Feature request: option to suppress thinking-block relays (CCGRAM_HIDE_THINKING)

[RedelRedel]

Problem

With extended-thinking models (e.g. Claude with thinking enabled), every intermediate thinking block is relayed to Telegram as its own message. On a longer task this means a steady stream of notifications for content that is only scratch reasoning, not an actual answer.

CCGRAM_HIDE_TOOL_CALLS (added in #64) solved exactly this for tool_use/tool_result noise, but it does not cover thinking blocks: in handlers/messaging_pipeline/message_queue.py the hidden-filter only matches content_type in ("tool_use", "tool_result"), and the only gate for thinking messages is the 20-char minimum length in handlers/messaging_pipeline/message_routing.py (_MIN_THINKING_LENGTH). Verified against 4.1.0 and current main.

Proposal

A global env toggle, symmetric to the existing one:

CCGRAM_HIDE_THINKING=true

When set, thinking-type messages are dropped before relay; text replies, tool messages (subject to their own setting) and status updates are unaffected. A per-window override via the /verbose cycle would be a nice bonus, but the global env var alone already solves the pain.

Workaround I'm running now

Three-line local patch in message_routing.py:

_HIDE_THINKING = os.getenv("CCGRAM_HIDE_THINKING", "").lower() in ("1", "true", "yes")

# in handle_new_message, inside the per-user loop:
        if msg.content_type == "thinking":
            if _HIDE_THINKING:
                continue

Works fine, but gets wiped on every uv tool upgrade ccgram. Happy to turn this into a PR (including the config plumbing in config.py analogous to hide_tool_calls) if you'd take it.

Thanks for ccgram — the terminal-first design is exactly right.

[alexei-led]

Fixed in v4.3.12: https://github.com/alexei-led/ccgram/releases/tag/v4.3.12