Repository navigation
[pull] main from can1357:main - #136
Merged
Merged
Conversation
…hropic route Bedrock Mantle serves the Anthropic Messages API at https://bedrock-mantle.{region}.api.aws/anthropic and accepts the compact-2026-09-04 beta: it returns a signed compaction block and adopts it on the next request. The compaction class rule and endpoint gate only allowed the Claude API, Vertex, Foundry, and Claude Platform on AWS, so Claude models routed through Mantle always fell back to local summaries. Extend the class rule to the bedrock-mantle provider and accept Mantle's /anthropic path as a compaction endpoint. The /openai routes and the bedrock-runtime Converse API stay excluded; Converse rejects the beta.
Amazon Bedrock serves the Anthropic Messages API under /anthropic on
bedrock-runtime (the route AWS recommends) as well as bedrock-mantle.
Every omp request to the runtime route failed with a 400, and neither
route allowed on-demand compaction for the amazon-bedrock provider.
- Detect both routes with isBedrockAnthropicRoute, including Mantle's
{region} template. Treat them as thinking-signing hosts.
- On both routes, drop tool `strict` (both reject it) and fit
metadata.user_id to Bedrock's request-metadata pattern (runtime
rejects other values). This runs on the effective request URL and
after the onPayload hook.
- Enable the compaction class rule for amazon-bedrock and trust
Bedrock's /anthropic URLs in the compaction gate.
- Document setup, request shaping, and the on-demand protocol with AWS
and Anthropic sources.
Amazon Bedrock's OpenAI Responses routes (`/openai/...` on
bedrock-runtime.<region>.amazonaws.com and bedrock-mantle.<region>.api.aws)
serve both OpenAI compaction methods: V1 `POST /responses/compact` and V2
streamed `compaction_trigger`. omp only used them for provider `openai` or
with an explicit `remoteCompaction` opt-in, so GPT models on Bedrock fell
through to snapcompact or handoff even with `remote` first in
`compaction.methodOrder`.
- Detect the route by URL (`isBedrockOpenAIUrl` in pi-catalog hosts) and
enable V1 and V2 for `openai-responses` models on it, for any provider id.
`remoteCompaction.enabled: false` and `v2StreamingEnabled: false` still win.
- Shape both compaction requests like a normal turn: resolve configured
headers, apply the transport fetch (provider proxy, extra CA, User-Agent,
request recording), then run the provider's `prepareModel`/`prepareRequest`
hooks (new `prepareProviderRequest`, also used by `stream()`). Without this
the bundled `bedrock-mantle` provider sent compaction to the unresolved
`bedrock-mantle.{region}.api.aws` host and skipped its bearer/SigV4 fetch.
- V2 takes its body model from the prepared model, so provider model-id
rewrites reach the request.
- V1 strips all trailing slashes from the base URL, appends `/compact` to a
`/responses` base, and no longer sends `Authorization: Bearer N/A` for
keyless credentials.
- Route OpenAI native compaction through provider request setup only for Bedrock OpenAI URLs (prepareBedrockCompactionRequest); other providers keep their original compaction transport. - Move Bedrock /anthropic strict-tool and metadata fitting into the amazon-bedrock and bedrock-mantle request hooks; anthropic.ts, stream.ts, and the registry are unchanged from main. - Recognize Bedrock FIPS, PrivateLink, and Mantle /v1 endpoints; resolve disableStrictTools for Bedrock /anthropic URLs in the catalog. - Move changelog entries out of the released 18.3.5 section.
…aping Replace the provider-id-scoped Bedrock /anthropic request hooks with a catalog compat flag. It is detected from a Bedrock /anthropic base URL under any provider id and can be set in models.yml to opt a proxy or an ANTHROPIC_BASE_URL reroute in, or to false to opt a Bedrock URL out. The Anthropic transport fits the payload only when the flag is set, and the on-demand compaction gate reads the same flag.
…ceholder x-api-key
…consumers Streamed text is not retracted, so ACP/RPC stream renderers keep the original. Also stop all extensions' handlers on abort and treat an undefined original textSignature as absent.
Only content was cloned, so in-place edits to usage and other fields leaked into the finalized message despite the documented detached copy. Also fix the docs example spacing.
…ge rewrite hook (@NaC-L) Expose finalized assistant message rewrite hook
- Priced openai-codex gpt-6.1-sol / gpt-6.1-sol-wm ($2/$10, $0.10 cached). - Added the `ultrafast` service tier and `/fast ultra`; Codex models receive it only when discovery lists it in `service_tiers` (Model.serviceTiers), and entering/leaving it starts a new websocket response chain, matching codex-rs. - Gated Codex priority/scale on the discovered tier list once known, and billed Codex cost from the tier the response reports.
- Restored the shouldSendServiceTier call in google-shared (narrowing guard only). - Telemetry records the request service tier only when it reaches the wire for the model. - Documented Ultrafast turns being costed at standard rates; changelog punctuation.
- Emit an OSC 1337 SetUserVar="omp_session_file" with the absolute session file path (base64) when TERM_PROGRAM=tern. - Introduce setTerminalSessionFileSource to supply a getter for the current session file to the reporter. - Call setTerminalSessionFileSource during InteractiveMode initialization so session changes trigger reporting. - Register reportTernSessionFile as a session-change callback, ensuring the variable updates on start, resume, and cwd switches. - Guard reporting to only occur in a Tern terminal, when stdout is a TTY, and not in headless mode. - Clear the session-file source and emit a removal of the variable on terminal title disposal.
- Removed the `env` property from the Bash tool schema and input interface, eliminating per-service environment customization. - Updated `startService` to always use the session shell environment and ignore any provided `env` object. - Revised the Bash tool prompt documentation to drop references to an `env` parameter. - Adjusted the test suite to remove `env` arguments and related expectations, confirming the new behavior.
… protocol (@H4vC) test(coding-agent): fix stubs broken by tern surface protocol
# Conflicts: # packages/coding-agent/src/modes/controllers/input-controller.ts
…ecall cache to their settings (@H4vC)
…ped tests that assert nothing
…confirm outside empty sessions (@H4vC) feat(coding-agent): opt-in bare slash commands with confirm outside empty sessions
…helpers, and dropped tests that assert nothing (@H4vC)
fix(coding-agent): bound bare-command confirmation to the session it was armed in
An empty service_tiers list (every model on free-plan accounts) now counts as not reported: priority/scale are dropped only when a non-empty list omits them. ultrafast still requires the list to name it.
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )