fix(media): add missing extensions to MEDIA delivery whitelist (#37318)

[luoxiao6645]

Problem

MEDIA:/path/to/file.md tags silently fail for common file types (.md,
.json, .yaml, .py, etc.) — the file is never delivered, send_message
returns misleading {success: true}.

Root Cause

Two of the three MEDIA regexes in the codebase used a stale
hardcoded extension list that never grew beyond the original
image/video/archive set. Only base.py's MEDIA_DELIVERY_EXTS had been
updated; run.py's _TOOL_MEDIA_RE and its inline clone at L17764
carried independent, narrower lists.

Changes

File	Change
gateway/platforms/base.py	Added 8 extensions: .markdown .toml .py .js .ts .sh .bash .zsh
gateway/run.py	Replaced 2x hardcoded regex extension lists with _MEDIA_EXT_ALTERNATION from base.py

All three MEDIA regexes now share one source of truth (60 extensions),
so new extensions never need syncing across files again.

Platforms affected

Feishu, Telegram, and any platform using send_document via MEDIA: tags.

Fixes #37318

[tonydwb]

Code Review Summary

Verdict: Approved ✅

Scope

Adds 8 missing extensions (.markdown, .toml, .py, .js, .ts, .sh, .bash, .zsh) to the MEDIA delivery whitelist and consolidates three separate regex lists into one shared source of truth.

Strengths

• DRY refactor: Replaced two hardcoded inline regex copies in run.py with _MEDIA_EXT_ALTERNATION from base.py — future extensions only need one change location.
• Good root cause analysis: Identified that only base.py had been updated while two regexes in run.py were independently stale.
• Well-scoped: 7 additions, 10 deletions. Clean diff, single concern.
• Correct by inspection: The approach (importing the shared alternation fragment) is syntactically and logically sound.

Minor Notes

• No tests added for the new extensions — acceptable for a whitelist change this simple; regression test would be testing that a constant contains a string.

---

Reviewed by Hermes Agent

[teknium1]

Thanks for narrowing this to the remaining MEDIA extension drift.

Problems

• Current main has only one live _TOOL_MEDIA_RE definition at gateway/run.py:1061; the PR's second replacement targets a duplicate that was removed. GitHub currently reports this branch as dirty, so that obsolete hunk should not be carried forward.
• The PR adds no regression coverage. tests/gateway/test_platform_base.py:616-630 exercises the earlier .md/.json/.yaml class only, and tests/gateway/test_run_tool_media_re.py:23-32 reconstructs the old hardcoded regex.

Suggested changes

• Salvage the current matcher replacement at gateway/run.py:1061 and drop the obsolete duplicate hunk.
• Add behavior tests for the new extensions through BasePlatformAdapter.extract_media() and the auto-append path.

Current gateway/platforms/base.py:1463-1471 still lacks the requested .markdown, .toml, and script extensions, while extract_media() derives its matcher from that tuple at gateway/platforms/base.py:3641-3645; the underlying extension addition remains valid.

Automated hermes-sweeper review.

[teknium1]

Closing as resolved on main: PR #65510 replaced the extension-allowlist approach with universal validated egress — ANY MEDIA: path (.py/.js/.ts/.sh/.toml and everything else) now delivers after passing validate_media_delivery_path. Verified live on main for your target extensions. Your idea of routing run.py's _TOOL_MEDIA_RE through the shared alternation is a reasonable residual improvement for the tool-result auto-append path — that narrow list is now only a secondary precision guard, but if you want to unify it, a focused PR for just that would be welcome. Thanks!