docs(skill): section 456 refuted the same day, and corrected into a stopping rule - #3028
Merged
Conversation
…topping rule The recipe that section shipped -- fetch, all checks in one shell call, an unbroken read, then publish -- worked once and then failed three times in a row on the next publish of the same artifact. fetch, merge, Read all 3061 lines, publish: refused as "not built on it". Resent unchanged: refused as "identical content already refused". Fetched again and published: refused with the same message. Two COMPLETE reads of one file in one turn, each with nothing between the last Read and the publish, and neither counted. Three candidate mechanisms and no way to separate them from inside the session: a per-path base version recorded by the one successful publish, a refusal invalidating prior reads, or a fetch invalidating them. Measured cost: roughly 250k tokens of context on one artifact in one turn, for a page already merged and verified on disk. So the section stops being a recipe. When a precondition cannot be inspected and following its own stated instruction twice does not satisfy it, a third attempt is a state change made while blind. Hand the artefact over and say what refused it. The durable repair is the merge command filed as #3023. Refs #3023
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
§456 shipped a recipe this morning for republishing the shared loop dashboard.
It worked once. It then failed three times in a row on the very next
publish, and the section is corrected in place rather than left standing.
Two COMPLETE reads of the same 3061-line file in one turn, each with
nothing between the last Read and the publish, and neither counted.
What is established, and what is not
path this session had never successfully published before. Every attempt
after that first successful publish was refused.
the successful publish, whether a refusal invalidates prior reads, or whether
a fetch does. Three candidate mechanisms, and no way to separate them from
inside the session.
turn, for a page that was already merged and verified on disk — 3 expected
line differences, counts closing at +22 chips / +8 entries / +3 options.
Why it stops being a recipe
A third read would have been a state change made while blind, which is the one
move this repository's debugging doctrine names outright. So the section now
says: when a precondition cannot be inspected and following its own stated
instruction twice does not satisfy it, hand the artefact over and name what
refused it.
The durable repair is #3023 — the merge command — not another reading of the
same file.
Refs #3023
🤖 Generated with Claude Code