feat(task): compact flow-style arrays in emitted task.md frontmatter - #967
Merged
Conversation
Frontmatter emitters used yaml.safe_dump, which renders every sequence as multiline block bullets — a hand-written 'tags: [parsing, nlp]' exploded into bullets on any migrate/normalize round-trip. Add _CompactDumper (SafeDumper subclass) in task/_document_parse.py whose sequence representer emits flow style for lists of short scalars (str/int/float/ bool/None, no embedded newlines) whose standalone flow rendering fits in 80 chars, and block style otherwise. One canonical dump_frontmatter_yaml helper is shared by all three frontmatter dump sites (render_task_md, render_normalized_task_md, skill_eval verifier.md); the machine-consumed litellm config dumps are intentionally untouched. Output remains deterministic and a dump/load/dump fixed point; PyYAML handles quoting for items with YAML metacharacters inside flow style.
…date scalars Review follow-ups: the 80-char cap measured the list standalone, so a list under a key prefix could render past the cap or wrap mid-flow at deep nesting (uglier than the block bullets it replaces) — charge a fixed 8-char allowance since PyYAML exposes no emit-time column. Also admit datetime.date/datetime items (SafeDumper renders them fine; date lists previously stayed block for no reason).
xdotli
temporarily deployed
to
pypi-internal-preview
August 9, 2026 19:34 — with 
GitHub Actions
Inactive
GitHub Actions
Inactive
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 85817e7421
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
|
|
||
| def test_frontmatter_short_scalar_list_renders_flow_style() -> None: | ||
| """Short scalar-only lists stay compact instead of exploding into bullets.""" |
There was a problem hiding this comment.
The newly added tests guard this commit’s compact-array behavior, but their docstrings only restate the expected behavior—and two tests have no docstring—so none identifies the PR or commit being protected. Add the required reference to each regression-test docstring so future maintainers can trace the intended contract.
AGENTS.md reference: AGENTS.md:L17-L17
Useful? React with 👍 / 👎.
Merged
xdotli
added a commit
that referenced
this pull request
Aug 9, 2026
Version 0.6.7 (from 0.6.7.dev0), CITATION date 2026-08-09, and the 0.6.7 CHANGELOG section covering #949-#967. Also in this cut: - Point the wheel-shipped env0 pins at the org-owned ghcr.io/benchflow-ai/env0:0.2.0 base image (they named a personal Docker Hub image while their own comments and the docs declared ghcr authoritative) and drop the PR-archaeology comments from files that ship to PyPI. Validated on Daytona: 8/8 services ready, reward 1.0. - Document mean_reward in the summary.json layout, the per-task failure block and live-token footer in the CLI reference, and soften the external-agents cost figure (the underlying rate table is a placeholder).
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.
User request: frontmatter YAML gets long — short lists like
tagsshould render as one-line arrays instead of multiline bullets. Parsing already accepted flow style (standard YAML); this fixes the emitters, so hand-writtentags: [parsing, nlp]survives migrate/normalize round-trips instead of exploding into bullets, and generated frontmatter is compact.What changed
_CompactDumper(yaml.SafeDumper)+ one representer intask/_document_parse.py: a list emits flow iff every item is a short scalar (str/int/float/bool/date/None, no embedded newlines) and its rendered flow form — measured with the real PyYAML emitter, so the predicate can never disagree with actual quoting — fits the width cap minus a fixed context allowance (PyYAML exposes no emit-time column, so the allowance stands in for key prefix/indentation). Block style otherwise. Deterministic; dump→load→dump is a fixed point.dump_frontmatter_yaml()replaces three independentyaml.safe_dumpcall sites (both_document_parserender paths +skill_eval/_corevia the existing façade).tags:/- parsing/- nlp→tags: [parsing, nlp]; mappings and scalars byte-identical.Review
Structural review (APPROVE-WITH-MINORS): measure-with-the-real-emitter called out as the right judo (no reimplemented quoting rules to drift); alias/anchor and self-reference safety verified empirically; the
tupleregistration confirmed an override, not a legalization (SafeDumper already serialized tuples); mutation-tested end-to-end throughrender_task_md. The one worth-taking minor — the standalone-width measurement could overflow or wrap in context — is fixed here with the context allowance; date scalars admitted per the same review.Gates: ruff format/check, ty; document/frontmatter/migrate/skill_eval sweep 244 passed (one pre-existing env-dependent collection error in test_daytona_download.py, untouched by this branch, excluded).