fix(ralph-loop): require tagged completion promise

[abhinavgautam01]

Closes #282

Summary

• Extract completion promises only from complete <promise>...</promise> tags.
• Use printf instead of echo to preserve response text during extraction.
• Add integration regression tests covering tagged, untagged, malformed, mismatched, empty and whitespace-normalized responses.

Why

The previous Perl expression used -p, which prints the complete response even when the <promise> substitution does not match.

As a result, an untagged response that exactly matched the configured completion promise could create the Ralph done flag and prematurely terminate the loop.

The updated extractor prints text only after finding a complete promise tag.

Validation

• git diff --check
• shfmt formatting check
• bash -n on the Ralph hook and test scripts
• bash ralph-loop/hooks/tests/capture-response.test.sh
  • All six regression cases passed
• node scripts/validate-plugins.mjs
  • Result: All plugins validated successfully.

---

Note

Low Risk
Scoped hook behavior fix with regression tests; only affects when the Ralph loop marks itself done, with stricter (safer) completion detection.

Overview
Fixes a bug where the Ralph afterAgentResponse hook could end the loop when the assistant’s plain text matched completion_promise, even without <promise>...</promise> tags. The old Perl -p path could leave the full response as “extracted” promise text on non-matches.

Promise extraction now uses printf (avoids echo mangling) and Perl that only emits inner text when a complete tag pair exists, still trimming and collapsing whitespace before comparing to the configured promise.

Adds capture-response.test.sh with six integration cases: untagged match must not set done, valid tagged matches (including multiline/whitespace), wrong promise, missing closing tag, and empty input.

^{Reviewed by Cursor Bugbot for commit 91ebae5. Bugbot is set up for automated code reviews on this repo. Configure here.}