What
Step 2 of the capture skill requires a knowledge base search before saving, and gives three branches: skip when the knowledge is already captured, edit the existing memory when the new material extends it, or save a new memory with a cross-link.
The pre-write checks in Step 4 have to print their work, and the skill says why: "Hidden reasoning is easy to skip; printed output is reviewable." Step 2 prints nothing, so a search that never ran looks the same as one that found nothing.
Why it matters
The branch Step 2 picks decides whether a session strengthens one memory or splits the same knowledge across two. The audit skill carries two DROP categories for cleaning that up afterward: C.2 for duplication, and H for a narrow learning that a sibling already covers. Every item those catch is a Step 2 that did not happen properly.
Possible fix
One logged line, matching the Step 4 checks:
KB search: "<query>" -> <n> hits, <what they covered> -> <branch taken, and the file edited or created>
This was drafted and reverted along with a larger change set. It adds about five lines to a file that has been shrinking on purpose, which is the argument against it. The question is whether the cheapest gate should be the only invisible one.
What
Step 2 of the capture skill requires a knowledge base search before saving, and gives three branches: skip when the knowledge is already captured, edit the existing memory when the new material extends it, or save a new memory with a cross-link.
The pre-write checks in Step 4 have to print their work, and the skill says why: "Hidden reasoning is easy to skip; printed output is reviewable." Step 2 prints nothing, so a search that never ran looks the same as one that found nothing.
Why it matters
The branch Step 2 picks decides whether a session strengthens one memory or splits the same knowledge across two. The audit skill carries two DROP categories for cleaning that up afterward: C.2 for duplication, and H for a narrow learning that a sibling already covers. Every item those catch is a Step 2 that did not happen properly.
Possible fix
One logged line, matching the Step 4 checks:
This was drafted and reverted along with a larger change set. It adds about five lines to a file that has been shrinking on purpose, which is the argument against it. The question is whether the cheapest gate should be the only invisible one.