Conversation
The generated cask template used the deprecated string-comparison form
`depends_on macos: ">= :{min_macos}"`, which Homebrew now warns against.
Switch to the canonical bare-symbol form `depends_on macos: :{min_macos}`
(min_macos is always a macOS codename, so the symbol is valid and already
means "this version or newer").
This is the root cause that reintroduced the deprecation into generated
casks (cf. Data-Wise/homebrew-tap#112). Also fixed in the reference SPEC.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Third orchestration mode /craft:orchestrate:workflow — code-driven parallel/pipeline/loop control flow with schema-validated JSON per agent. 8 design decisions locked via interrogation 2026-06-12. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Make the full doc/discoverability surface ship-blocking, mirroring the /craft:orchestrate:drive precedent (tutorial, command+help pages, REFCARD, cookbook recipe, modes-compared 2->3, skills-agents catalog, mkdocs nav). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…output brainstorm Step 5.5 now requires a 'Documentation & Discoverability' section in every captured spec; orchestrate:plan template gains a required docs phase + acceptance criterion. Mirrors the full doc surface a shipped craft feature carries (tutorial, help, REFCARD, hub, website). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ot cause confirmed) VALID_FIELDS omits deprecated/replaced-by; command-audit.sh --fix strips them, and test_audit_fix_mode_dry runs --fix against the real tree. Exact diffs + TDD plan for fix/command-audit-deprecated-fields. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…, spec-docs mandate, engine scaffold
…staleness --fix) (#147) Bundles Bug A (command-audit --fix stripped deprecated/replaced-by from 56 commands) and Bug C (docs-staleness --fix rewrote count refs in 3 docs), both via tests invoking --fix against the real tree. Adds isolation helpers + enforcement test, fixes the fix/* branch-type test, and an intentional 108->109 count sync. All CI green; full suite 1672 passed / 0 failed, tree clean.
…:workflow (3rd mode) (#148) * docs(plan): ORCHESTRATE plan for workflow engine (5 increments, FR9 docs surface) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(orchestrate): workflow-engine mechanical core — parser + schema (Increment 1) Increment 1 of the deterministic workflow engine (SPEC-workflow-engine-2026-06-12). TDD-built, stdlib-only determinism core (PyYAML used only for input reading, imported lazily inside parse_yaml so the core never touches it). scripts/workflow_parse.py: - YAML form + frozen shape-DSL form -> identical canonical wave plan (D1, D3). DSL is a tokenizer + recursive-descent reader (no eval), with map/flatMap path-[] agreement enforced so both forms provably converge. - Structural output validator (D2 layer 1, gating): required keys, primitives, homogeneous arrays; bool rejected as number; one error per offending key. - resolve_fanout: empty fan-out hard-aborts naming the upstream stage (D6/FR8). - cache_key content hash + cascade_invalidate downstream propagation (D4). - Run-wide semaphore as a counter FILE — read/acquire/release/reconcile (D5/FR7). commands/orchestrate/_workflow_schema.json: constrained v1 dialect (reference; enforcement is the stdlib validator, no jsonschema). tests/test_workflow_engine.py: 33 unit tests, red->green per decision (D1-D6). Full `pytest tests/` green (1703 passed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(orchestrate): workflow-engine skill executor + CLI/gate hooks (Increment 2) Increment 2 of the deterministic workflow engine — the prompt-driven executor behind /craft:orchestrate:workflow (no runtime dep, like drive-engine). skills/orchestration/workflow-engine/SKILL.md: - Compile -> resolve fan-out -> dispatch file-scoped agents under the run-wide semaphore -> hybrid-gate every output -> cache/replay -> reconcile counter at each wave boundary (D5 residual-risk mitigation). - Hard rule: delegate ALL deterministic mechanics to scripts/workflow_parse.py (plan, structural gate, cache key, cascade, fan-out, semaphore) — never improvise. Claude judges only the advisory semantic layer. - First-class verify gate (D8): runs the project's real verify command, exit status authoritative; reuses drive-engine's auto-detection table so drive stays a strict subset. scripts/workflow_parse.py: - parse_file() + main() CLI: emits the canonical wave plan as JSON (the skill shells out to this) — Increment 2 gate "dry-run emits a sane wave plan". - gate_output(): hybrid D2 gate, failure-isolating and non-raising. Structural miss -> ok=False (fail just that branch); semantic_warning surfaced but never flips the gate — proves the "semantic-warn non-blocking" gate. Counts: skill 38->39 pulled forward to here (plugin.json + CLAUDE.md) so validate-counts stays green now that the skill exists; command count + docs catalog remain Increment 3/5. Tests: +6 (39 total in the file). validate-counts green; new SKILL.md frontmatter valid, no duplicate trigger phrases; markdownlint clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(orchestrate): /craft:orchestrate:workflow thin command (Increment 3) Increment 3 — the thin entry point that owns args and delegates to the workflow-engine skill (mirrors orchestrate:drive). commands/orchestrate/workflow.md: - args: workflow file, --dry-run/-n, --resume <run-id>, --refine (FR6 parity). - 3-mode "which orchestration?" table (orchestrate / drive / workflow). - Step-ordered execution: resolve def -> compile (delegate, never eyeball) -> dry-run box -> execute via skill -> resume replay -> authoritative verify gate. - command-audit clean (0 errors/warnings); markdownlint clean. scripts/workflow_parse.py: - dry_run_actions(plan): one line per wave; statically-unknown fan-out width shown symbolically (xN), never a fabricated number (D1/FR5). - main(): -n/--dry-run flag prints the plan + run-wide ceiling, no JSON. .gitignore += .craft/workflow-runs/ (per-run agent outputs, manifest, semaphore). Counts: command 109->110 (plugin.json + CLAUDE.md); validate-counts green (110 cmds / 39 skills / 8 agents). Tests: +3 dry-run (42 total in file). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(orchestrate): D7 known-shape router suggestion (Increment 4) Increment 4 — task-analyzer detects the decompose→cover→verify→synthesize shape and SUGGESTS /craft:orchestrate:workflow. Never silently switches (routing-false-positive guard is the accepted D7 residual risk). scripts/workflow_parse.py: - detects_workflow_shape(text): conservative heuristic — fires only when ≥3 of the four stage categories (decompose/fan-out/verify/synthesize) appear, or the explicit decompose…synthesize chain. A lone "verify"/"parallel" stays silent. skills/orchestration/task-analyzer/SKILL.md: - New "Coded-Workflow Shape Detection (D7)" section: trigger-word table, suggest-don't-switch confirmation prompt, detected/not-detected examples. Tests: +4 (46 total in file) — full shape fires, parallel+verify+summarize fires, ordinary feature request silent, lone signal silent. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(orchestrate): runnable example + verify-stage + D8 convergence test (Increment 5a) scripts/workflow_parse.py: - verify stages (type: verify) now carry their `command` through to the wave plan (D8) — a static gate whose exit status is authoritative. examples/workflow-code-review/WORKFLOW-code-review-sweep.yaml: - the spec's runnable 5-dim review case (decompose -> cover N -> verify 2/finding -> synthesize); parses and dry-runs cleanly. tests/test_workflow_engine.py (+3, 49 total): - verify stage compiles to a command-carrying static gate. - D8 convergence guard: drive-engine AND workflow-engine skills both keep the "real command, exit status authoritative, green transcript insufficient" semantics — fails if either drifts. - the example file parses to the canonical 4-wave shape. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(orchestrate): FR9 docs surface for :workflow + count sync (Increment 5b) Ship-blocking documentation surface for /craft:orchestrate:workflow, mirroring the orchestrate:drive set (FR9). New pages: - docs/commands/orchestrate-workflow.md — command reference (forms, schema, replay). - docs/help/orchestrate-workflow.md — when-to-use, dry-run reading, failure table. - docs/tutorials/TUTORIAL-orchestrate-workflow.md — hands-on (YAML + shape-DSL, --dry-run, --resume, verify gate). - docs/reference/REFCARD-WORKFLOW.md — quick reference (precedent: REFCARD-CHECK). - docs/cookbook/recipes/run-a-coded-workflow.md — recipe. Updated surfaces: - docs/tutorials/orchestrator-modes-compared.md — new "three orchestration modes" section (orchestrate / drive / workflow) distinct from the 4 execution modes. - docs/REFCARD.md, docs/skills-agents.md (workflow-engine row + 38->39), commands/smart-help.md (Orchestration topic + Q&A + Quick Ref row), mkdocs.yml nav (5 entries), and the three orchestrator guides cross-linked. - CHANGELOG.md + docs/CHANGELOG.md [Unreleased]. Count sync: docs-staleness-check.sh --fix swept the Tier 2 long-tail (109->110 commands, 38->39 skills) across ~20 docs. Staleness now GREEN on all phases (nav, counts, coverage, freshness); validate-counts green; mkdocs builds. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * test(orchestrate): expand --refine scope contract to 6 commands (FR6) Increment 3 added --refine to /craft:orchestrate:workflow per spec FR6 ("consistent with other craft commands"). test_refine_flag_scope pins the exact sanctioned set, so the contract grows 5 -> 6. Update the expected set and the CLAUDE.md note. Caught only by the full suite (test_plugin_e2e). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(orchestrate): adversarial-review hardening of the workflow engine Adversarial pre-PR review surfaced real gaps; closed the genuine ones (TDD): - build_deps(plan) (NEW): derives the stage dependency graph from over/input bindings. cascade_invalidate was tested in isolation but had no way to get `deps` from a real plan — the D4 cascade was effectively unreachable on the resume path. Now wired (skill references build_deps(plan)). - Parse-time binding validation: a binding to an unknown or FORWARD stage now hard-errors at compile time, naming the stage — was silently accepted and only caught at runtime. - flatten() DSL operator: D3's vocabulary lists map/flatMap/flatten; flatten was rejected as "unknown operator". Now accepted (requires a [] path). - loop max_iter: carried into the wave plan (was dropped). Disclosed as v1 limitations (command doc): no dry-run static-max-vs-ceiling warning (FR7 runtime semaphore is the core), loop is represented-not-executed, and shape-DSL agent() requires an options object. Tests: +6 (55 total in file), all green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore(orchestrate): remove ORCHESTRATE working artifact before merge The ORCHESTRATE-*.md plan is a feature-branch working artifact and must not land on dev (per the plan's Done checklist + global workflow rule). The spec remains at docs/specs/SPEC-workflow-engine-2026-06-12.md. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(orchestrate): guard binding resolution + sync counts (PR #148 review B1/B2) B1 — _resolve_binding now fails closed as a structured WorkflowError when a binding walks a missing/non-dict field (both the []-flatten and .field paths), instead of letting a bare KeyError/TypeError abort the whole resolver on heterogeneous object[] outputs. +2 regression tests (57 total in file). B2 — count drift: package.json 109→110 commands / 38→39 skills; plugin.json breakdown 96+13 → 96+14 workflow (the PR's new orchestrate/workflow.md command) so it sums to the authoritative 110. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Test User <test@example.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rker CI flake - PR #148 (deterministic workflow engine) shipped → dev caadd8d (squash, --admin) with B1/B2 review fixes (d1d0588); worktree + remote branch removed. - Next Action A reset to the now-confirmed dev-red root cause: mermaid/mermaid_mcp pytest markers unregistered in pyproject → --strict-markers collection error when the env plugin install flakes. Tiny root-cause fix noted (register both markers). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Documentation gap analysis follow-up for the shipped workflow engine (PR #148): - Count drift (110 commands / 39 skills): README.md (6 refs), docs/index.md (live landing), docs/cookbook/common/find-the-right-command.md, plus the categorical-subtotal headers bump-version.sh doesn't sweep — commands/hub.md `ORCHESTRATE (4)→(5)` + `Skills (38→39)`, and docs/skills-agents.md `### Orchestration (4)→(5)`. - Discoverability: added the missing `/craft:orchestrate:workflow` row to the hub's orchestrate command list (it existed nowhere in hub.md). - Flowcharts (the feature shipped with zero, while sibling orchestrate docs have them): a mode-selection decision flowchart in docs/help/orchestrate-workflow.md and an execution-pipeline flowchart in docs/commands/orchestrate-workflow.md. Both pass the mermaid pre-commit validator; no <br/> tags (repo is phasing them out). validate-counts: PASS. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
test_mermaid_dogfood.py (@pytest.mark.mermaid) and test_mermaid_e2e.py (@pytest.mark.mermaid_mcp) used markers absent from pyproject's markers list. With --strict-markers, that's only safe while an env pytest plugin registers them at install time; when that install flakes, collection hard-errors and the ENTIRE "Craft CI" run aborts (exit 2) — a CI-wide failure unrelated to any diff (observed taking down PR #148 and dev's own merge-commit run on 2026-06-13). Registering both markers makes collection deterministic regardless of plugin availability. Full suite: 1729 passed, 38 skipped (mermaid_mcp tests skip cleanly when the MCP server is absent), 1 xfailed. Co-authored-by: Test User <test@example.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… gaps shipped - PR #149 merged → dev d21a1e4 (markers registered; --strict-markers flake gone). - Doc-gap closure 8854470: counts 110/39, hub entry, 2 flowcharts. - Next Action A resolved; flagged 2 leftover local branches needing manual -D. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Minor release. Highlights (full detail in CHANGELOG): - Added: /craft:orchestrate:workflow — 3rd orchestration mode (deterministic, schema-gated, resumable) + workflow-engine skill + stdlib parser. - Fixed: test-suite source mutation (command-audit + docs-staleness --fix), mermaid/mermaid_mcp --strict-markers CI flake, Homebrew cask depends_on macos symbol form. - Counts: 110 commands, 39 skills, 8 agents. Version stamped across 14 files; [Unreleased] → [2.36.0] in both CHANGELOGs. pre-release-check 2.36.0: PASSED. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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.
v2.36.0 — deterministic workflow engine
Minor release. 18 commits since v2.35.0.
Added
/craft:orchestrate:workflow— a third orchestration mode that runs a coded, fixed-control-flow program (parallel/pipeline/loop/verify) with stdlib-enforced structural output schemas, data-driven fan-out, a run-wide concurrency semaphore, and cached/resumable replay by run-ID.workflow-engineskill +scripts/workflow_parse.py(stdlib-only core).task-analyzersuggests:workflowon decompose→cover→verify→synthesize phrasing.Fixed
command-audit.sh --fix(strippeddeprecated/replaced-by) anddocs-staleness-check.sh --fix(rewrote count strings) both retargeted to isolated temp trees + structural guard.pytest --strict-markersno longer red-lists all of CI when themermaid/mermaid_mcpmarker-provider plugin flakes — both markers registered inpyproject.toml.depends_on macos: :codenameform.test_on_expected_branch_typeacceptsfix/*branches.Counts
110 commands · 39 skills · 8 agents
Pre-flight
pre-release-check.sh 2.36.0: PASSED[Unreleased] → [2.36.0] — 2026-06-13🤖 Generated with Claude Code