General-purpose coding policy for Baruch's AI agents
95
91%
Does it follow best practices?
Impact
96%
1.31xAverage score across 10 eval scenarios
Advisory
Suggest reviewing before use
[skip ci] to commit messages<type>/<description> (e.g., feat/add-auth, fix/null-pointer, chore/update-deps)gh run watch or equivalent to monitor the run in real timeLatest Version BEFORE the merge as a baseline (tessl tile info <workspace>/<tile> for Tessl tiles, or the equivalent for other registries); after merge, wait for the publish workflow to reach a terminal state (gh run watch <publish-run-id> — used to time the check, not as the gate); then confirm the registry's Latest Version advanced past the pre-publish baseline. Registry-advanced = the new version landed; registry-not-advanced = it didn't, regardless of the workflow's exit code. Do not derive a specific expected version from the merge SHA's manifest (the bump runs inside the publish workflow after that SHA is read, so the merge SHA carries the pre-bump value), and do not compare against a specific expected version (interleaved merges in busy repos may advance the registry past your specific version — still success). The Tessl registry never rejects a published version: moderationPassed: false means the version still lands and is fetchable but won't surface in tessl search; a security finding means tessl install requires an override flag to install — neither is rejection. If the registry didn't advance, the publish failed; do not invent moderation states as a hedge explanation. Naively re-running a failed publish can create an extra release when the workflow includes a version-bump step (as tesslio/patch-version-publish does); the safer recovery is a follow-up commit that fires a fresh publish on mergemain or master