- Introduced a rejection interception mechanism to capture unhandled promise rejections from eval cell code.
- Attributed floating rejections to specific runs to fail the owning cell instead of crashing the process or worker.
- Downgraded rejections occurring after a cell finished to warn logs to prevent silent failures.
- Added `onTurnError` hook to `AdvisorRuntime` to handle failed turns before retries.
- Integrated credential blocking in `AgentSession` to prevent retrying usage-limited accounts when advisor turns fail.
- Included account key in `codex-auto-reset` debug logs to improve skip reason visibility.
artifactsDirsFromRegistry scanned only each registered agent's adopted (root-wide) ArtifactManager dir, but a subagent's own children are written one level deeper under its sessionFile-derived dir (task/index.ts). Any spawn chain 2+ levels deep therefore had its live, addressable output unresolvable via agent:// and the output() eval helper (Not found / Available: none). Collect both candidate dirs per ref; addDir dedup collapses the depth-0 case.
Fixes#4650
computeNonMessageTokens / computeNonMessageBreakdown re-tokenize the system
prompt and every tool's wire schema (per-tool JSON.stringify) on each call,
but the per-turn compaction and context-threshold paths call them several
times (getContextBreakdown twice, #estimateStoredContextTokens once) over
inputs that change at most once per turn. Memoize on the identity of
(systemPrompt, tools, skills) -- the same stable refs the StatusLineComponent
cache already trusts -- so the expensive parts run at most once per input
change instead of per call.
Synthesized project .omp/RULES.md as a distinct sticky rule name so capability dedup no longer shadows it behind the user sticky rule.
Added a public rules capability regression test covering user and project sticky RULES.md files loading together.
Fixes#4739
Left unresolved internal URLs unchanged during bash command expansion so quoted literal mentions can execute verbatim.
Skipped expansion for URL tokens embedded inside larger quoted shell text while preserving resolvable path-argument expansion.
Fixes#4737
- Replaced tool-based `set_title` invocation with XML-style `<title>` marker tags for session title discovery.
- Implemented robust JSON-unwrapping logic to handle and sanitize title generation outputs.
- Updated model registry in catalog with new model support, provider prefixes, and metadata adjustments.
- Synchronized system prompt documentation and test suites to reflect the new marker-based generation flow.
Filtered Bun-autoloaded launch .env.local entries out of child shell environments so nested commands can load their own dotenv files.
Added a regression test covering Convex-style inherited deployment variables while preserving ordinary inherited env values.
Fixes#4723
- Honored per-model architecture.input_modalities from llama.cpp /v1/models during discovery and selected model refresh.
- Added regression coverage for full refresh and cached selected-model metadata refresh.
- Updated the coding-agent changelog.
Fixes#4719
Read the Linux CPU model from /proc/cpuinfo during system prompt construction instead of calling os.cpus(), which can synchronously probe every logical CPU frequency file on many-core hosts.
Fixes#4712
can1357