Closes#137. The 0.36 -> 0.38 bump (skipping 0.37) introduces Tower-based
middleware support and a fix to the ReasoningItem `type` field, but
neither affects baibot's call sites — no code adaptation was needed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
matrix-sdk 0.17.0 brings a few breaking API changes that we adapt to:
- Drop the `native-tls` feature on matrix-sdk; it was removed upstream
in 2026-04 (matrix-sdk now hardwires rustls). The previous workaround
comment referencing rust-mxlink issue #1 is no longer relevant.
- `Relation::Reply` is now a tuple variant wrapping a `Reply` struct;
pattern matches updated to bind through `reply.in_reply_to.event_id`.
Also bumps the Rust toolchain from 1.93.0 to 1.95.0 (closes#85, which
proposed the Dockerfile bump in isolation). rustc 1.94+ trips a
query-depth overflow when computing async layouts in the matrix-sdk
timeline future graph; matrix-rust-sdk PR #6489 raises the limit, but
\`recursion_limit\` is per-crate, so we repeat \`#![recursion_limit = "256"]\`
on this crate root.
Bumps the mxlink floor to 1.14.0, which carries the matching adapter.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Run the publish workflow from the workflow_run event so Docker publishing happens only after the CI workflow completes successfully for push events on main or v* tags.
Check out the exact SHA validated by CI and derive Docker metadata from the upstream CI ref, so publishing follows the tested revision instead of the default branch tip.
This supersedes 7d183b9 ("Run CI for pull requests"), which mixed validation and publishing in one workflow and regressed docker-manifest by dropping the package-write permission it needs to publish the manifest.
Split the workflows so CI handles pull requests, branch pushes, tags, and manual runs, while publishing stays focused on Docker delivery with the manifest permission fixed explicitly at the job level.
Add a per-room/global `sender_context_mode` setting that optionally
prefixes conversation messages with sender metadata before sending them
to the model provider.
This helps models distinguish between participants in multi-user rooms.
Three modes are supported:
- `disabled` (default, no change)
- `matrix_user_id` (prefixes with `[sender=@user:server]`)
- `matrix_user_id_and_timestamp` (adds `send_at`; Example: `[sender=@user:server sent_at=<ISO 8601>]`)
Sender context is applied to user and assistant text messages only,
skipping system prompts and non-text content.
Mixed-sender merged turns (something we intentionally do for Anthropic)
have their `sender_id` cleared to avoid misattribution.
async-openai 0.34 adds a required phase field to EasyInputMessage,
which broke our Responses API request construction.
Set phase to None because baibot does not currently model assistant commentary vs final-answer turns,
so omitting phase preserves the previous behavior while keeping the request compatible with the new crate.
Files with unrecognized MIME types (application/octet-stream) are
not supported by any LLM provider and would cause errors that
permanently break the conversation thread. Instead, represent them
as a text message describing the attachment so the LLM is still
aware a file was sent without the thread becoming unusable.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the hand-maintained extension-to-MIME mapping with the
mime_guess crate, which was already in the dependency tree.
This covers hundreds of file extensions out of the box.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
