fix(deps): update vstorm-co packages - autoclosed

[renovate]

This PR contains the following updates:

Package	Change	Age	Confidence
subagents-pydantic-ai (changelog)	>=0.2.9 → >=0.2.10
summarization-pydantic-ai	>=0.1.10 → >=0.1.11

---

Release Notes

vstorm-co/subagents-pydantic-ai (subagents-pydantic-ai)

v0.2.10

Compare Source

Fixed

• wait_tasks truncated results silently, making orchestrators re-delegate finished work (#​55). A completed task's result was hard-sliced to 2000 characters with no ellipsis, length, or marker of any kind. The orchestrator saw a well-formed answer that stopped mid-sentence, concluded the subagent had been cut off, and dispatched a new task asking it to finish — burning a full extra round-trip on work that was already complete and stored intact. Truncated results now end with an explicit marker stating that the cut is a display limit, that the stored answer is complete, and which check_task(...) call returns the full text; check_task and the wait_tasks tool description say the same, so the orchestrator knows the rule before it ever meets a cut result.

Added

• max_result_chars on create_subagent_toolset and SubAgentCapability (default 2000, matching the previous hard-coded limit). Sets the per-result character budget in the wait_tasks listing, so a fan-out of verbose subagents can't flood the orchestrator's context. Pass None to never truncate; a negative value raises ValueError.

vstorm-co/summarization-pydantic-ai (summarization-pydantic-ai)

v0.1.11

Compare Source

Fixed

• Compression no longer produces a history that providers reject (#​40, reported by @​drcrallen). execute_plan emitted the summary as a SystemPromptPart, so with keep=("messages", 0) — the ContextManagerCapability default — the rebuilt history was a single ModelRequest whose parts were all system parts. Anthropic and Google route system parts into a top-level system parameter instead of the message list, so that history mapped to zero provider messages: an IndexError in pydantic-ai's _apply_message_cache_control under anthropic_cache_messages, and a hard 400 without it. The failure was deterministic for any run long enough to cross the compression threshold. The summary is now carried by a UserPromptPart, so a compressed history always maps to at least one provider message on every provider.
• A zero keep no longer summarizes away the in-flight request. Compression runs in before_model_request on the full history, which includes the request that triggered the call. At keep=("messages", 0) that request — carrying the pending tool returns, or the new user prompt — was folded into the summary along with everything else, so the model was asked to act with the data (or the question) removed. find_safe_cutoff gained a keep_in_flight_request flag, set by SummarizationProcessor, that holds a trailing ModelRequest back and lets the existing tool-pair check pull its matching ToolCallPart across the cutoff with it. SlidingWindowProcessor leaves the flag off: with no summary to fall back on, a zero keep stays the no-op established in 0.1.10.
• Summaries no longer accumulate in the system channel. _extract_system_prompts carries every leading SystemPromptPart into the compressed message, so a summary stored as a system part was re-extracted on the next compression and kept next to the new one — one stale copy per compression, growing without bound for the life of the conversation. Keeping the summary in a UserPromptPart stops the extraction walk at the summary.

Changed

• The generated summary is now a UserPromptPart, not a SystemPromptPart. Code that reads the summary back out of a compressed history by scanning for SystemPromptPart needs to look for UserPromptPart instead. The carried-over system prompts are unaffected: they still lead the rebuilt ModelRequest.

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • At any time (no schedule defined)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

👻 Immortal: This PR will be recreated if closed unmerged. Get config help if that's undesired.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.

[coveralls]

Coverage Report for CI Build 30124256673

Coverage remained the same at 100.0%

Details

• Coverage remained the same as the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	6939
Covered Lines:	6939
Line Coverage:	100.0%
Coverage Strength:	1.0 hits per line

---

💛 - Coveralls