feat(stories): spec-to-loop contract layer (Phase 1) - #65
Merged
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (4)
🚧 Files skipped from review as they are similar to previous changes (3)
WalkthroughThis PR adds stories-mode support: plan-halt synthesis handling, a strict ChangesStories Mode Feature
Estimated code review effort: 3 (Moderate) | ~30 minutes Poem
🚥 Pre-merge checks | ✅ 3 | ❌ 2❌ Failed checks (2 warnings)
✅ Passed checks (3 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
🤖 Augment PR SummarySummary: Adds an opt-in “stories mode” contract layer (Phase 1) for reading/validating Changes:
Tests: Adds extensive coverage for stories parsing/scheduling/spec resolution, the plan-halt terminal behavior, and stories-mode verification paths. 🤖 Was this summary useful? React with 👍 or 👎 |
![augmentcode[bot]](https://cdn.github.com/in/1027498?s=60&v=4){kind=small}
![coderabbitai[bot]](https://cdn.github.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
🧹 Nitpick comments (3)
src/bmad_loop/verify.py (2)
1113-1113: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueUndocumented deferred import.
from . import storiesis placed inside the function body with no comment. This is presumably to avoid a circular import (stories.py importsread_frontmatter/status_offrom this module at the top level). Worth a one-line comment so a future refactor doesn't hoist it to module scope and reintroduce the cycle.🤖 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/bmad_loop/verify.py` at line 1113, Add a brief one-line comment above the deferred import inside the verify flow to explain that `from . import stories` is intentionally kept local to avoid a circular dependency with `stories.py` importing `read_frontmatter` and `status_of` from this module. Keep the import in the function body and make the intent explicit so future refactors don’t move it to module scope and reintroduce the cycle.
1094-1169: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winDuplicate gate logic between
verify_devandverify_dev_stories.Workflow-tag, expected-status, baseline-match, and proof-of-work checks (Lines 1137-1166) are copy-pasted from
verify_dev(Lines 991-1020). Consider factoring the shared post-resolution gates into a helper both functions call, taking the resolvedspec_pathas a parameter, so the sprint-mode and stories-mode gates can't silently drift apart.♻️ Sketch of a shared helper
def _verify_common_gates( spec_path: Path, rj: dict, task: StoryTask, paths: ProjectPaths, review_enabled: bool ) -> VerifyOutcome | None: """Shared workflow/status/baseline/proof-of-work gates. Returns a failing VerifyOutcome, or None when all gates pass (caller then sets task.spec_file).""" workflow = rj.get("workflow") if workflow != DEV_WORKFLOW: return VerifyOutcome.retry(...) expected = "in-review" if review_enabled else "done" fm = read_frontmatter(spec_path) status = status_of(fm) if status != expected: return VerifyOutcome.retry(...) ... return None🤖 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/bmad_loop/verify.py` around lines 1094 - 1169, The post-resolution verification logic is duplicated between verify_dev_stories and verify_dev, which risks drift in the workflow/status/baseline/proof-of-work gates. Extract the shared checks into a helper such as _verify_common_gates that takes spec_path, rj, task, paths, and review_enabled, performs the common gates, and returns either a failing VerifyOutcome or None. Then have verify_dev_stories call that helper after stories.resolve_story_spec and only set task.spec_file / pass when it succeeds, keeping the stories-specific resolution and id-prefix validation in verify_dev_stories.tests/test_stories.py (1)
49-49: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueUnescaped regex metacharacters in
pytest.raises(match=...).Ruff RUF043: these
match=strings contain a literal.(in "stories.yaml") that is a regex metacharacter but isn't escaped. It happens to still match correctly, butre.escape()makes intent explicit and avoids future accidental collisions.🔧 Suggested fix
- with pytest.raises(stories.StoriesError, match="no stories.yaml found"): + with pytest.raises(stories.StoriesError, match=re.escape("no stories.yaml found")):(apply similarly at lines 214 and 310)
Also applies to: 214-214, 310-310
🤖 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/test_stories.py` at line 49, The pytest.raises(match=...) assertions in stories tests use literal error text containing regex metacharacters, so the match strings should be escaped to satisfy Ruff RUF043. Update the match arguments in the relevant tests in stories-related functions to use a regex-safe form, such as escaping the literal “stories.yaml” text, and apply the same fix to the other matching assertions noted in the review. Keep the expected message content the same while making the regex intent explicit.Source: Linters/SAST tools
🤖 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.
Nitpick comments:
In `@src/bmad_loop/verify.py`:
- Line 1113: Add a brief one-line comment above the deferred import inside the
verify flow to explain that `from . import stories` is intentionally kept local
to avoid a circular dependency with `stories.py` importing `read_frontmatter`
and `status_of` from this module. Keep the import in the function body and make
the intent explicit so future refactors don’t move it to module scope and
reintroduce the cycle.
- Around line 1094-1169: The post-resolution verification logic is duplicated
between verify_dev_stories and verify_dev, which risks drift in the
workflow/status/baseline/proof-of-work gates. Extract the shared checks into a
helper such as _verify_common_gates that takes spec_path, rj, task, paths, and
review_enabled, performs the common gates, and returns either a failing
VerifyOutcome or None. Then have verify_dev_stories call that helper after
stories.resolve_story_spec and only set task.spec_file / pass when it succeeds,
keeping the stories-specific resolution and id-prefix validation in
verify_dev_stories.
In `@tests/test_stories.py`:
- Line 49: The pytest.raises(match=...) assertions in stories tests use literal
error text containing regex metacharacters, so the match strings should be
escaped to satisfy Ruff RUF043. Update the match arguments in the relevant tests
in stories-related functions to use a regex-safe form, such as escaping the
literal “stories.yaml” text, and apply the same fix to the other matching
assertions noted in the review. Keep the expected message content the same while
making the regex intent explicit.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: d0fc846e-8174-423c-8c66-4a7ca85a0f14
📒 Files selected for processing (7)
src/bmad_loop/devcontract.pysrc/bmad_loop/stories.pysrc/bmad_loop/verify.pytests/fixtures/stories.yamltests/test_devcontract.pytests/test_stories.pytests/test_verify.py
Pure contract layer for the spec-to-loop "stories mode" (BMAD-METHOD PR flow. - stories.py: StoryEntry/Stories + load_stories() — a strict typed parser for the flat, linear stories.yaml (str()-normalized ids as int/float coercion defense, charset + prefix-free + no-status + unique validation, independent spec_checkpoint/done_checkpoint bools, verbatim invoke_dev_with; NO depends_on/DAG). Plus resolve_story_spec() (id-keyed disk state) and a linear schedule() distinguishing run-complete from run-wedged (blocked/sentinel stops the scan). - devcontract.synthesize_result: plan_halt seam so a `Halt after planning.` dispatch treats frontmatter status ready-for-dev as success-terminal (with a plan_halt marker), while the default keeps it non-terminal (died-mid-flight). Composes with _reconcile_generic_terminal_status, which only reconciles done-prose specs and so never clobbers this leg. - verify.verify_dev_stories: verify_dev minus the sprint-status gate, with deterministic id-keyed resolution and an id-prefix assertion. - Tests + a stories.yaml fixture from the gist's dogfooded example. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…fest" wording Two contract-layer hardenings surfaced by the post-hoc stories-mode audit: - Reject ids that differ only by case. Story specs resolve by the `<id>-*.md` glob, which is case-insensitive on Windows/macOS filesystems (both in the CI matrix), so `Auth` and `auth` — or `Auth` and `auth-2` — would cross-match the same files and make resolution filesystem-dependent. load_stories now rejects equal-casefold duplicates, and _validate_prefix_free folds case so a case-insensitive prefix (`Auth` vs `auth-2`) is caught too. - The forbidden-'status'-key error said "never in the manifest"; "manifest" is the plugin layer's word for plugin.toml. Use the locked user-facing vocabulary "stories.yaml" instead. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
pbean
force-pushed
the
feat/spec-to-loop-stories-phase1
branch
from
July 7, 2026 03:46
b7e2b97 to
032027a
Compare
There was a problem hiding this comment.
🧹 Nitpick comments (1)
tests/test_stories.py (1)
49-49: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueEscape regex metacharacters in
pytest.raises(match=...).Static analysis flags unescaped
.in match patterns (e.g.,"stories.yaml"), which is technically a wildcard rather than a literal dot. Functionally harmless here, but usere.escape()for precision.🧹 Proposed fix
- with pytest.raises(stories.StoriesError, match="no stories.yaml found"): + with pytest.raises(stories.StoriesError, match=re.escape("no stories.yaml found")):Apply similarly at lines 243 and 339.
Also applies to: 243-243, 339-339
🤖 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/test_stories.py` at line 49, The pytest.raises match strings in the Stories tests use unescaped regex metacharacters, so update the affected assertions to use re.escape() for the literal error text. Apply this in the tests around stories.StoriesError and the other matching checks in the same test module, especially the assertions in the relevant test functions near the current match patterns, so the regex matches the exact message rather than treating punctuation like "." as wildcards.Source: Linters/SAST tools
🤖 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.
Nitpick comments:
In `@tests/test_stories.py`:
- Line 49: The pytest.raises match strings in the Stories tests use unescaped
regex metacharacters, so update the affected assertions to use re.escape() for
the literal error text. Apply this in the tests around stories.StoriesError and
the other matching checks in the same test module, especially the assertions in
the relevant test functions near the current match patterns, so the regex
matches the exact message rather than treating punctuation like "." as
wildcards.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 5934f5ad-251d-443c-91bc-c9de4c97f7b8
📒 Files selected for processing (7)
src/bmad_loop/devcontract.pysrc/bmad_loop/stories.pysrc/bmad_loop/verify.pytests/fixtures/stories.yamltests/test_devcontract.pytests/test_stories.pytests/test_verify.py
🚧 Files skipped from review as they are similar to previous changes (6)
- tests/fixtures/stories.yaml
- src/bmad_loop/devcontract.py
- src/bmad_loop/verify.py
- tests/test_devcontract.py
- tests/test_verify.py
- src/bmad_loop/stories.py
Collaborator
Author
|
@CodeRabbit review |
|
✅ Action performedReview finished.
|
…ories (PR #65 review) Address the augmentcode review on the Phase 1 contract layer: - resolve_story_spec: guard the `<id>-*.md` glob with ID_RE so a non-charset id (glob metachar / path separator) fails clean to PENDING instead of an injected or mis-widened match (medium). - verify_dev_stories: normalize `story_key` via str().strip() before the filename-prefix check, matching the resolver, so a padded key can't spuriously fail resolution (low). - verify_dev_stories: honor the devcontract plan_halt terminal — a `plan_halt` leg expects `ready-for-dev` and skips proof-of-work (a plan writes only its own spec). Keeps Phase 1 consistent with the plan_halt seam it already ships (medium). Nitpicks: document the deferred `from . import stories` (verify<->stories cycle avoidance); re.escape() the literal-dot pytest.raises match patterns. Tests: metachar/path-sep id -> PENDING; whitespace story_key passes; plan_halt accepts ready-for-dev + skips proof-of-work, and rejects a non-plan status. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
🧹 Nitpick comments (1)
src/bmad_loop/verify.py (1)
1135-1139: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winDuplicated literal risks silent drift with
devcontract.PLAN_HALT_STATUS.The comment explains the duplication is intentional to avoid an import cycle, but nothing guards against the two literals diverging later (e.g., if
devcontract's plan-halt status token changes). A cheap cross-module test assertingverify.PLAN_HALT_STATUS == devcontract.PLAN_HALT_STATUSwould catch drift without reintroducing the import cycle.Suggested test addition
def test_plan_halt_status_matches_devcontract(): from bmad_loop import devcontract assert verify.PLAN_HALT_STATUS == devcontract.PLAN_HALT_STATUS🤖 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/bmad_loop/verify.py` around lines 1135 - 1139, Add a cross-module test to guard the duplicated plan-halt status literal from drifting. Update the test suite around verify.PLAN_HALT_STATUS so it imports bmad_loop.devcontract inside the test and asserts verify.PLAN_HALT_STATUS matches devcontract.PLAN_HALT_STATUS, using the existing PLAN_HALT_STATUS symbol to locate the check without changing the no-cycle design.
🤖 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.
Nitpick comments:
In `@src/bmad_loop/verify.py`:
- Around line 1135-1139: Add a cross-module test to guard the duplicated
plan-halt status literal from drifting. Update the test suite around
verify.PLAN_HALT_STATUS so it imports bmad_loop.devcontract inside the test and
asserts verify.PLAN_HALT_STATUS matches devcontract.PLAN_HALT_STATUS, using the
existing PLAN_HALT_STATUS symbol to locate the check without changing the
no-cycle design.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 5b8c4733-9c1e-4b4e-913c-34f80c373f66
📒 Files selected for processing (4)
src/bmad_loop/stories.pysrc/bmad_loop/verify.pytests/test_stories.pytests/test_verify.py
🚧 Files skipped from review as they are similar to previous changes (2)
- tests/test_stories.py
- src/bmad_loop/stories.py
pbean
added a commit
that referenced
this pull request
Jul 7, 2026
… rebase Rebasing phase2 onto the updated phase1 (PR #65 review fixes) kept the source fixes (id-guard, story_key normalization, deferred-import comment) but favored phase2's overlapping test additions, dropping two new-test functions. Re-add them and finish CR-3: - test_resolve_charset_invalid_id_is_pending_not_glob: a metachar/path-sep id resolves to PENDING, never an injected glob match. - test_verify_dev_stories_whitespace_story_key: a padded story_key still resolves and passes (normalized id feeds the filename-prefix check). - re.escape() the phase2-added `no stories.yaml found` match pattern too. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Collaborator
Author
|
Addressed the CodeRabbit nitpicks in 3fcc182:
|
Collaborator
Author
|
@CodeRabbit review |
|
✅ Action performedReview finished.
|
| # caller already handles, matching the module's fail-loud-not-slip rule. | ||
| return StoryState(kind=KIND_PENDING) | ||
| stories_dir = Path(spec_folder) / STORIES_SUBDIR | ||
| matches = sorted(stories_dir.glob(f"{sid}-*.md")) if stories_dir.is_dir() else [] |
There was a problem hiding this comment.
In resolve_story_spec() (src/bmad_loop/stories.py:290), the <id>-*.md glob can match a differently-cased filename on Windows/macOS (case-insensitive FS), which would make story resolution/scheduling OS-dependent. Consider filtering matches to those whose name starts with the exact-case f"{sid}-" before deciding PRESENT/AMBIGUOUS.
Severity: medium
🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.
Collaborator
Author
There was a problem hiding this comment.
Fixed in d7e0267: resolve_story_spec now filters the <id>-*.md glob to names starting with the exact-case <id>- prefix, so resolution is deterministic across filesystems — a case-insensitive FS (Windows CI, macOS) can no longer match a differently-cased file that a case-sensitive FS never would. This also keeps it in step with the exact-case sentinel comparison and verify's id-prefix gate (a wrong-case hit would previously resolve here but then fail those as a spurious retry). Covered by test_resolve_wrong_case_id_is_pending.
| # Generic path always self-finalizes to done (no in-review handoff); the | ||
| # review_enabled arm mirrors verify_dev for symmetry. A plan-halt leg instead | ||
| # expects the ready-for-dev plan gate (the plan is done, not the code). | ||
| if plan_halt: |
There was a problem hiding this comment.
In verify_dev_stories() (src/bmad_loop/verify.py:1205), when plan_halt=True you accept ready-for-dev as a success terminal and skip proof-of-work, but there’s no cross-check that result_json actually carries the plan_halt marker. If a caller accidentally passes plan_halt=True, a died-mid-flight ready-for-dev could be treated as a successful plan-halt leg.
Severity: medium
🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.
Collaborator
Author
There was a problem hiding this comment.
Fixed in d7e0267: a plan_halt=True leg now also requires result.json to carry the plan_halt marker devcontract.synthesize_result emits only on a clean plan-halt (plan_halt and status==ready-for-dev and not escalations). A died-mid-flight ready-for-dev never carries it, so it can no longer be promoted to a "successful plan" by a caller that flips the flag — this mirrors the defensive id-prefix gate just above. Covered by test_verify_dev_stories_plan_halt_requires_marker.
…+ shared verify gates (PR #65 review) Second-round PR #65 review fixes. - resolve_story_spec (stories.py): filter the `<id>-*.md` glob to the exact-case `<id>-` prefix so resolution is deterministic across filesystems. On a case-insensitive FS (Windows CI, macOS) the glob would otherwise match a differently-cased file that a case-sensitive FS never would — and a wrong-case hit would then fail the exact-case sentinel comparison / verify id-prefix gate as a spurious retry. (augment comment 3538118749) Covered by test_resolve_wrong_case_id_is_pending. - verify_dev_stories (verify.py): a plan_halt leg now also requires result.json to carry the `plan_halt` marker devcontract emits on a clean plan-halt, so a died-mid-flight `ready-for-dev` can't be promoted to a "successful plan" by a caller that flips plan_halt=True. Mirrors the defensive id-prefix gate above. (augment comment 3538118753) Covered by test_verify_dev_stories_plan_halt_requires_marker. - Extract _verify_shared_gates: the workflow/status/baseline/proof-of-work gates copy-pasted across verify_dev / verify_dev_bundle / verify_dev_stories are factored into one helper (expected_status + proof_exclude params; proof_exclude=None skips proof-of-work for a plan leg) so sprint- and stories-mode gates can't drift. Behavior-preserving; existing suite is the net. (CodeRabbit nitpick) - Add test_plan_halt_status_matches_devcontract: guards verify.PLAN_HALT_STATUS (a deliberate literal copy, kept to avoid an import cycle) against drifting from devcontract.PLAN_HALT_STATUS. (CodeRabbit nitpick) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
pbean
added a commit
that referenced
this pull request
Jul 7, 2026
… rebase Rebasing phase2 onto the updated phase1 (PR #65 review fixes) kept the source fixes (id-guard, story_key normalization, deferred-import comment) but favored phase2's overlapping test additions, dropping two new-test functions. Re-add them and finish CR-3: - test_resolve_charset_invalid_id_is_pending_not_glob: a metachar/path-sep id resolves to PENDING, never an injected glob match. - test_verify_dev_stories_whitespace_story_key: a padded story_key still resolves and passes (normalized id feeds the filename-prefix check). - re.escape() the phase2-added `no stories.yaml found` match pattern too. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Collaborator
Author
|
Note on the pre-merge Linked Issues / Out of Scope Changes warnings: these are false positives. This PR is Phase 1 of the stories-mode stack (pure |
| task, | ||
| paths, | ||
| expected_status=expected, | ||
| proof_exclude=None if plan_halt else artifact_relpaths(paths), |
There was a problem hiding this comment.
src/bmad_loop/verify.py:1231 — In verify_dev_stories(), the proof-of-work exclusion uses artifact_relpaths(paths), which excludes the entire planning/implementation artifacts trees. This differs from verify_dev/verify_dev_bundle’s narrower exclude and may cause false “no changes since baseline” retries when legitimate work is artifact-only (e.g., deferred-work ledger or other story specs).
Severity: medium
🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.
Collaborator
Author
There was a problem hiding this comment.
Valid — this is the #79 false-negative (KNOWN-BUG-ledger-only-story-false-no-changes) re-surfacing in stories mode: the whole-folder artifact_relpaths exclude swallows a ledger/spec-only diff under implementation_artifacts so it reads as "no changes since baseline".
Fixed in the stacked Phase 2 (PR #77) by d55993f, which swaps artifact_relpaths(paths) → verify_dev_exclude_relpaths(paths, spec_path) + _stories_relpaths(paths.project, spec_folder) (the file-granular post-#79 contract: only the session's own spec + sprint_status, plus the spec folder's stories/ + stories.yaml). Regression test: tests/test_verify.py::test_verify_dev_stories_ledger_only_counts_as_real_work.
Kept in Phase 2 rather than back-ported here because the fix requires _stories_relpaths (a Phase 2 helper) and verify_dev_stories is not yet wired to any engine in this pure contract layer — the exclude breadth has no runtime effect until Phase 2 makes it reachable.
3 tasks
pbean
added a commit
that referenced
this pull request
Jul 8, 2026
… rebase Rebasing phase2 onto the updated phase1 (PR #65 review fixes) kept the source fixes (id-guard, story_key normalization, deferred-import comment) but favored phase2's overlapping test additions, dropping two new-test functions. Re-add them and finish CR-3: - test_resolve_charset_invalid_id_is_pending_not_glob: a metachar/path-sep id resolves to PENDING, never an injected glob match. - test_verify_dev_stories_whitespace_story_key: a padded story_key still resolves and passes (normalized id feeds the filename-prefix check). - re.escape() the phase2-added `no stories.yaml found` match pattern too. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
pbean
added a commit
that referenced
this pull request
Jul 8, 2026
…Phase 2 + 3) (#77) * feat(stories): engine, adapter + HITL checkpoints for folder+id dispatch (Phase 2) Wire "stories mode" (BMAD-METHOD #2549) into the loop as an opt-in engine variant alongside the default sprint-status flow — building on the Phase 1 contract layer. Covers the folder+id engine/adapter plus the human-in-the-loop plan/story checkpoints and sentinel recovery. Engine + adapter: - StoriesEngine(Engine): a thin override layer like SweepEngine. `_pick_next` runs the linear schedule from stories.py (re-validated fresh every pick, within-run skip set mirroring sprint's base_skip, blocked/sentinel/ambiguous → pause for resolve). `_dev_prompt` emits the folder+id dispatch (`/bmad-dev-auto Spec folder: <rel>. Story id: <id>.` + verbatim `invoke_dev_with`). `_post_dev_state_sync` is a no-op (no sprint board), `_verify_dev_artifacts`→verify_dev_stories, and `_verify_review`→ verify_review_stories (drops the sprint-status gate). - Deterministic adapter read-back: GenericDevAdapter resolves the id-keyed story spec via stories.resolve_story_spec when BMAD_LOOP_SPEC_FOLDER is set (new env seam on Engine._run_session), skipping the mtime scan; a relative folder is rebased against spec.cwd for worktree isolation. - [stories] policy (source = sprint-status|stories, spec_folder; no continue_independent) mirroring [review] — dataclass, validation, core.toml schema section, template. RunState pins source + spec_folder so resume/resolve rebuild the right engine without re-reading policy. - Preflight content-probe (install.missing_stories_support): stories mode needs a bmad-dev-auto whose step-01 carries folder+id dispatch; fail loud, not at dispatch time. `run --spec <folder>` forces stories mode; `--dry-run` prints the linear schedule (checkpoints, live on-disk state); `--story <id>` filters. - verify_dev_stories proof-of-work also excludes the spec folder's stories/ + stories.yaml so a spec-only story never reads as implementation work. HITL checkpoints (per-story, independent — a story may set both and pause twice): - spec_checkpoint (two-leg plan-halt): `_plan_halt_leg` reads on-disk state — leg 1 dispatches `Halt after planning.` + BMAD_LOOP_PLAN_HALT (adapter synthesizes ready-for-dev as a `plan_halt` terminal), verify_dev_stories( plan_halt=True) gates the plan (ready-for-dev, no proof-of-work, no build/test via the `_run_verify_commands_after_dev` seam), then StoriesEngine pauses at PAUSE_PLAN_CHECKPOINT. Resume re-drives leg 2 (plain folder+id → implement) via `_resume_after_dev_verify`, keyed off StoryTask.plan_checkpoint_pending; the on-disk status (not a flag) keeps prompt + env in lock-step. - done_checkpoint: after a story commits, `_after_story` pauses at PAUSE_STORY_CHECKPOINT — skipped when the story was the last to dispatch. Fires from both _loop and _finish_inflight, always after worktree integration, so a committed unit is merged before the run stops. - Blocked/sentinel/ambiguous wedge (_pause_wedged) records an ESCALATED task (spec path attached), so `resolve`/rearm_escalation and the resolved re-drive flow through the same machinery as an in-run escalation — no defer-and-continue. - Sentinel recovery in runs.rearm_escalation: a fixed-slug <id>-unresolved.md / <id>-ambiguous.md is preserved under {run_dir}/sentinels/, journaled `sentinel-cleared` with its blocking condition, then deleted so the re-dispatch starts clean (PENDING → re-plan). - New pause consts PAUSE_PLAN_CHECKPOINT / PAUSE_STORY_CHECKPOINT + StoryTask .plan_checkpoint_pending (serialized). Journal events: plan-halt, checkpoint-pause, sentinel-cleared, stories-validated. Base engine seams are no-ops for sprint/sweep; the TUI keeps reusing the CLI resume/resolve paths. Tests: StoriesEngine happy path / scheduling / prompt seams / resume round-trip, plan-checkpoint pause/resume round-trip, story-checkpoint pause incl. skip-if-last, additive spec+done double-pause, sentinel re-arm (preserved copy + journal + clean re-dispatch), adapter id-keyed + plan_halt read-back, [stories] policy matrix, install probe, CLI dry-run/validate, RunState + plan_checkpoint_pending round-trip, verify_dev_stories plan_halt gate, verify_review_stories. Full sandbox E2E matrix is deferred to Phase 4. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(verify): port stories proof-of-work exclusions onto the file-granular #79 contract (T3) verify_dev_stories excluded the whole implementation_artifacts/planning_artifacts folders from its proof-of-work gate (the pre-#79 artifact_relpaths blanket). After (verify_dev_exclude_relpaths: only the session's own spec + the sprint-status ledger), stories mode kept the exact false-negative #79 fixed: a story whose entire authorized scope is ledger/spec reconciliation (e.g. deferred-work.md under implementation_artifacts) always read as "no changes since baseline" and got rolled back/deferred. Swap artifact_relpaths for verify_dev_exclude_relpaths in verify_dev_stories, keeping _stories_relpaths (the spec folder's stories/ subdir + stories.yaml hold only specs and the human-authored manifest, never implementation work). Test: a stories-mode story whose only diff is the deferred-work ledger now passes proof-of-work. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(stories): stop a bare resume leapfrogging a wedged story (MAJOR-A) _compute_schedule built the scheduler skip set from every recorded task (`set(self.state.tasks)`), including the ESCALATED task _pause_wedged persists for a blocked/sentinel/ambiguous story. Because schedule() consults the skip set before classifying a story's on-disk state, a plain `bmad-loop resume` that never resolved the wedge would skip past the blocked story and dispatch the next one onto a tree missing the blocked story's work — violating the linear "a blocked story cannot be leapfrogged" invariant the branch documents. Restrict the skip set to stories actually retired this run — DONE or DEFERRED — matching schedule()'s own documented contract. An unresolved wedge is left out, so resume re-classifies it from disk (still blocked) and re-pauses on the same story; once `resolve` re-arms it to PENDING it re-dispatches normally. Engine test: wedge → bare resume with sessions available → still pauses on the same story, no dev session runs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(stories): a spec_checkpoint story can never commit without a plan review (MAJOR-B) The plan-review pause keyed entirely off ephemeral, result-derived state: the per-leg plan_checkpoint_pending (set from result.json's plan_halt marker) and _plan_halt_leg (which reads the on-disk spec status). Both go stale the moment the plan reaches ready-for-dev on disk, so the checkpoint was silently skipped whenever the pause was not consumed on that same leg: (a) host dies after the plan is written but before the durable session record → resume re-drives on the already-planned spec straight to implementation; (b) leg 1's verify fails non-fixably → the tree resets but the (rollback-kept) plan survives, so attempt 2 is an implement leg; (c) the skill overruns `Halt after planning.` and drives leg 1 to done. Latch a durable obligation, StoryTask.plan_review_owed, at the story's first dispatch — before the session runs, keyed off the entry's spec_checkpoint flag, not the leg's status/result — and persist it so it survives all three. Clear it ONLY when a plan-review pause actually raises. After any dev leg that did not itself pause, if the obligation is still owed, pause before _review_and_commit with a distinct "plan review owed but already implemented" message (reusing PAUSE_PLAN_CHECKPOINT so CLI/TUI resume is unchanged; plan_checkpoint_pending stays unset so resume commits the approved work rather than re-driving). Tests cover all three scenarios. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(stories): withhold story-spec env from injected workflow sessions (MAJOR-C) _extra_session_env exported BMAD_LOOP_SPEC_FOLDER for every dev/review-role session, including injected plugin-workflow sessions (e.g. a TEA pre_commit_gate). The generic adapter treats that env as a signal to short-circuit to id-keyed story-spec synthesis — but at pre_commit_gate the story spec is already `done`, so a gate session that did no work would read `completed:done` and bypass the completion-marker + monotonic stall-nudge contract that the TEA-livelock fix depends on. Stories-mode only; no sprint regression. Thread the session `label` (None for the primary dev/review session, set for an injected workflow) from the _run_session call site into _extra_session_env, and export the story-spec env only for primary sessions. A labeled workflow session then keeps the generic marker contract. Unit test (label withholds the env) + an integration test (a pre_commit_gate workflow session in a stories run carries no BMAD_LOOP_SPEC_FOLDER, while the primary dev session still does). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(runs): gate sentinel-clear on stories mode; snapshot baseline after spec block (MINOR-G) rearm_escalation's fixed-slug sentinel handling (delete + preserve a copy, rather than a status flip) ran for every run source. A *sprint* spec that merely happened to be named `<key>-unresolved.md` / `<key>-ambiguous.md` would be DELETED instead of re-opened — the one shared-path change from the stories work reachable from default sprint mode. Gate _sentinel_condition on `state.source == "stories"`; a sprint spec is now always status-flipped and kept, whatever its name. Also move #78's baseline-advance (HEAD + untracked snapshot) to AFTER the spec block so a just-cleared stories sentinel — an untracked file removed there — is not captured into baseline_untracked as a phantom pre-existing untracked file (the "order snapshot after sentinel-clear" merge note). The advance's own atomic-pair guarantee is unchanged. Sprint regression test: a sentinel-shaped sprint spec is flipped, not deleted; the stories sentinel tests now pin source="stories". Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(stories): shared status-projection helpers + mode-aware status/dry-run The Phase-2/3-scope portion of the original Phase-4 commit (67b75e3), split out so #77 stays pure Phase 2+3 and the TUI surface + docs + E2E matrix ride the stacked Phase-4 branch. These pieces are core infrastructure that Phase-2/3 code (the stories-aware validate + --story preflight + dry-run parity in item 8, the resolve escalation context in item 9, and the engine folder-render in item 10) already depends on, so they belong below those commits on this branch. - stories.py: the disk-derived status projection shared by the CLI and TUI — resolve_spec_folder (folder anchoring), state_label, StoryRow, story_rows — plus the pure, engine/RunState-free read model they build on. - cli.py: `status` is mode-aware (a stories run prints its board via story_rows) and `run --dry-run` is refactored onto the same helper so the two never drift. - tests: test_cli / test_stories / test_stories_engine / test_install coverage for the projection + the mode-aware status/dry-run paths. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(cli): stories-aware validate + selector preflight + dry-run parity Wave-1b item 8 + the CLI parts of item 10 (audit plan your-two-plans-were-delightful-phoenix.md). - `bmad-loop validate` is now mode-aware: in stories mode (`[stories].source == "stories"` or `--spec <folder>`) it skips the sprint-status gate — which a stories-only project fails on — and instead validates stories.yaml + SPEC.md and probes bmad-dev-auto for folder+id dispatch, with remediation text (MAJOR, flagged by 3 audit agents). `validate --spec` forces it; TUI `v` inherits. - `--story` selector preflight: an unknown id fails the run at preflight (exit 1) instead of crashing the scheduler mid-flight (MINOR-E). - dry-run parity (NOTE): render the project-relative folder the engine actually dispatches, and emit a pending spec_checkpoint story's leg-1 `Halt after planning.` + BMAD_LOOP_PLAN_HALT markers. - stories.py: shared relativize_spec_folder / is_plan_halt_leg / recorded_blocking_condition helpers + PLAN_PRODUCED_STATUSES so run, dry-run, validate and resolve all agree on the same strings. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(resolve): stories escalation context + SKILL sentinel guidance Wave-1b item 9 + the sentinel-cleared journal fix (MINOR-5). - resolve.build_context adds a `stories` block in stories mode: the spec folder, the manifest entry (title/description/checkpoints/invoke_dev_with), and — for a sentinel-escalated story — a sentinel indicator with its kind + recorded blocking condition. The resolver now sees the manifest intent and knows a sentinel has no frozen spec to edit (MAJOR-2). Sprint mode is unchanged. - bmad-loop-resolve SKILL.md: document the stories context block and add a "sentinels and the preserved copy" section (resolve the upstream ambiguity; the orchestrator preserves+deletes the sentinel on re-arm). Reseeded to the .claude/.agents working-tree copies. - runs._clear_sentinel journals sentinel-cleared with the recorded blocking condition parsed from `## Auto Run Result`, plus the fixed slug as sentinel_kind — not the slug alone (MINOR-5). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(stories): engine polish — selector/sentinel/max-stories (item 10) Wave-1b item 10 (audit plan your-two-plans-were-delightful-phoenix.md), the engine portion. The doc-note + TUI + e2e-assert parts of item 10 target Phase-4-introduced text/code, so they ride the stacked Phase-4 branch. - rename engine-side StoriesError -> StoriesModeError, distinct from the contract-side stories.StoriesError so the parse vs drive seams never conflate. - an unknown --story id mid-run pauses for resolve instead of crashing the run (MINOR-E); keyed on the selector so rearm/TUI act on it uniformly. - emit sentinel-detected at read-back (pick-time wedge + post-dev verify), carrying the recorded blocking condition (MINOR-6) — no longer only the later stories-wedged/escalation trace. - done_checkpoint skip-if-last honors --max-stories durably from state (MINOR-F): a bounded run no longer leapfrogs past its cap after a checkpoint pause/resume. - _entry_for journals a one-time stories-manifest-unreadable warning per story when the manifest is hand-broken (NOTE-10). - tests: gate+checkpoint additive double-pause (MINOR-4) plus the audit-named test per fix. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(stories): re-add PR #65 review-fix regression tests after phase1 rebase Rebasing phase2 onto the updated phase1 (PR #65 review fixes) kept the source fixes (id-guard, story_key normalization, deferred-import comment) but favored phase2's overlapping test additions, dropping two new-test functions. Re-add them and finish CR-3: - test_resolve_charset_invalid_id_is_pending_not_glob: a metachar/path-sep id resolves to PENDING, never an injected glob match. - test_verify_dev_stories_whitespace_story_key: a padded story_key still resolves and passes (normalized id feeds the filename-prefix check). - re.escape() the phase2-added `no stories.yaml found` match pattern too. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(install): catch UnicodeDecodeError in stories-support probe (C1) read_text(encoding="utf-8") raises UnicodeDecodeError (a ValueError, not an OSError) on a binary/non-UTF-8 step-01 file, so the content probe's except OSError let it escape and crash the whole preflight. Catch it alongside OSError and report the tree as a problem, mirroring the missing-file case. * fix(stories): launch-floor + ambiguous guard in adapter read-back (A1, A2) A1: _stories_result_json read back the id-keyed story spec with no mtime floor, so a spec left terminal by a prior step — notably the dev's `done` spec that a follow-up review session (also a GenericDevAdapter with BMAD_LOOP_SPEC_FOLDER set) re-opens — was misread as THIS session's completion even when the session produced nothing. Require the spec's mtime to be >= handle.launched_ns, the same launch floor devcontract.find_result_artifact applies on the mtime-scan path; stories-mode dev/review read-back now matches sprint mode. A2: a KIND_AMBIGUOUS match (>1 <id>-*.md file) now returns None immediately instead of polling the full grace — waiting can't collapse the anomaly; the next _pick_next re-classifies it into an actionable wedge for resolve. * fix(engine): make --max-stories cap durable across pause/resume (A5) The _loop dispatch gate counted a local `started` that resets to 0 every time the loop is re-entered. With HITL checkpoints, a healthy run now pauses (plan/story checkpoint) and resumes routinely, so after a checkpoint pause the reset counter let the run dispatch past its --max-stories cap. max_stories is already persisted on RunState and re-passed on resume, so only the counter was non-durable. Gate on a new Engine._dispatched_count() = len(state.tasks) — every picked story is recorded before its session runs (the same set _pick_next keys base_skip on), so the task count is the durable dispatch tally that survives resume. StoriesEngine's _max_stories_reached() (the done_checkpoint skip-if-last guard) now consults the same count, so skip-if-last fires exactly when the loop will stop — no drift. Sprint-mode behavior is identical in-run and only gains resume durability. * fix(stories): clear sentinels by recorded verdict, not basename (C2, C3) C2: rearm_escalation identified a sentinel purely from the spec_file basename (<key>-unresolved.md / <key>-ambiguous.md), so a real story spec that happened to match the convention — or a non-sentinel escalation whose spec matched — was DELETED on re-arm. Data loss for a legitimately-named spec. Record the sentinel verdict at detection time instead: StoriesEngine stamps a new durable StoryTask.sentinel_kind in both detection points (_pause_wedged pick-time wedge and _verify_dev_artifacts post-dev read-back). rearm_escalation now clears a sentinel only when task.sentinel_kind is set (stories mode, defensively re-confirming the on-disk name still matches the recorded slug), and clears the field once discharged. A spec the run never classified as a sentinel is status-flipped and kept. C3: test_rearm_non_sentinel_spec_still_flips_status now runs with source="stories" so it actually exercises the stories-mode sentinel branch (it defaulted to sprint-status, which skips that branch entirely). Adds a sentinel-named-but-never- detected regression, a model round-trip, engine detection assertions, and a full detection->rearm E2E. * fix(cli): warn when --epic is passed with stories mode (A4) --epic has no effect in stories mode (StoriesEngine nulls epic_filter — the stories.yaml manifest is a single flat schedule), so `run --spec ... --epic N` silently dropped the flag. Print a one-line stderr note pointing to --story for per-id filtering instead of surprising the caller with an unfiltered run. * fix(stories): catch UnicodeDecodeError in spec/sentinel read paths Non-UTF-8 story specs/sentinels raised UnicodeDecodeError (a ValueError, not OSError), so the existing `except OSError` guards missed it and crashed the escalation-resolution + re-arm flows instead of degrading gracefully. Broaden the guards across the stories-mode reads: - resolve._stories_context: load_stories, resolve_story_spec, and the sentinel blocking-condition read now tolerate a decode failure (best-effort context). - runs._clear_sentinel: wrap the blocking-condition read; still preserve+delete the sentinel so re-arm completes with an empty recorded condition. - stories_engine._journal_sentinel_detected: same defect, unflagged sibling swept in the same pass. Same bug class as the install.py stories-support probe (74f9b35). Addresses the two open augmentcode threads on PR #77. Regression tests per site (invalid UTF-8). * fix(stories): convert non-UTF-8 manifest/spec reads to clean errors load_stories read the manifest and resolve_story_spec read a PRESENT spec's frontmatter with read_text(encoding="utf-8") but neither guarded the UnicodeDecodeError (a ValueError, not caught by their except yaml.YAMLError / except OSError). A binary/non-UTF-8 stories.yaml or story spec therefore crashed stories-mode preflight, --dry-run, status, validate, and the engine loop with a traceback instead of the clean "stories mode: ..." error the mode contracts on — since every call site (cli.py, stories_engine.py) catches only StoriesError. Fix at the two source functions rather than each of the 5 call sites (DRY, keeps load_stories's "raises StoriesError" contract, also covers verify.py's review-path resolve_story_spec): - load_stories: UnicodeDecodeError -> StoriesError ("not valid UTF-8"). - resolve_story_spec: an undecodable PRESENT spec degrades to status="", which _classify treats as wedged (-> pause for resolve, never silent skip) and state_label renders as "present". Same bug class as the 528171f sweep, which missed these two paths. Leaves verify.read_frontmatter (shared with the sprint path) untouched. Addresses the open augmentcode thread on PR #77. Regression tests per site (non-UTF-8 manifest + PRESENT spec + clean-error CLI boundary). * fix(stories): re-pause an in-run escalation on bare resume, never re-derive it from disk _compute_schedule deliberately leaves ESCALATED tasks out of the skip set (MAJOR-A) so a pick-time wedge re-classifies from disk and re-pauses. That is only safe when the escalation's subject IS the disk state. An *in-run* escalation is not: a CRITICAL verify outcome (e.g. a proof-of-work GitError, which fires only after the status gate passed at in-review/done) or a resolved-redrive exhaustion pause escalates with the spec at a resumable or done status. A bare `bmad-loop resume` (the CLI gates only on liveness) then skipped the terminal ESCALATED task in _finish_inflight, re-classified the story from disk, and either re-dispatched it — overwriting the escalated task with a fresh StoryTask, destroying the escalation record and the resolved_redrive guard (a later exhaustion would DEFER the human-resolved work) — or, at done, dispatched the NEXT story past the unresolved escalation, the exact leapfrog MAJOR-A forbids. New _repause_inrun_escalation guard in _pick_next: a task at ESCALATED with attempt > 0 (a session ran — pick-time wedge/unknown-selector tasks stay at 0 and keep their designed fix-by-hand-then-resume lifecycle) re-raises the escalation pause before disk classification, mutating nothing so resolve still has the full record. rearm_escalation resets the task to PENDING, after which the guard no longer matches and the normal re-drive proceeds. Sprint mode is unaffected (its base_skip includes ESCALATED; proceeding past is its designed defer-and-continue). Regression tests: bare resume with an in-run-escalated spec at in-review (re-pauses on itself, task + resolved_redrive survive) and at done (story 2 never dispatched). * fix(stories): sweep the last three UnicodeDecodeError read paths Round 3 of the 528171f/52df1bd bug class (UnicodeDecodeError is a ValueError, so except-OSError guards miss it). Three sites remained: - adapters/generic._stories_result_json: devcontract.synthesize_result re-reads the resolved spec as UTF-8, and 52df1bd's degrade in resolve_story_spec keeps an undecodable spec kind=PRESENT with a path — so the read-back poll crashed on the very state the engine wedges-and-pauses on. Treat the decode failure as no-result-yet: a torn mid-write glimpse is retried by the poll, a genuinely corrupt spec expires the grace result-less and the next _pick_next raises the actionable wedge. (Closes the open augment thread on generic.py.) - verify.read_frontmatter: fixed at the source rather than the flagged verify_review_stories call site — an undecodable file now degrades exactly like unparseable YAML ({} → status "" → clean retry). Covers verify_review_stories AND _verify_shared_gates (verify_dev_stories), and de-crashes the pre-existing sprint/bundle verify paths for free. (Closes the open augment thread on verify.py.) - runs.rearm_escalation non-sentinel flip: set_frontmatter_status / strip_auto_run_result both re-read the spec as UTF-8, so re-arming an undecodable PRESENT-spec escalation aborted `bmad-loop resolve` with a traceback after mutating in-memory state. Convert to an actionable RearmError raised BEFORE save_state — nothing persists, the escalation stays armed, and the message says to fix/replace the file and re-run resolve. Regression tests per site (invalid UTF-8 bytes): read-back returns None, read_frontmatter returns {}, verify_review_stories retries clean, rearm fails clean and stays armed. * fix(stories): make the dry-run --max-stories cap count dispatchable stories, like the run story_rows truncated the RAW manifest list (entries[:max_stories]), counting already-done stories against the cap, while the real run's cap is the durable dispatch count (len(state.tasks)) that only ever counts stories it drives. On a partially-complete manifest the preview was disjoint from reality: [1..4] with 1-2 done and --max-stories 2, dry-run printed stories 1-2 (both done), the run dispatched 3-4. The docstring already claimed run-limit parity. The cap now counts entries whose on-disk state is not done (_classify), so done rows before the cap stay in view as skipped context and the preview stops after the cap's worth of driveable stories. A non-positive cap now previews an empty schedule — the run dispatches nothing — instead of Python's negative-slice dropping entries from the END of the list. Covers cmd_status and run --dry-run via the shared projection; regression test with a partially-done manifest + zero/negative caps. * fix(stories): fire the done_checkpoint even when the manifest goes unreadable after commit _after_story resolved the entry via _entry_for, which conflates "manifest unreadable" with "id absent" (both return None) — so a stories.yaml that became unreadable between the commit and the after-story check silently dropped the human review pause. Its sibling guard _schedule_complete treats the exact same fault the opposite, deliberately conservative way ("not complete" so the checkpoint still fires). Distinguish the two in _after_story: a readable manifest without the id (or without done_checkpoint) still skips, an unreadable one journals stories-manifest-unreadable and pauses. The run cannot proceed past a broken manifest anyway — the next pick halts on it loud — so the only thing a skip could save is the review itself. Regression test: a dev session that corrupts stories.yaml before commit still lands PAUSE_STORY_CHECKPOINT on the committed story. * docs(resolve-skill): distinguish the multi-file ambiguous-id wedge from the ambiguous sentinel SKILL.md documented the single-file <id>-ambiguous.md sentinel but nothing about the OTHER ambiguous state: >1 file matching <id>-*.md wedges the id with no sentinel — no stories.sentinel context block, no spec path, and no auto-preserve-and-delete on re-arm, so a bare re-arm just re-wedges on the same duplicates. The only prose containing "ambiguous" implied the deletable sentinel, inviting the resolver to conflate the two and leave the duplicate in place. New subsection: how to recognize the state (escalation reason + missing sentinel block) and that the cleanup IS the resolution — merge/remove/rename with the human until at most one file matches, then marker + re-arm as usual. Mirrors re-copied to .claude/.agents skill trees (untracked dogfood copies). * test(engine): sprint-mode --max-stories durability across a pause/resume The e56f9db fix rewired the SHARED dispatch gate (base Engine._dispatched_count replaced the resume-reset _loop counter) but was regression-tested only through stories mode. Lock the sprint path in too: cap=2, one story committed before an epic-boundary gate pause, resume dispatches exactly one more — the third ready-for-dev story never dispatches. --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This was referenced Jul 8, 2026
This was referenced Jul 9, 2026
This was referenced Jul 19, 2026
This was referenced Jul 28, 2026
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.
Phase 1 — pure contract layer for "stories mode"
First phase of adopting BMAD-METHOD PR #2549 (
stories.yaml+ folder+id dispatch) as an opt-in stories mode alongside the default sprint mode. This phase is pure: no engine or sprint-mode edits, so there is zero risk to the existing sprint flow. Engine/adapter/HITL wiring is gated on the upstream merge and lands in Phases 2–4.What's here
src/bmad_loop/stories.py(new) — the strict, typed parser the orchestrator readsstories.yamlthrough:StoryEntry/Storiesdataclasses +load_stories(spec_folder). Validates required fields, unique ids, prefix-free ids, nostatuskey, and the id charset^[A-Za-z0-9]+(-[A-Za-z0-9]+)*$. Ids arestr()-normalized before validation (defense against an LLM-authored unquotedid: 1→ int orid: 3.5→ float).spec_checkpoint/done_checkpointare independent bools (defaultfalse);invoke_dev_withis verbatim free text. Nodepends_on/ DAG — the list is strictly linear.resolve_story_spec(spec_folder, id)→ deterministic id-keyed on-disk state (pending / present+status / ambiguous / sentinel).schedule(stories, states, selector)→ linear scan returning the first PENDING-or-resumable entry, skippingdone, and stopping on ablocked/sentinel/ambiguous entry — distinguishing run-complete from run-wedged.devcontract.synthesize_result— aplan_haltexpected-terminal seam. AHalt after planning.dispatch leaves the spec atready-for-dev; withplan_halt=Truethat becomes a success terminal (carrying aplan_haltmarker), while the default keepsready-for-devnon-terminal (died-mid-flight) exactly as today. It composes with_reconcile_generic_terminal_status— that path only reconciles a spec whose prose saysdone, so a plan-haltready-for-dev(no such prose) is never clobbered. Default path is byte-identical.verify.verify_dev_stories— modeled onverify_dev: keeps the spec-exists / workflow / status-expected / baseline / proof-of-work gates, drops the sprint-status gate, resolves the spec deterministically by id, and asserts the resolved filename's id prefix equals the task id.Tests
tests/test_stories.py(new, 47 tests): parse/normalize incl. unquoted-int ids, float/charset/prefix-free/duplicate/status-key rejections, checkpoint bool defaults, linear schedule with resume states + blocked/sentinel/ambiguous stops + selector,resolve_story_spec.tests/test_devcontract.py:ready-for-devexpected-terminal both ways + reconcile-composition guard.tests/test_verify.py:verify_dev_storiesgates incl. the no-sprint-gate differentiator.tests/fixtures/stories.yaml: the gist's dogfooded example (quoted ids, both checkpoint flags,invoke_dev_with, nodepends_on).Full suite green (1472 passed, 1 skipped);
trunk checkclean.🤖 Generated with Claude Code
Summary by CodeRabbit
stories.yaml, deterministic story scheduling, and automatic story-spec discovery with clear retry/wedge behaviors.ready-for-dev, including result marking when plan-halt is enabled.