fix(force-model): skip leading injected tag blocks before leading-line guard#238
Merged
Merged
Conversation
…e guard Claude Code prepends <system-reminder>, <command-name>, etc. ahead of the user's typed text on every turn, which buried /force-model past the first non-empty line and caused parseForceModelCommand to bail. The pin was never written, cluster routing picked the default, and sticky session affinity latched onto whatever the scorer chose — observed in prod as /force-model silently no-op'ing with reason=no_pin. leadingInjectedPrefixEnd skips complete <tag>...</tag> blocks at the head of the message before applying the leading-line check. Strict tag-name matching (identifier only, no attributes) preserves the original paste- safety guard: arbitrary HTML/XML, unclosed tags, and attribute-bearing tags still terminate the prefix scan.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
/force-modeland/unforce-modelsilently no-op'd in prod (reason=no_pinon every request) because Claude Code prepends<system-reminder>,<command-name>,<local-command-stdout>, etc. ahead of the user's typed text on every turn. The parser's leading-non-empty-line guard saw the injected tag instead of the command and bailed; the pin was never written, cluster routing picked the default, sticky affinity latched the session onto whatever the scorer chose.leadingInjectedPrefixEndskips complete<tag>...</tag>blocks at the head of the message before applying the leading-line check. Strict tag-name matching (identifier only — letters, digits,-,_, no attributes) preserves the original paste-safety guard: arbitrary HTML/XML, unclosed tags, and attribute-bearing tags still terminate the prefix scan, so a stray/force-modelburied in pasted content is still ignored.Skipped blocks are preserved in the stripped body so downstream prompt context (system-reminders, command metadata) reaches the upstream intact.
Test plan
go test ./internal/translate/... ./internal/proxy/...— all greenforce_model_test.go:<system-reminder>before command