fix(langchain): convert prompt variables with non-word characters

[i-anubhav-anand]

When converting a Langfuse prompt to a LangChain template, _get_langchain_prompt_string only matched \w+ variable names:

re.sub(r"{{\s*(\w+)\s*}}", r"{\g<1>}", json_escaped_content)

But Langfuse variable names are permissive (the template parser just .strip()s the name), so a valid prompt like Hello {{user-name}}! is left unchanged. LangChain then reads {{ as a literal brace, so the variable is silently dropped and can never be filled:

client.get_langchain_prompt()                     # "Hello {{user-name}}!"  (not converted)
PromptTemplate.from_template(...).input_variables  # []
.format(...)                                       # "Hello {user-name}!"   (un-fillable)

Fix broadens the capture to [^{}"]+?, so any variable name (hyphens, spaces, unicode, …) converts, while still excluding already-escaped JSON — which appears as {{"key": ...}} after _escape_json_for_langchain and must stay doubled. Verified the JSON case stays correctly escaped (the existing JSON-escaping tests still pass).

Test

Adds a unit test in TestLangchainPromptCompilation for a hyphenated variable (fails before: left as {{user-name}}). All 33 tests in the file pass. ruff check/format clean.

Type of change

• Bug fix (non-breaking change which fixes an issue)

Greptile Summary

This PR fixes a silent variable-drop bug in _get_langchain_prompt_string where Langfuse variable names containing non-word characters (hyphens, spaces, unicode) were never converted to LangChain single-brace format, leaving them as un-fillable {{...}} literals.

• The regex capture group is broadened from \w+ to [^{}"]+?, so any variable name not containing braces or double-quotes now converts correctly.
• A new unit test (test_get_langchain_prompt_preserves_special_variable_names) covers the hyphen case end-to-end through PromptTemplate.format.
• The fix is one line in model.py; all existing JSON-escaping tests continue to pass per the PR description.

Confidence Score: 3/5

The core regex fix is correct for double-quoted JSON content, but it introduces a regression for single-quote–prefixed content that _escape_json_for_langchain also escapes.

The escape function at line 213 doubles braces for both ' and " prefixes, so a prompt containing {'key': 'value'} is escaped to {{'key': 'value'}}. The new [^{}"]+? character class excludes " but not ', so that escaped pair is still matched and incorrectly converted to a LangChain placeholder instead of being left as a literal. The old \w+ was accidentally immune to this. The fix is one character away ([^{}"']+?), but as written the PR leaves this gap.

langfuse/model.py — specifically the character class in _get_langchain_prompt_string and whether the new test suite covers single-quote–delimited content going through _escape_json_for_langchain.

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
langfuse/model.py:170
`_escape_json_for_langchain` doubles braces for both `"` **and** `'`-prefixed content (line 213: `text[j] in {"'", '"'}`). A prompt like `The result is {'key': 'value'}` is escaped to `{{'key': 'value'}}`, but `[^{}"]+?` does not exclude `'`, so the new regex matches it and converts it to `{'key': 'value'}` — a LangChain placeholder — instead of leaving the brace-pair literal. The old `\w+` was accidentally safe here; this PR introduces a regression for single-quote–prefixed content. Adding `'` to the exclusion class keeps parity with the escaper.

_{Reviews (1): Last reviewed commit: "fix(langchain): convert prompt variables..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[greptile-apps]

_escape_json_for_langchain doubles braces for both " and '-prefixed content (line 213: text[j] in {"'", '"'}). A prompt like The result is {'key': 'value'} is escaped to {{'key': 'value'}}, but [^{}"]+? does not exclude ', so the new regex matches it and converts it to {'key': 'value'} — a LangChain placeholder — instead of leaving the brace-pair literal. The old \w+ was accidentally safe here; this PR introduces a regression for single-quote–prefixed content. Adding ' to the exclusion class keeps parity with the escaper.

Prompt To Fix With AI

This is a comment left during a code review.
Path: langfuse/model.py
Line: 170

Comment:
`_escape_json_for_langchain` doubles braces for both `"` **and** `'`-prefixed content (line 213: `text[j] in {"'", '"'}`). A prompt like `The result is {'key': 'value'}` is escaped to `{{'key': 'value'}}`, but `[^{}"]+?` does not exclude `'`, so the new regex matches it and converts it to `{'key': 'value'}` — a LangChain placeholder — instead of leaving the brace-pair literal. The old `\w+` was accidentally safe here; this PR introduces a regression for single-quote–prefixed content. Adding `'` to the exclusion class keeps parity with the escaper.

How can I resolve this? If you propose a fix, please make it concise.

[i-anubhav-anand]

Good catch — fixed in the latest commit. _escape_json_for_langchain doubles both quote styles, so I broadened the exclusion class to [^{}"']+? and added a regression test (test_get_langchain_prompt_leaves_quoted_json_escaped) covering single-quote JSON.