refactor(llm): migrate iMessage + Telegram call_llm onto codec_llm (A-12 bridges)#71
Merged
Conversation
…-12 bridges) PR-3E-bridges. The two outbound bridges hand-rolled the same OpenAI chat/completions text call; both now route through codec_llm.call. Bridges WANT graceful degradation, so the default never-raise behavior is exactly right: codec_llm.call returns "" on error/no-choices/timeout/empty, which maps back to the bridge's None-on-failure contract via `content if content else None`. enable_thinking=False is preserved by passing llm_cfg["kwargs"] minus chat_template_kwargs; <think> stripping moves into codec_llm. Auth parity (Bearer iff api_key) and the 1500/0.7 tuning are kept. The vision sites in both bridges (llm_cfg["vision_url"]) are A-11 / codec_vision, not this tranche — left untouched. Removed a now-dead `import re` in codec_imessage (its only use was the inline <think> strip). Tests: tests/test_llm_bridges.py (8 — each bridge returns content on success, returns None on empty/failure (graceful-degradation contract), passes base_url/model/api_key through and filters chat_template_kwargs out of extra_kwargs; source invariants confirm codec_llm.call + only the vision /chat/completions site remains). Full suite 1431 passing, 23 known-baseline failures, zero new. Net -50 LOC. Zero net-new ruff. No skills/ touched. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
5 tasks
AVADSA25
added a commit
that referenced
this pull request
May 23, 2026
…omplete) (#79) PR-3F, Option 1 (scoped). The audit's full BridgeRouter (unifying process_message) was high-risk/low-value: the headline call_llm dup was already fixed in #71, and the two process_message flows have intentionally drifted (telegram audio/Gemini/ briefing; imessage goals/intent) — forcing them into one pipeline would risk regressing a working channel. So this extracts ONLY the genuinely-identical helpers and leaves process_message per-bridge. New codec_bridges.py: - load_dispatch() / try_skill(text) — the lazy codec_dispatch loader + skill match (+ the _SKIP_SKILLS terminal/GUI set), one copy. - call_llm(channel, text, llm_cfg, ...) — the canonical bridge LLM call: persona selected by channel, routed through codec_llm.call, None-contract preserved, chat_template_kwargs filtered. - save_to_memory(channel, conv_id, user, assistant) — the memory.db write under session_id "<channel>-<conv_id>". codec_telegram + codec_imessage now `from codec_bridges import try_skill` and keep thin channel-injecting wrappers for call_llm / save_to_memory, so every process_message call site is unchanged (telegram audio + imessage goal-tracking untouched). Removed telegram's now-dead sqlite3 import. codec_bridges.py is the documented "add a channel" seed (CLAUDE.md §1: future WhatsApp/Discord). Updated 2 stale #71 source-invariants (the bridges now delegate via codec_bridges rather than calling codec_llm.call directly). Tests: tests/test_bridges.py (10 — call_llm persona/None/kwargs-filter/history+ override, try_skill skip-set/run/no-match, save_to_memory channel-prefixed session_id, source invariants). Full suite 1502 passing, 23 known-baseline failures, zero new. Net -148 LOC in the bridges. Zero net-new ruff. **Audit-A (Wave 3, code quality) is complete** — A-1..A-22 all closed or verified. Co-authored-by: Mickael Farina <farina.mickael@gmail.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
PR-3E-bridges. The two outbound bridges hand-rolled the same OpenAI
chat/completionstext call — bothcall_llmtext sites now route throughcodec_llm.call. Design-first →docs/PR3E-BRIDGES-DESIGN.md. The simplest tranche: never-raise is the default, so this is a near-mechanical swap.Migrated
codec_telegram.call_llm(text site:482)codec_imessage.call_llm(text site:343)Bridges want graceful degradation, so the default never-raise is exactly right:
codec_llm.callreturns""on error/no-choices/timeout/empty, mapped back to the bridge'sNone-on-failure contract viacontent if content else None. Preserved:enable_thinking=False(by passingllm_cfg["kwargs"]minuschat_template_kwargs), Bearer-iff-api_keyauth, the 1500/0.7 tuning.<think>stripping moves into codec_llm. Removed a now-deadimport reincodec_imessage.The vision sites in both bridges (
llm_cfg["vision_url"]) are A-11 / codec_vision, not this tranche — left untouched.Test plan
tests/test_llm_bridges.py— 8 tests: each bridge returns content on success; returnsNoneon empty/failure (graceful-degradation contract); passesbase_url/model/api_keythrough and filterschat_template_kwargsout ofextra_kwargs; source invariants (each callscodec_llm.call(,/chat/completionscount down to 1 = only the vision site remains).reimport that the migration orphaned).skills/touched → no manifest regen.🤖 Generated with Claude Code