Twinned from https://github.com/cfsh/ninefold/issues/420 by tools/beads/import_github.py's reconcile pass.
Filed at ninefold-b's wrap-up. Unclaimed and unlabelled — backlog.
#401 added ~+116 net lines to CLAUDE.md (after a −14 trim in 526d536). Every addition answers a review finding or one of Aedan's inline questions, so none of it is filler — but the PR's own thesis is that this file is re-read by every session on every tick, which makes its length a cost line rather than a style preference.
Aedan was asked to rule three times during the review and didn't, which was the right call mid-flight: the design was still moving. It has now landed, so a dedicated pass is possible without mixing prose edits into behavioural fixes — which is exactly why it was deferred each time.
Candidate targets, from the review's own residuals:
- **Ticks are not free.** and - **Cron**:, so the poll section's bullet list reads as two lists with prose wedged between. Flagged twice by the reviewer, deferred twice. Cosmetic, but this is the section every session is asked to read carefully.The constraint that makes this delicate: the reviewer found the intake rule surviving verbatim in *four* places after #401 removed it, including the cold-start summary. Cutting the wrong contrast is how that happens. A rule that is likely to be reinvented needs its "don't reinvent this" clause; a rule nobody would think to invent does not.
Related: #396, #401, #414 (shell tier has no test harness), #418 (digest scoping), #419 (cron → Monitor).
None.
No comments.