ace-assign 0.42.4 → 0.55.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 214f383018bb2b9861e03cedf52c8eca4df627d9787357d00998725be36cbe36
-  data.tar.gz: 5f8d577f6f0eacd399c2a317d8dee720f237194af70968b65fcca8a1de676dea
+  metadata.gz: 1f0f554ace326dc469a945fc93f86f81d087b7c4e6197cb59a23b38bd913c821
+  data.tar.gz: faf2665d5a3e0e88bf626ec35951bfac2af109fb465de44215c32aed766e3d54
 SHA512:
-  metadata.gz: 3be4c93c1264e83f808bd0fed4d92780a65e00da59561c428bf3d7a6f65ce63c21398d256113d601e3849ba25dc546c1eb841e6dbf9db21804809c3afac35239
-  data.tar.gz: daab1a233b937ef79fa3d3b406404350b6b34ca0ade254cb7ff96b6f11c810f0c7dc255fe271e8e00cd8c194c6a941330ed8af725e22d49dfcceb0c627ad1ddf
+  metadata.gz: d41c5204d7060d3359e1adcfb91a3a77adfc6fd3c193a531f3128a6e1e08759d835180fae64be3203acf5f8d0bb94905f278d61a731536702769a1f2fade90a9
+  data.tar.gz: 04c7d468bc85e49fccbc9b4250957803fbc7a204878e5fc47a0b804615aae66bcdf057b6cd7ed682aedbb530a204ae6274de2a77a6210f2c506bb8ebe6d2c742

data/.ace-defaults/assign/catalog/composition-rules.yml

@@ -108,11 +108,6 @@ ordering:
     after: create-retro
     note: "Retro should follow all feedback application in review cycles"
 
-  - rule: e2e-before-release
-    before: verify-e2e
-    after: release
-    note: "E2E must pass before bumping version — catch failures early"
-
   - rule: update-docs-after-release
     before: release
     after: update-docs
@@ -123,11 +118,6 @@ ordering:
     after: create-pr
     note: "Updated docs should be visible to PR reviewers"
 
-  - rule: e2e-after-verify
-    before: verify-test-suite
-    after: verify-e2e
-    note: "Unit tests should pass before E2E runs"
-
 pairs:
   - name: review-apply-cycle
     steps: [review-pr, apply-feedback]
@@ -167,11 +157,6 @@ pairs:
     cap: 1
     note: "Re-plan produces 1-2 sub-tasks, then reflect runs once more (no further re-plan)"
 
-  - name: e2e-review-run-pair
-    steps: [verify-e2e]
-    pattern: sequential
-    note: "E2E review (coverage matrix) and targeted run are a single step with two steps"
-
   - name: docs-update-validate-pair
     steps: [update-docs]
     pattern: sequential
@@ -191,9 +176,9 @@ conditional:
     note: "Test suite verification with profiling catches slow tests and regressions before release. In explicit-step mode this remains advisory unless policy elevates it."
 
   - when: "assignment includes work-on-task and touches CLI commands or public API"
-    suggest: [verify-e2e, update-docs, record-demo]
+    suggest: [update-docs, record-demo]
     strength: recommended
-    note: "Public contract changes require E2E verification, doc updates, and demo recording"
+    note: "Public contract changes require doc updates and demo recording"
 
   - when: "task involves auth, security, or data handling"
     suggest: [security-audit]

data/.ace-defaults/assign/catalog/steps/create-pr.step.yml

@@ -1,28 +1,2 @@
 name: create-pr
 skill: as-github-pr-create
-description: Create a pull request for the implemented changes
-intent:
-  phrases:
-    - "create pr"
-    - "create a pr"
-    - "open pr"
-    - "open pull request"
-
-prerequisites:
-  - name: work-on-task
-    strength: required
-    reason: "Must have code changes to create a PR"
-
-produces: [pull-request]
-consumes: [code-changes, commits]
-
-context:
-  default: null
-  reason: "PR creation needs access to git state in main context"
-
-when_to_skip:
-  - "PR already exists for this branch"
-  - "Changes don't warrant a PR (e.g., direct push allowed)"
-
-effort: light
-tags: [git, pr, publishing]

data/.ace-defaults/assign/catalog/steps/create-retro.step.yml

@@ -1,5 +1,5 @@
 name: create-retro
-skill: null
+skill: as-create-retro-internal
 description: Create a retrospective summarizing what was accomplished and lessons learned
 
 prerequisites:

data/.ace-defaults/assign/catalog/steps/mark-task-done.step.yml

@@ -1,7 +1,6 @@
 name: mark-task-done
-skill: null
+skill: as-mark-task-done-internal
 description: Mark task as done in ace-task after successful implementation
-workflow: wfi://task/update
 
 prerequisites:
   - name: work-on-task

data/.ace-defaults/assign/catalog/steps/onboard.step.yml

@@ -1,19 +1,2 @@
 name: onboard
 skill: as-onboard
-description: Load project context and understand the codebase
-
-prerequisites: []
-
-produces: [project-context]
-consumes: []
-
-context:
-  default: null
-  reason: "Onboarding needs access to the main conversation context"
-
-when_to_skip:
-  - "Already onboarded in a previous assignment"
-  - "Working in a context where project is already loaded"
-
-effort: light
-tags: [setup, context-loading]

data/.ace-defaults/assign/catalog/steps/plan-task.step.yml

@@ -1,13 +1,2 @@
 name: plan-task
 skill: as-task-plan
-description: Analyze task requirements and create an implementation plan
-
-produces: [implementation-plan]
-consumes: [project-base-context, task-spec]
-
-when_to_skip:
-  - "Task is trivial and doesn't need a plan"
-  - "Implementation plan already exists"
-
-effort: light
-tags: [planning, analysis]

data/.ace-defaults/assign/catalog/steps/pre-commit-review.step.yml

@@ -1,6 +1,9 @@
 name: pre-commit-review
 skill: null
 description: Run /review agent command (conversation slash command, not bash) when supported; fall back to ace-lint
+migration_state: transitional-exception
+migration_target: internal orchestration skill/workflow
+migration_reason: "Nested review mode and runtime client detection still encoded in step-template YAML."
 
 prerequisites:
   - name: work-on-task

data/.ace-defaults/assign/catalog/steps/reflect-and-refactor.step.yml

@@ -1,5 +1,6 @@
 name: reflect-and-refactor
-skill: null  # No dedicated skill — the driving agent follows the steps directly, using ace-review and ace-git-commit as tools
+skill: as-reflect-and-refactor-internal
+render: step_template
 description: Analyze implementation against ATOM principles and execute targeted refactoring before PR creation
 
 prerequisites:
@@ -29,7 +30,7 @@ steps:
   - name: demo-validate
     optional: true
     description: "Run functional demo proving the feature works end-to-end"
-    skip_when: "verify-test-suite already covers functional validation (E2E tests exist and passed)"
+    skip_when: "verify-test-suite already covers the required deterministic validation for this change"
     on_failure: "Trigger re-plan mechanism instead of proceeding to analysis"
 
   - name: analyze

data/.ace-defaults/assign/catalog/steps/review-pr.step.yml

@@ -1,18 +1,2 @@
 name: review-pr
 skill: as-review-pr
-description: Review code changes for correctness, style, and best practices
-
-prerequisites:
-  - name: create-pr
-    strength: required
-    reason: "Must have a PR to review"
-
-produces: [review-feedback]
-consumes: [pull-request]
-
-when_to_skip:
-  - "No code changes since last review"
-  - "Changes are trivial (typo fix, config update)"
-
-effort: medium
-tags: [review, quality]

data/.ace-defaults/assign/catalog/steps/split-subtree-root.step.yml

@@ -10,9 +10,11 @@ instructions:
     - "If subtree execution fails: use adaptive minimal-safe replay (do not replay whole subtree by default)."
     - "Inject only required recovery steps before retrying failed or affected steps."
   fork:
-    - "Delegate this subtree into forked context:"
+    - "Unscoped driver action: delegate this subtree into forked context:"
     - "- ace-assign fork-run --assignment <assignment-id>@{{parent_number}}"
-    - "Inside the forked agent, continue execution within this subtree scope only."
+    - "Scoped forked agent action: do not call fork-run again for {{parent_number}}."
+    - "If this scoped root is active and no child step is active yet: ace-assign start --assignment <assignment-id>@{{parent_number}}"
+    - "After a child step becomes active, continue execution within this subtree scope only."
     - "Note: fork-run may take 10-30 minutes. If your environment has bash timeout limits, run in background and poll scoped status with stepped backoff (5m -> 3m -> 1m -> every 60s)."
     - "After fork-run completes: read all subtree report files before continuing. The driver is the guard — review reports to verify quality before advancing."
     - "For failed subtrees: recovery typically includes context refresh (onboard + report review) and verify-test-suite before retry."

data/.ace-defaults/assign/catalog/steps/task-load.step.yml

@@ -1,5 +1,5 @@
 name: task-load
-skill: null
+skill: as-task-load-internal
 description: "Load task behavioral spec: run `ace-bundle task://<taskref>`"
 
 prerequisites:

data/.ace-defaults/assign/catalog/steps/verify-test-suite.step.yml

@@ -1,48 +1,21 @@
 name: verify-test-suite
-skill: as-test-verify-suite
+workflow: wfi://assign/verify-test-suite
 render: step_template
-description: Run package test suites with profiling to verify correctness and performance
-intent:
-  phrases:
-    - "run tests"
-    - "run test suite"
-    - "verify tests"
-    - "test changes"
-    - "check tests"
-
-prerequisites:
-  - name: work-on-task
-    strength: recommended
-    reason: "Should have code changes to verify"
-
-produces: [test-results]
-consumes: [code-changes]
-
-context:
-  default: null
-  reason: "Test execution needs access to project environment"
-
-when_to_skip:
-  - "No code changes that could affect tests (documentation-only)"
-  - "Tests were already run and profiled in a previous step"
-
-effort: light
-tags: [testing, verification, performance]
 
 steps:
   - name: run-package-tests
-    description: "Run ace-test --profile 6 for each modified package"
-    tool: "ace-test --profile 6"
-    note: "Profile reveals slow tests; run per-package, not full monorepo"
+    description: "Run ace-test <package> all --profile 6 for each modified package"
+    tool: "ace-test <package> all --profile 6"
+    note: "Profile reveals slow tests; verify modified packages with their explicit full deterministic target first"
 
   - name: check-performance-budgets
     description: "Verify no test exceeds documented thresholds"
-    note: "Atoms <50ms, molecules <100ms, integration <1s, full package <30s. Tests must not make real network calls or unnecessary filesystem I/O."
+    note: "Fast atoms <50ms, fast molecules <100ms, feat tests <1s, full package <30s. Tests must not make real network calls or unnecessary filesystem I/O."
 
   - name: fix-violations
     description: "Stub I/O and mock external calls for tests exceeding budgets"
     conditional: "performance violations detected"
 
   - name: run-suite
-    description: "Run ace-test-suite to verify no cross-package regressions"
-    tool: "ace-test-suite"
+    description: "Run ace-test-suite --target all to verify no cross-package regressions in the full deterministic suite"
+    tool: "ace-test-suite --target all"

data/.ace-defaults/assign/catalog/steps/verify-test.step.yml

@@ -1,6 +1,9 @@
 name: verify-test
 skill: null
 description: Run profile-guided tests for modified packages only; no full-suite run in subtree
+migration_state: transitional-exception
+migration_target: internal orchestration skill/workflow
+migration_reason: "Nested per-package test sequencing and budget checks still encoded in step-template YAML."
 
 prerequisites:
   - name: work-on-task
@@ -23,13 +26,13 @@ tags: [testing, verification, performance]
 
 steps:
   - name: run-package-tests
-    description: "Run ace-test --profile 6 for each modified package in this subtree"
-    tool: "ace-test --profile 6"
-    note: "Execute per-package; do not run full-stack suite here"
+    description: "Run ace-test <package> all --profile 6 for each modified package in this subtree"
+    tool: "ace-test <package> all --profile 6"
+    note: "Execute per-package only; do not run the monorepo suite in this subtree step"
 
   - name: check-performance-budgets
     description: "Verify no test exceeds documented thresholds"
-    note: "Atoms <50ms, molecules <100ms, integration <1s, full package suite <30s. Tests must not make real network calls or unnecessary filesystem I/O."
+    note: "Fast atoms <50ms, fast molecules <100ms, feat tests <1s, full package suite <30s. Tests must not make real network calls or unnecessary filesystem I/O."
 
   - name: fix-violations
     description: "Fix slow or environment-coupled tests when violations are detected"

data/.ace-defaults/assign/catalog/steps/work-on-task.step.yml

@@ -1,19 +1,2 @@
 name: work-on-task
 skill: as-task-work
-description: Implement task changes following project conventions
-intent:
-  phrases:
-    - "work on task"
-    - "implement task"
-    - "task work"
-    - "build feature"
-
-produces: [code-changes, commits]
-consumes: [project-base-context, task-spec]
-
-when_to_skip:
-  - "Task is documentation-only"
-  - "Changes are already implemented"
-
-effort: medium
-tags: [implementation, core-workflow]

data/.ace-defaults/assign/config.yml

@@ -26,6 +26,7 @@ workflow_source_paths:
 execution:
   provider: "claude:opus@yolo"
   timeout: 1800
+  launch_mode: auto # auto | headless | tmux
 
 # Subtree execution behavior
 subtree:

data/.ace-defaults/assign/presets/fix-bug.yml

@@ -34,10 +34,11 @@ steps:
       - "Commit the archive file moves: `ace-git-commit .ace-tasks/ -i 'archive completed task'`"
 
   - name: verify-test-suite
-    skill: as-test-verify-suite
+    workflow: wfi://assign/verify-test-suite
     instructions:
-      - Run the test suite to verify the fix doesn't break anything.
-      - Ensure all tests pass.
+      - Run `ace-test <package> all --profile 6` for each modified package first.
+      - Then run `ace-test-suite --target all` to verify the fix does not break the wider deterministic suite.
+      - Ensure all required checks pass before continuing.
 
   - name: create-pr
     skill: as-github-pr-create

data/.ace-defaults/assign/presets/quick-implement.yml

@@ -36,6 +36,6 @@ steps:
       - Use a descriptive commit message following project conventions.
 
 customization_hints:
-  - "Add verify-test-suite before commit for complex changes"
+  - "Add verify-test-suite before commit for complex changes; that step should run `ace-test <package> all --profile 6` for modified packages and then `ace-test-suite --target all`."
   - "Add lint after work-on-task for style enforcement"
   - "Skip onboard if already onboarded"

data/.ace-defaults/assign/presets/work-on-task.yml

@@ -22,26 +22,13 @@ steps:
 
   - name: verify-test-suite
     number: "012"
-    workflow: wfi://test/verify-suite
+    workflow: wfi://assign/verify-test-suite
     instructions:
       - "Verify test coverage and performance across batch tasks: {{taskrefs}}."
-      - "Focus on packages modified by this batch before any broader regression checks required by the workflow."
+      - "First run `ace-test <package> all --profile 6` for each package modified by this batch."
+      - "Then run `ace-test-suite --target all`."
       - "If the batch is documentation-only or otherwise does not affect runnable code, mark this step skipped with a clear reason."
 
-  - name: verify-e2e
-    number: "015"
-    context: fork
-    instructions:
-      - "Run E2E tests for packages modified across batch tasks: {{taskrefs}}."
-      - "Detect modified packages from the current diff: `git diff origin/main --name-only | grep '^ace-' | cut -d/ -f1 | sort -u`."
-      - "Run `ace-test-e2e <package>` for each modified package that has test/e2e/ scenarios."
-      - "If more than 5 packages were modified with cross-dependencies, run the full suite: `ace-test-e2e-suite`."
-      - "If any scenarios fail, use /as-e2e-fix to diagnose and fix each failing scenario."
-      - "After fixing, rerun the failing scenarios to confirm they pass."
-      - "Repeat fix→rerun cycle up to 3 times if new failures appear."
-      - "Final gate: all previously-failing scenarios must pass before completing this step."
-      - "Skip only when no modified package has test/e2e/ scenarios."
-
   - name: release-minor
     number: "020"
     instructions:
@@ -176,6 +163,9 @@ expansion:
   batch-parent:
     name: batch-tasks
     number: "010"
+    batch_parent: true
+    parallel: false
+    fork_retry_limit: 1
     instructions: |
       Batch container for multiple task implementations.
       This step auto-completes when all child tasks are done.

data/CHANGELOG.md

@@ -6,6 +6,222 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
+## [0.55.0] - 2026-04-23
+
+### Changed
+- Migrated tmux-backed `fork-run` orchestration to consume the shared `ace-tmux` runtime/control surface instead of maintaining a private `ace-assign` tmux runner.
+- Added step-level `fork.mode` plus config-level `execution.launch_mode` so `ace-assign fork-run` can default subtree launch behavior to `auto`, `headless`, or `tmux` while still letting `--launch-mode` override both.
+- Added `ace-assign fork-run --callback` for tmux-backed parent/child agent flows so a fork can capture the origin pane, launch the child subtree, and let the child send one final status sentence back with `ace-tmux send` instead of keeping the parent on a polling loop.
+- Updated the `/as-assign-drive` fork workflow contract to document callback-mode idle waiting, direct `ace-tmux send` callback delivery, and assignment-state recovery when a callback is missing on re-entry.
+
+### Fixed
+- Updated the `ace-tmux` runtime dependency constraint to `~> 0.17` so assignment tmux delegation stays aligned with the new runtime inspection release line.
+
+### Technical
+- Kept assignment queue state as the authoritative fork completion signal while limiting tmux pane capture to launch and failure diagnostics.
+- Persisted fork callback pane metadata in tmux session files, exported it into tmux fork launch environments as `ACE_ASSIGN_CALLBACK_PANE`, and added command/launcher/runner contract coverage for the new non-blocking callback path.
+- Added parser/model/launcher/command regression coverage for launch-mode precedence across CLI, step frontmatter, and assign config defaults.
+
+## [0.53.4] - 2026-04-20
+
+### Technical
+- Tightened `TS-ASSIGN-002` hierarchy E2E contracts so auto-completion uses derived assignment IDs, valid cross-assignment `finish --assignment` syntax, and explicit scoped status artifacts for subtree verification.
+- Clarified `ace-assign finish` usage docs so positional step targeting is reserved for the active assignment while cross-assignment/scoped completion uses `--assignment <id[@step]>`.
+
+## [0.53.3] - 2026-04-18
+
+### Technical
+- Corrected `TS-ASSIGN-001` lifecycle fixture-path guidance so Goal 1 resolves the real sandbox fixture files and no longer records a false create failure before the lifecycle flow completes.
+
+## [0.53.2] - 2026-04-16
+
+### Technical
+- Reworked retained fork-context E2E capture around explicit regular, fork, return, and final checkpoints so lifecycle transitions are proven from stable artifacts instead of generic completion text.
+
+## [0.53.1] - 2026-04-16
+
+### Fixed
+- Hid failed retry history from compact `ace-assign status` pending previews after assignment completion so terminal state no longer mixes `Status: completed` with a misleading `Pending steps:` block.
+
+### Technical
+- Added a status-command regression covering completed assignments with failed retry ancestry and validated the retained lifecycle E2E path against the corrected compact status output.
+
+## [0.53.0] - 2026-04-15
+
+### Changed
+- Rewrote retained assignment E2E runner/verifier contracts to emphasize public command journeys and impact-first state oracles over rigid capture choreography.
+- Added `TS-ASSIGN-003` operations coverage for multi-assignment operator flow and scoped `fork-run` delegation behavior.
+
+### Technical
+- Refreshed hierarchy/lifecycle/fork scenario guidance so debug artifacts remain fallback evidence and CLI-visible status transitions are primary evidence.
+
+## [0.52.0] - 2026-04-15
+
+### Changed
+- Tightened the committed tmux fork demo contract so the recording is aimed at the visible `work -> work-fs -> agent starts working` transition instead of a synthetic transcript.
+
+### Technical
+- Made tmux fork wrappers print the translated skill handoff before launching the interactive agent, keeping the fork pane bootstrap visible and testable.
+
+## [0.51.0] - 2026-04-15
+
+### Changed
+- Reworked tmux fork execution to start a real interactive agent session via `ace-llm --interactive` in the `<current-window>-fs` pane and auto-send the scoped `/as-assign-drive <assignment>@<root>` handoff instead of running the old headless fork-session wrapper in tmux.
+
+### Technical
+- Switched tmux fork completion tracking from pane-process exit to subtree-state polling so interactive agents can stay open while forked assignment progress is still tracked to completion.
+
+## [0.50.3] - 2026-04-15
+
+### Fixed
+- Fixed tmux fork-window lookup by correcting `TmuxForkRunner` result normalization, so explicit-session launches resolve the current window instead of failing under detached-session demo/recording runs.
+
+### Changed
+- Refined the committed `fork-provider` asciinema tape/setup and refreshed the demo artifacts (`.gif`, `.cast`, and recording manifest) after rerunning the verified tmux fork demo against PR-ready behavior.
+
+## [0.50.2] - 2026-04-15
+
+### Fixed
+- Resolved tmux-backed fork window detection when `ACE_TMUX_SESSION` is set without a live `TMUX` variable, so visible fork demos and detached-session launches still resolve the current window correctly.
+
+### Changed
+- Replaced the stale `fork-provider` demo tape with an asciinema/YAML recording that verifies the real `work -> work-fs` tmux fork transition.
+
+## [0.50.1] - 2026-04-15
+
+### Changed
+- Refreshed the tmux-backed fork demo asset so the recorded flow starts in the operator work window and visibly opens the fork staging window during `fork-run`.
+
+## [0.50.0] - 2026-04-15
+
+### Changed
+- Added `ace-assign fork-run --launch-mode auto|headless|tmux`, keeping the existing headless subprocess path while allowing tmux-backed fork execution in the current tmux session.
+- Tmux-backed fork runs now create or reuse a `<current-window>-fs` window, launch forked provider sessions in panes there, and persist tmux session/window/pane metadata alongside existing fork session files.
+
+### Technical
+- Added an internal `ace-assign fork-session` entrypoint plus launcher/test coverage to reuse the existing provider-backed fork session path from tmux panes.
+- Added a recorded tmux fork demo GIF under `docs/demo/` and updated usage docs for the new launch-mode contract.
+
+## [0.49.2] - 2026-04-13
+
+### Fixed
+- Rewrote `TS-ASSIGN-002` around sandbox end-state and runner observations so hierarchy renumbering, auto-completion, and audit coverage no longer depend on synthetic helper artifacts under `results/`.
+
+## [0.49.1] - 2026-04-13
+
+### Fixed
+- Tightened `TS-ASSIGN-002` hierarchy scenario evidence so cascade renumbering is proven from concrete step-file artifacts and display-audit checks read the copied sandbox fixture path instead of a repo-relative path.
+
+## [0.49.0] - 2026-04-13
+
+### Changed
+- Split assignment status from instruction loading: `ace-assign status` now defaults to compact status-only output with `compact`, `progress`, and `full` text modes, while the new `ace-assign step` command prints current/next/exact step instructions.
+- Made `ace-assign start` and `ace-assign finish` concise by pointing operators to `ace-assign step` instead of printing raw instruction blocks inline.
+
+### Technical
+- Updated drive/fork guidance, deterministic command coverage, and fork-context E2E specs to treat `status` as state-only and `step` as the instruction-loading surface.
+
+## [0.48.0] - 2026-04-13
+
+### Changed
+- Removed the shipped `verify-e2e` assignment step from the public catalog, presets, and composition defaults so standard `ace-assign` workflows no longer schedule package E2E verification.
+- Narrowed assignment-level `verify-test-suite` to deterministic verification only: modified packages run `ace-test <package> all --profile 6`, then the monorepo gate runs `ace-test-suite --target all`.
+
+### Technical
+- Added an `ace-assign`-local `wfi://assign/verify-test-suite` workflow and preserved local step runtime bindings during canonical catalog metadata merges so assignment-specific verification contracts are not overwritten by shared skill defaults.
+
+## [0.47.3] - 2026-04-13
+
+### Changed
+- Completed the batch i05 migration follow-through for this package and aligned it with the restarted `fast` / `feat` / `e2e` verification model.
+
+### Technical
+- Included in the coordinated assignment-driven patch release for batch i05 package updates.
+
+
+## [0.47.2] - 2026-04-12
+
+### Fixed
+- Hardened `wfi://assign/drive` so forked subtree completion is driven by scoped `ace-assign status`, with explicit 6-minute polling, detached resume guidance, and immediate parent-queue re-entry after terminal child status.
+
+### Technical
+- Updated the canonical and projected `as-assign-drive` skill contracts plus workflow contract coverage to lock in status-driven fork completion pickup after terminal or session interruption.
+
+## [0.47.1] - 2026-04-12
+
+### Fixed
+- Simplified `TS-ASSIGN-001` to keep only lifecycle and fork-context E2E coverage, moving help, prepare, and no-skip policy assertions down to deterministic fast and feat contracts.
+- Tightened `TS-ASSIGN-002` into a 4-goal hierarchy scenario by moving hierarchy error-path checks to command tests and requiring explicit multi-level auto-completion and audit-trail evidence.
+
+### Technical
+- Added regression coverage for invalid `add --after` references, pending-parent `finish` rejection after child injection, the canonical `work-on-task` preset contract, and the `assign/drive` no-skip workflow contract.
+
+## [0.47.0] - 2026-04-11
+
+### Changed
+- Taught `wfi://assign/drive` to prefer an E2E-specific `fix-e2e` recovery step after `ace-test-e2e` failures so assignment retries route into `/as-e2e-fix` instead of generic test-fix instructions.
+- Updated the shipped work-on-task verification fixture to require `/as-e2e-fix` to bootstrap `wfi://e2e/analyze-failures` automatically when scenario-failure analysis is missing or incomplete.
+
+## [0.46.0] - 2026-04-11
+
+### Changed
+- Migrated package test layout to the `fast` / `feat` / `e2e` model by moving deterministic ATOM and command tests under `test/fast/`, adding deterministic feature-level coverage under `test/feat/`, and keeping scenario assets under `test/e2e/`.
+- Updated package documentation and E2E scenario metadata to reference the new testing contract (`ace-test ace-assign`, `ace-test ace-assign feat`, `ace-test ace-assign all`, `ace-test-e2e ace-assign`) and `test/fast/*` deterministic coverage paths.
+- Tightened E2E goal runner/verifier contracts to reduce brittle artifact-name coupling and keep policy/lifecycle evidence checks aligned with current command outputs.
+
+## [0.45.0] - 2026-04-11
+
+### Changed
+- Updated assignment verification presets, step templates, and runtime child-step instructions so modified packages run `ace-test <package> all --profile 6`, batch suite checks stay on plain `ace-test-suite`, and E2E verification stays package-targeted via `ace-test-e2e <package>`.
+
+### Technical
+- Synced the shipped work-on-task prepare fixture and executor rendering coverage to the new verification contract.
+
+## [0.44.5] - 2026-04-10
+
+### Fixed
+- Updated verify-E2E guidance to support both `test-e2e/scenarios/` and legacy `test/e2e/` paths during migration in assignment presets and runtime executor output.
+- Synced the work-on-task E2E fixture contract with the dual-path migration guidance.
+
+## [0.44.3] - 2026-04-07
+
+### Fixed
+- Added missing `Bash(ace-bundle:*)` permissions to internal helper skills that execute `ace-bundle` workflows (`as-create-retro-internal`, `as-mark-task-done-internal`, `as-reflect-and-refactor-internal`).
+- Restored required canonical skill metadata headers (`# bundle`, `# agent`) for internal helper skills `as-create-retro-internal` and `as-reflect-and-refactor-internal`.
+- Added source-based canonical step-definition fallback in assignment step materialization so custom-named steps with explicit `source` preserve canonical metadata.
+
+### Technical
+- Added regression coverage for custom-named explicit-source step materialization preserving canonical fork metadata.
+
+## [0.44.2] - 2026-04-07
+
+### Fixed
+- Preserved project-level `.ace/assign/catalog/steps/*.step.yml` overrides when canonical skill metadata is present by applying project definitions after canonical merge in assignment step catalog resolution.
+- Restored canonical step metadata in `CatalogLoader.load_all` results for migrated public steps so direct catalog consumers retain description/prerequisite/artifact semantics, while keeping raw-YAML opt-out support via `canonical_steps: false`.
+
+## [0.44.1] - 2026-04-07
+
+### Fixed
+- Preserved canonical step-level `assign.steps` presentation fields (`name` and `description`) when resolving skill-backed step rendering.
+
+### Technical
+- Added required canonical metadata headers (`# bundle`, `# agent`) to internal helper skills `as-task-load-internal` and `as-mark-task-done-internal`.
+- Added regression coverage for step-level description preservation in `SkillAssignSourceResolver`.
+
+## [0.44.0] - 2026-04-07
+
+### Changed
+- Migrated helper-step ownership for `task-load`, `mark-task-done`, `reflect-and-refactor`, and `create-retro` from permanent `skill: null` templates to internal canonical helper skills.
+- Migrated public step metadata ownership for `onboard`, `plan-task`, `work-on-task`, `review-pr`, `create-pr`, and `verify-test-suite` into canonical skill `assign.steps`, with public workflow binding resolved from `skill.execution.workflow` (legacy `assign.source` fallback retained).
+- Normalized runtime execution and materialization paths around canonical step `source` (`skill://...` and explicit `wfi://...`) while preserving legacy `skill`/`workflow` compatibility as migration fallback.
+- Updated shipped assign compose/create/prepare workflow docs to use source-first step contract examples.
+
+### Fixed
+- Restricted public assign-step discovery to canonical skills with `user-invocable: true`, preventing internal helper skills from appearing in public assignment composition inventory.
+
+### Technical
+- Added internal helper workflows and skills under `ace-assign/handbook` to preserve helper execution contracts while keeping them non-discoverable.
+- Marked `pre-commit-review` and `verify-test` helper templates as explicit transitional exceptions with migration metadata.
 
 ## [0.42.4] - 2026-04-05