Routed up from Preponderous-Software/acsf-dev-loop#36, which was classified template-rule during triage and is therefore not implemented in that instance.
The pattern (positive signal — retain)
In an ACSF cycle, a 7-line docstring fix and a 219-line new test file — filed minutes apart in Phase 1, logically independent, in different directories (core/ vs tests/) — were shipped as one PR. It landed clean with one CI run, one squash commit, and two issues auto-closed. Split into two PRs, the same work would have cost two CI runs and two review rounds for no benefit.
Phase 2 already says "1–2 is the default" and that coherent batching takes precedence. What is missing is the positive heuristic that tells an agent what to actively look for when batching, rather than only accepting a bundle when it happens to arise.
Suggested template text (Phase 2 — Work selection)
Bundle a small docstring/annotation fix with a co-located test or doc update when both are under ~250 lines and touch different directories. This is a sweet spot: one PR, one CI run, two issues closed, and a diff a reviewer scans in thirty seconds. Look for this pairing deliberately during work selection, not only when it arises by chance.
Why upstream
Skills evolved from failure signals alone drift toward over-correction, and this template's "What went well that was non-obvious" rubric item exists for exactly this reason. The heuristic is not specific to the ACSF project — it applies to any repo whose tests live in a separate directory from the code they cover.
This issue body was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).
drafted by Claude on behalf of Daniel Stephenson
Routed up from
Preponderous-Software/acsf-dev-loop#36, which was classifiedtemplate-ruleduring triage and is therefore not implemented in that instance.The pattern (positive signal — retain)
In an ACSF cycle, a 7-line docstring fix and a 219-line new test file — filed minutes apart in Phase 1, logically independent, in different directories (
core/vstests/) — were shipped as one PR. It landed clean with one CI run, one squash commit, and two issues auto-closed. Split into two PRs, the same work would have cost two CI runs and two review rounds for no benefit.Phase 2 already says "1–2 is the default" and that coherent batching takes precedence. What is missing is the positive heuristic that tells an agent what to actively look for when batching, rather than only accepting a bundle when it happens to arise.
Suggested template text (Phase 2 — Work selection)
Why upstream
Skills evolved from failure signals alone drift toward over-correction, and this template's "What went well that was non-obvious" rubric item exists for exactly this reason. The heuristic is not specific to the ACSF project — it applies to any repo whose tests live in a separate directory from the code they cover.
This issue body was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).
drafted by Claude on behalf of Daniel Stephenson