0.43.0 · core 2026-07-07

release 0.43.0 on GitHub →

Record evidence convention (v1) + glob pattern expansion + Shape Up task tracking + Muxy cross-workflow scope view. No npm publish at any name; install via pi install git:github.com/calionauta/stelow or npx skills add ....

Added

• Record evidence convention (v1, advisory). cali-product-scope-executor SKILL Step 3e-bis ships a ## Record template that scopes fill before close. Template lives in the scope's iteration-state markdown file; the machine-checkable mirror (completed_at, files_count, commands_count, verified, suggested_commit — all snake_case to match the rest of stelow.json) lands in wf.scopes[i].record. Convention only — no enforcement in v1; execution-critique Criterion 6 (Record Evidence) flags missing or unverified records as critique findings (block / warning / minor). Rationale: weakest-true-claim discipline borrowed from Skill-Steward ADR 0023 — without a non-vacuous record, the ✅ is unearned.
• Glob pattern expansion in matchesDeclaredGlob(). extensions/stelow/scope.ts now supports:
• in-segment wildcard (e.g. src/auth/.ts)
• as a cross-segment wildcard (e.g. src//*.{ts,tsx})
• brace expansion {a,b,c} for OR alternatives (single-level, no nesting) Existing trailing /* and / patterns unchanged. Net effect: the [TARGET_FILES] block in spec-tech.md can now declare src/*/.{ts,tsx} instead of enumerating extensions.
• ScopeRecord interface in extensions/stelow/types.ts — TypeBox-friendly shape for the mirror fields above.
• cali-product-execution-critique Criterion 6 (Record Evidence). New criterion (renumbered subsequent criteria 7–10) checks every completed scope for: record present, verified: true, files_count > 0, commands_count > 0, suggested_commit set, and iteration-state-{SCOPE-ID}.md has a ## Record section.
• 5 new test cases in tests/integration/scope-overlap-classify.test.ts covering brace expansion (2-way, 3+way, combined with /**, empty alternatives, unclosed brace as literal).
• Task tracking inside scopes (Shape Up hill chart collapse). New wf.scopes[i].tasks?: ScopeTask[] for runtime task checkboxes that emerge as the scope executes. Each task is source: 'planned' | 'discovered'; discovered tasks always carry a note: explaining the discovery trigger. Two scopes stay at the appetite ceiling (Lean ≤2, Core ≤5, Complete ~10) while individual scopes can carry many tasks — tasks emerge from reality, scopes are committed up front.
• Scope-executor SKILL Step 3e-ter documents seeding, appending, and marking-done patterns for the task checklist. Iteration-state markdown renders the live checklist so a future agent or human can see what's actually done vs. what was planned.
• Muxy cross-workflow scope tab. New "Scopes" view in the muxy panel (integrations/muxy/stelow/src/panel/{app,data}.js) flattens wf.scopes[] across all workflows in the active worktree into hill-chart columns (Pending / In Progress / Escalated / Failed / Completed). Filter strip: status chips + free-text search across scope id, name, type, workflow name, project path. Each card shows scope id, name, workflow, project (last 2 path segments), iteration counter, and a Record badge (✅ verified / ○ unverified) when present. Build verified via npm run build.
• Sandboxing note (muxy limitation): cross-workflow is scoped to the active worktree because muxy.files.read is sandboxed to the workspace root (per Muxy docs). True cross-project would require Muxy to expose a projects.read.files API; tracked for later cycles.

Changed

• Scope interface now has two new optional fields: record?: ScopeRecord and tasks?: ScopeTask[]. Both are advisory in v1 (convention only). Future v2 makes record mandatory for status: 'completed' scopes (TypeBox schema enforcement).
• Criterion numbering in cali-product-execution-critique: Record Evidence is now 6; Gap Registry 7; Lessons Learned 8; Gap-to-Scope 9; Decision Matrix 10. All references updated.
• ScopeRecord field naming: snake_case throughout (completed_at, files_count, commands_count, suggested_commit) to match the rest of stelow.json schema (target_files, actual_files, start_sha, lock_ttl_seconds). v1 advisory references updated.

Tests

• 8 new unit tests in tests/unit/scope-record-tasks.test.ts covering ScopeRecord + ScopeTask shape, planned vs discovered, files_count ↔ actual_files invariance.
• 8 new unit tests in tests/unit/scope-panel-data.test.ts covering SCOPE_COLUMNS order, flattenScopesForView identity preservation, groupScopesByStatus column stability.
• Total: 1099 tests passing (was 1049 in v0.42.1).

Notes for next cycle

• Record v2 → pre-commit hook. The current opt-in runtime validation (STELOW_VALIDATE=1) works but requires an env var. The next cycle should ship a scripts/pre-commit-record.sh that runs the validators from schema-record.ts via require() and blocks commits with missing/incomplete records. Also check discovered_tasks_count > 5 as a project-level threshold.
• Execution-critique Criterion 6 → block threshold. Currently flags missing records as warning/minor. Bump to block after ≥3 workflows ship with v1 discipline (proves the convention is real, not just paper).