slopclean

Files (24)

• SKILL.md
• openai.yaml · agents/
• criteria.json · evals/agent-first-current-state-and-context-budget/
• task.md · evals/agent-first-current-state-and-context-budget/
• criteria.json · evals/agents-md-content-and-structure/
• task.md · evals/agents-md-content-and-structure/
• criteria.json · evals/config-source-links/
• task.md · evals/config-source-links/
• criteria.json · evals/near-miss-typo-pass/
• task.md · evals/near-miss-typo-pass/
• criteria.json · evals/prune-suite-with-evidence/
• task.md · evals/prune-suite-with-evidence/
• criteria.json · evals/source-boundary-and-local-discovery/
• task.md · evals/source-boundary-and-local-discovery/
• criteria.json · evals/trigger-reads-like-ai/
• task.md · evals/trigger-reads-like-ai/
• criteria.json · evals/unslop-mechanism-and-compression/
• task.md · evals/unslop-mechanism-and-compression/
• criteria.json · evals/unslop-readme-prose/
• task.md · evals/unslop-readme-prose/
• code.md · references/
• docs.md · references/
• prose.md · references/
• tests.md · references/

SKILL.md (first lines)

# Slopclean

Clean the artifact in place, in one pass. Keep meaning, facts, behavior, and
public API; house style wins.

Read the matching reference (source-and-test diffs need both):
[prose](references/prose.md), [code](references/code.md),
[tests](references/tests.md), [docs and agent guides](references/docs.md).

Every contract keeps one automated proof that fails when it breaks: a test at
its strongest boundary, or a type, schema, or lint check. A drive recipe the
gate doesn't run is not that proof. Report real defects and contract changes
separately instead of fixing them in the pass.

From uinaf/ffss, MIT.

More from uinaf/ffss