feat: variant task-scoping + tactical-ddd skill benchmark example#54
Merged
Conversation
szjanikowski
force-pushed
the
experiment/tactical-ddd-focus
branch
2 times, most recently
from
da8ec5b to
c5c0961
Compare
szjanikowski
changed the title
szjanikowski
force-pushed
the
experiment/tactical-ddd-focus
branch
7 times, most recently
from
1e8c524 to
56cf9b8
Compare
- Variant task-scoping: a variant.toml may declare `tasks = [...]` to restrict that variant to specific tasks (e.g. a repo-specific skill that should never run against the wrong codebase). With --all-variants, a scoped variant runs only against its declared tasks; the scope wins even over an explicit --tasks filter. (runner: load_variant_task_scope / scope_tasks_for_variant; cli wiring; scaffold template gains a commented example; tests in test_runner.py.) - Raise default evaluator max_turns 30 → 60 (error_max_turns on DDD-heavy workspaces). - Bump starlette 1.0.0 → 1.1.0 (PYSEC-2026-161, transitive via harbor/fastapi/mcp). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reworks the ddd-architectural-challenges example into a worked study of a public DDD skill (Nick Tune's tactical-ddd) and a repo-tuned version, measured with NASDE across four variants (vanilla / guided / public-skill / repo-tuned) on two tasks: a clean-feature task (weather discount) and a legacy anemic→rich refactor (movie-rental), both on .NET 8. - Tasks: add csharp-movie-rental-anemic, modernize csharp-anemic-to-rich-domain to .NET 8, align rubrics + verifiers. - Variants: public skill (repo-leaks removed, idiomatic C#), three repo-tuned skills; explicit skill invocation in instructions so activation is deterministic. - README: new Claude/skill deep-dive section — per-dimension radars, increment-over- vanilla line (absolute cross-task scores aren't comparable; increments are), and average token/time charts. Existing cross-agent table kept. - Lessons surfaced: judge per-dimension not one aggregate; a mounted skill isn't a used skill (verify activation). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
szjanikowski
force-pushed
the
experiment/tactical-ddd-focus
branch
from
56cf9b8 to
5a7076b
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two logical changes, squashed into two commits.
1.
feat: variant task-scoping (+ two small fixes) — toolkit codeA
variant.tomlmay now declaretasks = [...]to restrict a variant to specifictasks. Use it for a repo-specific variant (e.g. a skill whose examples reference one
repo's conventions) so it never runs against the wrong codebase. With
--all-variants, a scoped variant runs only against its declared tasks; thescope wins even over an explicit
--tasksfilter. Absent/empty → unscoped (default).Bundled with two small fixes that belong with the toolkit:
max_turns30 → 60 (avoidserror_max_turnson DDD-heavy workspaces),starlette1.0.0 → 1.1.0 (PYSEC-2026-161, transitive via harbor/fastapi/mcp).Covered by tests (
test_runner.py,test_config.py); scaffold template documents the new field.2.
example: tactical-ddd skill benchmark —examples/onlyReworks the
ddd-architectural-challengesexample into a worked study of a publicDDD skill (Nick Tune's
tactical-ddd) and a repo-tuned version, measured across fourvariants (vanilla / guided / public-skill / repo-tuned) on two tasks — a clean-feature
task (weather discount) and a legacy anemic→rich refactor (movie-rental), both .NET 8.
New README section with per-dimension radars, an increment-over-vanilla chart, and
token/time charts. No toolkit code touched by this part.
Quality gates green (ruff/format/mypy/pytest ×4 platforms, CVE audit, benchmark configs).