped ·ruff check .✓ ·ruff format .✓ ·mypy .✓ 421 files ·claudetm doctor✓ ·docker compose config✓ on all six compose files ·verify_docs_links291/291Built by nine path-disjoint agents in one checkout. Every slice's tests were written failure-case-first and verified to fail before the change; the
merge-prand branch-cleanup regressions run end-to-end against a real temp git repo.Closes #153, closes #152, closes #147, closes #146, closes #116, closes #100
🤖 Generated with Claude Code
Need help on this PR? Tag
@codesmith-botwith what you need. Autofix is disabled.Summary by CodeRabbit
New Features
--parallel, with--no-parallelavailable to disable it.merge-pr --create-pr.Bug Fixes
CHANGES_REQUESTED.Documentation