Bug Description
When an agent generates a file (e.g. a .docx report) and the LLM responds to the user, it often wraps the file path in Markdown bold syntax: **MEDIA:/path/to/file.docx**.
The extract_media() function in gateway/platforms/base.py uses a regex that captures the path group, but the subsequent rstrip() does NOT strip trailing asterisks (*). This causes the literal ** characters to remain in the extracted path, resulting in a file-not-found error.
Root Cause
File: gateway/platforms/base.py
Line: ~2020
Current code:
path = path.lstrip("`\"'").rstrip("`\"',.;:)}")The rstrip character set does not include *, so when the LLM outputs **MEDIA:/path/to/file.docx**, the captured path becomes /path/to/file.docx** — a non-existent file.
Reproduction
- Use any platform that inherits
BasePlatformAdapter.extract_media() (QQ, WeChat, Telegram, etc.)
- Have a Skill generate a document file (e.g. .docx, .xlsx)
- The LLM responds with
**MEDIA:/临时/报告.docx** (Markdown bold wrapping the path)
- The extracted path becomes
/临时/报告.docx** — file not found
- Send fails with error:
Media file not found: /临时/报告.docx**
Affected Platforms
All platforms that use BasePlatformAdapter.extract_media():
- QQ (qqbot)
- WeChat (weixin)
- Telegram
- And likely all other platform adapters
Why This Happens
The PLATFORM_HINTS for QQ/WeChat platforms encourage Markdown formatting:
"QQ supports markdown formatting and emoji"
This leads LLMs (especially Chinese-language models) to naturally bold paths using **...**. The extract_media() regex correctly identifies MEDIA:/path/file.docx but the trailing ** from the bold syntax leaks into the path group because rstrip does not strip *.
Suggested Fix
Add * to the rstrip character set in base.py line ~2020:
path = path.lstrip("`\"'").rstrip("`\"',.;:)}*")This is a one-character change with minimal risk — * is extremely rare in valid file paths.
Impact
- Severity: Medium — prevents file delivery on all platform adapters
- Frequency: Any time the LLM uses bold syntax around a MEDIA path
- Workaround: Manually edit base.py after each Hermes update to add
* to rstrip
Related Issues
Bug Description
When an agent generates a file (e.g. a .docx report) and the LLM responds to the user, it often wraps the file path in Markdown bold syntax:
**MEDIA:/path/to/file.docx**.The
extract_media()function ingateway/platforms/base.pyuses a regex that captures the path group, but the subsequentrstrip()does NOT strip trailing asterisks (*). This causes the literal**characters to remain in the extracted path, resulting in a file-not-found error.Root Cause
File:
gateway/platforms/base.pyLine: ~2020
Current code:
The
rstripcharacter set does not include*, so when the LLM outputs**MEDIA:/path/to/file.docx**, the captured path becomes/path/to/file.docx**— a non-existent file.Reproduction
BasePlatformAdapter.extract_media()(QQ, WeChat, Telegram, etc.)**MEDIA:/临时/报告.docx**(Markdown bold wrapping the path)/临时/报告.docx**— file not foundMedia file not found: /临时/报告.docx**Affected Platforms
All platforms that use
BasePlatformAdapter.extract_media():Why This Happens
The PLATFORM_HINTS for QQ/WeChat platforms encourage Markdown formatting:
This leads LLMs (especially Chinese-language models) to naturally bold paths using
**...**. Theextract_media()regex correctly identifiesMEDIA:/path/file.docxbut the trailing**from the bold syntax leaks into the path group becauserstripdoes not strip*.Suggested Fix
Add
*to therstripcharacter set inbase.pyline ~2020:This is a one-character change with minimal risk —
*is extremely rare in valid file paths.Impact
*to rstripRelated Issues