* fix(channels): dismiss tool feedback animation when turn ends via ResponseHandled
When a tool sets ResponseHandled=true (e.g., send_file), the turn ends
without producing a final assistant response. This meant no outbound
message triggered FinalizeToolFeedbackMessage, leaving the animation
goroutine running indefinitely — editing the Feishu card every 3 seconds
with "." / ".." suffixes long after the tool had finished.
Fix: call DismissToolFeedback at "Tool output satisfied delivery" so the
tracker is cleared and the animation goroutine is stopped immediately.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(adapters): add DismissToolFeedback to channelManagerAdapter
The adapter must implement the new interface method added in the
previous commit, otherwise the package fails to compile.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(channels): pass InboundContext to DismissToolFeedback for topic-aware keys
Telegram forum topics use scoped tracker keys like "chatID/topicID",
resolved via ToolFeedbackMessageChatID with the InboundContext. The
previous nil context caused the lookup to fall back to the raw chatID,
missing the topic-scoped entry and leaving the animation goroutine
orphaned in forum-topic conversations.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* style: wrap long function signatures for golines
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(feishu): fix image download with API fallback and post image support
- Add Image.Get API fallback when MessageResource.Get fails (different
permission scope: im:resource vs im:message:readonly)
- Extract and download images from post (rich text) messages
- Extract images from interactive card messages
- Deduplicate post image keys across locales
- Add comprehensive tests for new helpers
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(media): add image path tags alongside base64 for LLM file access
Images are still base64-encoded into msg.Media for multimodal LLMs,
but now also get [image:path] tags injected into message content so
the LLM knows the local file path for save/forward operations.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor(media): only auto-inject images for tool results, not user messages
Channel-received images (role=user) now get path tags only, letting
the LLM decide whether to view via load_image or just operate on
the file. Tool result images (role=tool, e.g. load_image) are
base64-encoded into a synthetic user message appended after the tool
message, since many LLM APIs don't support image_url in tool messages.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(media): preserve tool-message ordering for multi-tool-call scenarios
Move synthetic user message (carrying base64 tool images) to after the
entire contiguous tool-message block instead of immediately after each
tool message. This preserves the assistant→tool→tool ordering required
by OpenAI-compatible APIs.
Also fix load_image to use generic [image: photo] placeholder so
injectPathTags can properly replace it with the actual path.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(test): update load_image test for [image: photo] placeholder
The test was checking ForLLM for the media:// ref, but load_image now
emits the generic [image: photo] placeholder instead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(media): match all channel image placeholders in injectPathTags
Different channels emit different placeholder formats — Telegram/Feishu
use [image: photo], WeCom/WeChat/Line use bare [image], QQ/Discord use
[image: <filename>]. The previous string-match code only handled
[image: photo], so for the other channels the path tag was appended as
a duplicate, producing content like "[image] [image:/path]".
Switch to per-type regex that matches all generic placeholder shapes
while leaving path tags ([image:/path]) untouched. Also fixes the same
issue for [audio], [video], [file] tags. Added test coverage for the
various placeholder shapes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(media): skip path tag append for JSON content (Feishu cards/posts)
When content is structured JSON (interactive cards, post messages),
injectPathTags now skips the fallback append — only placeholder
replacement is attempted. This prevents corrupting JSON payloads
like {"schema":"2.0",...} with appended [image:/path] tags.
Adds looksLikeJSON() helper and three test cases covering JSON
objects, arrays, and an end-to-end resolveMediaRefs scenario with
Feishu card content.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(media): prepend path tags for JSON content, narrow looksLikeJSON
Two fixes from code review:
1. looksLikeJSON now only checks for '{' prefix (not '['), avoiding
false positives on regular text like "[update] see attached".
2. For JSON content (Feishu cards/posts), path tags are prepended
before the JSON instead of being silently dropped. This ensures
the LLM can discover attached images via the path tag while the
JSON payload stays valid for downstream parsing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Add detection for 'unknown variant' + 'image_url' error pattern used by
DeepSeek and other strict providers when vision is not supported.
These providers reject the image_url field at the JSON schema level
rather than returning a semantic 'not supported' message.
- Use fmt.Sprintf fallback instead of {} on encoding errors
- Normalize nil args to {} in FormatArgsJSON for consistent output
- Update tests to expect {} instead of null for nil args
Based on PR #2670 review feedback from afjcjsbx
Add a non-blocking runtime publish path and switch hot-path publishers to it.
Enforce subscription timeout boundaries, keep ordered subscriber snapshots up to date on subscribe changes, expose all runtime kinds to process hooks, add safe log attrs for non-agent events, and close the gateway message bus on full shutdown.
Replace leftover SubTurnOrphanResultEvent and short subturn event references with runtime event kinds in comments, tests, and hook design notes.
Validation: GOCACHE=/tmp/picoclaw-go-cache go test ./pkg/agent -run TestSpawnSubTurn_OrphanResultRouting; make lint
Remove the legacy EventKind/Event envelope mapping and let agent event emission build pkg/events.Event values directly.
Keep HookMeta as the shared hook metadata shape and preserve legacy observe string aliases by mapping them to runtime event kinds.
Validation: GOCACHE=/tmp/picoclaw-go-cache go test ./pkg/agent; make lint
Drop the old agent EventBus, SubscribeEvents/EventDrops public surface, legacy hook observer dispatch, and hook.event process notification path. Agent observations now flow through pkg/events runtime events.
Validation: go test ./pkg/agent; make lint
Move agent domain event payload structs out of the legacy event envelope file so the remaining EventKind/Event/EventMeta compatibility layer can be removed independently later.
Validation: go test ./pkg/agent; make lint
Use RuntimeEventObserver for the normal in-process hook observer path and make the process-hook helper assert hook.runtime_event notifications.
Validation: go test ./pkg/agent; make lint
Move AgentLoop event assertions to the runtime event stream and keep the legacy SubscribeEvents test only for dual-publish compatibility.
Validation: go test ./pkg/agent; make lint
Deprecate the legacy agent event APIs and add a runtime event test helper, then migrate the follow-up queued test to the runtime event stream.
Validation: go test ./pkg/agent; make lint
Migrate hook observation to runtime events and update the process hook notification protocol. Add runtime event publication for message bus failures, channel lifecycle/outbound flow, gateway reloads, MCP server state, and MCP tool calls.
Validation: go test ./pkg/events/... ./pkg/bus ./pkg/agent ./pkg/channels ./pkg/mcp ./pkg/tools/integration ./pkg/gateway; make lint
Introduce pkg/events with filtered channels, subscription policies, backpressure, and stats. Wire AgentLoop to dual-publish legacy agent events into runtime events while preserving old event APIs.
Validation: go test ./pkg/events/... ./pkg/agent; go test -race ./pkg/events/...; make lint
- Add PrettyPrint and DisableEscapeHTML config options to ToolFeedbackConfig
- Add FormatArgsJSON helper function with configurable pretty printing and HTML escaping
- Add toolFeedbackArgsPreviewWithOptions to pass formatting options
- Update pipeline_execute.go to use new formatting options for tool feedback
This fixes the issue where '&&' would be displayed as '\u0026' in tool
feedback messages and provides optional pretty-printing for better
readability.
- Add isNetworkError detection for connection reset, broken pipe, read/write tcp, EOF
- Add retry logic with configurable exponential backoff for network errors
- Add config options max_llm_retries and llm_retry_backoff_secs in agents.defaults
- Network errors now retry with backoff (was previously not retried)
- Timeout errors now use configurable backoff instead of hardcoded 5s
- Default: 2 retries with 2s backoff (3 total attempts)
Resolves conflicts after the agent loop refactor on main:
- pkg/agent/loop.go was deleted upstream (logic split into agent.go,
agent_init.go, pipeline.go, etc.); accepted the deletion.
- Moved the delegate tool registration block from the old loop.go
into registerSharedTools in pkg/agent/agent_init.go, immediately
after the spawn/spawn_status block. Logic and gating
(len(registry.ListAgentIDs()) > 1) are unchanged.
- pkg/agent/subturn.go and pkg/agent/subturn_test.go merged cleanly
on their own; TargetAgentID field, validation, registry lookup,
and tests all preserved.
Verified locally:
- go build ./pkg/agent/... ./pkg/tools/... clean
- go vet clean
- TestDelegateTool* (17 cases) pass
- TestSpawnSubTurn_TargetAgentID_* (3 cases) pass
- TestDelegateToolRegistered_MultiAgent / _SingleAgent pass
- full pkg/agent + pkg/tools test suites green
- centralize web search provider readiness and resolution logic
- fall back when the configured provider is unavailable or invalid
- allow native-search-capable models to use built-in search without the client tool
- simplify the tools page and add direct access to web search settings
- add backend, agent, and integration tests for the new selection behavior
* refactor: support explicit model list providers
* fix(web): preserve explicit model providers
* fix(web): preserve legacy provider prefixes on model updates
fix(models): normalize explicit provider-prefixed ids
fix(api): preserve legacy model updates across providers
fix(agent): preserve config identity for explicit provider refs
* fix ci
* feat(web): download attachments in frontend
* fix: proxy pico media and force svg downloads
* feat(web): hide ephemeral media refs from persisted session history
Add a context window usage indicator to the web chat UI and a /context
slash command that works across all channels.
Backend:
- Add computeContextUsage() estimating history + system + tool tokens
- Attach ContextUsage to outbound messages via the pico WebSocket protocol
- Add /context command showing context stats as formatted text
- Add EstimateSystemTokens() on ContextBuilder for system prompt estimation
Frontend:
- Add ContextUsageRing component (SVG ring + hover/tap popover)
- Show usage percentage, token counts, and compression threshold
- Hover on desktop (150ms leave delay), tap on mobile
- "View Details" sends /context with 1s cooldown
- i18n support (en/zh) for popover labels
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Replace generic mockProvider with modelRecordingProvider that captures
the model parameter passed to Chat(). After delegation from alpha to
beta, assert the recorded model is "model-beta" — proving the child
turn actually ran with the target agent's configuration, not the
caller's.
Also add wiring tests:
- TestDelegateToolNotRegistered_SingleAgent: single-agent has no
delegate in its tool registry
- TestDelegateToolRegistered_MultiAgent: both agents in a two-agent
setup have the delegate tool
Ref: #2148
Remove the IsToolEnabled("delegate") check — there is no "delegate"
entry in ToolsConfig, so the check was always true. The only real
gate is len(agents) > 1, which is the intended behavior: delegate
is auto-registered in multi-agent setups.
Ref: #2148
Register the delegate tool in registerSharedTools when multiple agents
are configured. Gated independently from the subagent tool — delegate
uses SubTurn directly and does not depend on SubagentManager.
Self-delegation is prevented by injecting the current agent ID.
Permission is enforced via CanSpawnSubagent (reuses allow_agents config).
Single-agent setups are unaffected: the tool is not registered when
only one agent exists in the registry.
Ref: #2148
Add multi-agent test setup (newMultiAgentLoop) with two agents using
distinct models (model-alpha, model-beta).
Three new tests:
- UsesTargetAgent: parent=alpha delegates to beta, event log confirms
child runs as agent_id=beta with model=model-beta
- NotFound: TargetAgentID pointing to nonexistent agent returns error
- EmptyModelAccepted: empty Model field accepted when TargetAgentID
provides the model implicitly
Ref: #2148
When TargetAgentID is set, spawnSubTurn resolves the target AgentInstance
from the registry and uses it as the base for the child turn. This gives
the child turn the target's workspace, model, tools, and system prompt
instead of inheriting from the caller.
Model validation is relaxed: empty Model is accepted when TargetAgentID
provides the model implicitly via the resolved agent instance.
Ref: #2148
Scope tool result deduplication to each assistant tool-call block so providers
that reuse call IDs across separate turns do not lose valid tool results. Also
drop invalid empty tool call IDs and orphaned tool messages after validation.
- Add Clear(ctx, sessionKey) to ContextManager interface
- Implement Clear for legacy (JSONL) and seahorse (DB + JSONL)
- Add Engine.ClearSession + Store.ClearConversation
- Fix FTS5 DELETE trigger syntax in schema (was using wrong
external-content FTS5 syntax; now uses standard DELETE FROM)
- Fix ClearSession to skip sessions never ingested (was creating
blank conversations record via GetOrCreateConversation)
- Simplify summary_parents DELETE into single OR statement
- Add TestStoreClearConversation unit test
When the message tool sent to a different chat (e.g., a group), the
agent's final response to the originating chat was incorrectly skipped
because HasSentInRound() was a simple bool that didn't distinguish
targets. Replace with HasSentTo(channel, chatID) that tracks all
send targets per round and only suppresses when the target matches.
Fixes cross-conversation message causing "Processing..." to hang.
* * completed
* * optimzie
* * fix format
* * fix pr check
* try to fix ci
* * Indicates that Windows does not support expos_paths, adding more mount paths for the Linux platform.
* fix isolation startup lifecycle and MCP transport wrapping
* fix isolation startup cleanup and optional Linux mounts
* fix isolation path handling for relative hooks
Preserve relative command and working-directory semantics when Linux isolation wraps subprocesses, and restore absolute argv path exposure to avoid startup regressions. Add hook coverage and docs updates so isolation-enabled process hooks keep working as configured.
* * fix ci
* fix(gateway): validate PID ownership and clean stale pid files
- include `pid` in health responses for runtime PID verification
- add `RemovePidFileIfPID` to safely delete PID files only on PID match
- sanitize gateway PID data via process-command checks with health fallback
- ignore and remove stale/non-gateway PID files before gateway operations
- refuse stop/restart actions when the attached process is not a gateway
- update gateway and websocket tests to cover PID validation and safety paths
* test(seahorse): use shared in-memory SQLite DB in tests to fix async compaction failures
* test: remove unused sendMediaErr field from hook test mock
* feat(hooks): add respond action for tool execution bypass
Add a new HookActionRespond that allows hooks to return tool results directly, skipping actual tool execution. This enables plugin tool injection, caching, and mocking capabilities.
- Add HookActionRespond constant and support in HookManager
- Extend ToolCallHookRequest with HookResult field
- Implement respond action handling in process hooks and agent loop
- Add comprehensive tests for respond and deny_tool actions
- Update documentation with hook actions table and examples
* docs(hooks): add JSON-RPC protocol and plugin tool injection documentation
Add comprehensive documentation for hook JSON-RPC protocol and plugin tool injection capabilities:
- Add "Hook Actions" section to README.zh.md explaining respond action for tool execution bypass
- Create hook-json-protocol.md/.zh.md detailing JSON-RPC 2.0 protocol for all hook methods
- Create plugin-tool-injection.md/.zh.md with complete examples for external tool implementation
- Document how hooks can inject tool definitions and return results via respond action
- Include Python and Go examples for weather query plugin implementation
* feat(agent): emit tool events and feedback for hook results
Add ToolExecStart event emission and tool feedback for hook results to ensure consistent behavior between normal tool execution and hook bypass scenarios. This maintains parity in event tracking and user feedback when tools are executed via hooks.
* style(agent): format whitespace in hook structs and constants
Remove trailing whitespace and standardize spacing in JSON struct tags, constants, and test data for improved code consistency.
* feat(hooks): add media support for plugin tool injection
Extend the hook respond action to support media file handling:
- Add `media` field for returning images and files from hooks
- Add `response_handled` field to control turn completion behavior
- When response_handled=true, media is automatically delivered to user
- When response_handled=false, media is passed to LLM for vision requests
This enables plugins to directly return generated images, downloaded
files, and other media content either to users or for LLM analysis.
* docs(hooks): document security implications of respond action
Add security boundary documentation explaining that the respond action
bypasses ApproveTool checks, allowing hooks to return results for any
tool without approval. Include recommendations for secure hook
implementation and code comments marking the security considerations.
Changes:
- Add "Security Boundaries" section to plugin-tool-injection docs
- Document bypass of approval checks and associated risks
- Provide security recommendations and example code
- Add inline security comments in hooks.go and loop.go
* refactor(agent): improve completeness of tool result cloning and hook processing
Extend cloneToolResult to properly copy ArtifactTags and Messages fields,
ensuring deep copies of all ToolResult data. Consolidate event emission
and user message handling to match the normal tool execution flow.
* fix(agent): align hook respond path with normal tool execution flow
The hook respond code path was missing several critical behaviors that
existed in normal tool execution:
- Add logging for tool calls with arguments preview
- Add is_tool_call metadata to user-facing messages
- Handle attachment delivery failures by setting error state and
notifying LLM
- Set ResponseHandled=false when using bus for media delivery
- Check for steering messages and graceful interrupts after tool
execution, skipping remaining tools when appropriate
- Poll for SubTurn results that arrived during tool execution
This ensures consistent behavior between hook-responded tool calls and
normally executed tool calls.
* test(agent): add tests for hook respond media error handling
Add comprehensive tests for the hook respond code path when media
delivery fails. Tests cover error media channel scenarios and verify
proper error state handling.
Also document that AfterTool is not called when using respond action,
as it provides the final answer directly (design decision).
* fix(agent): disable seahorse context manager on freebsd/arm
Exclude freebsd/arm from the seahorse-enabled build and route it to the
unsupported stub implementation.
This avoids freebsd/arm build failures caused by modernc sqlite/libc while
keeping picoclaw buildable on that target.
* build: bump Go version from 1.25.8 to 1.25.9
* ci: install and run govulncheck directly in PR workflow
* fix: use per-candidate provider for model_fallbacks
Each fallback model now uses its own api_base and api_key from
model_list instead of inheriting the primary model's provider config.
Previously, a single LLMProvider was created from the primary model's
ModelConfig and reused for all fallback candidates — only the model ID
string was swapped. This caused all fallback requests to be routed to
the primary provider's endpoint, making cross-provider fallback chains
non-functional (e.g., OpenRouter primary with Gemini fallback would
send the Gemini request to OpenRouter's API).
Fix: pre-create a per-candidate LLMProvider at agent initialization
time by looking up each candidate's ModelConfig from model_list. The
fallback run closure now selects the correct provider per candidate
via CandidateProviders map, falling back to agent.Provider when no
override is found.
Fixes#2140
Made-with: Cursor
test: add test for instance.go
fix: fix test
refactor: optimize
fix: fix Golang lint issues
chore: comment cleanup
* refactor: use resolvedModelConfig() instead of buildModelIndex()
* fix
- add build tags to exclude context_seahorse.go on mipsle and netbsd
- add context_seahorse_unsupported.go to keep registration and return a clear runtime error
- remove unused indirect dependency github.com/reiver/go-porterstemmer from go.mod and go.sum
* feat(seahorse): implement short-term memory engine of seahorse
Add pkg/seahorse/ module implementing a SQLite-backed DAG-based summary
hierarchy for context management, ported from lossless-claw's LCM design:
- types.go + short_constants.go: core types (Message, Summary, Conversation,
ContextItem) and configuration constants (fanout, token targets, thresholds)
- migration.go: idempotent DB schema with FTS5 trigram tokenizer for CJK
- store.go: full SQLite CRUD (conversations, messages, summaries DAG,
context_items with ordinal gap numbering, FTS5 search)
- short_engine.go: Engine lifecycle (NewEngine, Ingest, Assemble, Compact),
session pattern filtering (ignore/stateless glob→regex compilation),
per-session mutex via sync.Map
- short_assembler.go: budget-aware context assembly with fresh tail protection
(32 messages), oldest-first eviction, summary XML formatting, RebuildContextItems
- short_compaction.go: leaf compaction (messages→summary) and condensed
compaction (summaries→higher-level summary), 3-level LLM escalation,
CompactUntilUnder for emergency overflow
- short_retrieval.go: lookupByID, FTS5/LIKE search, recursive expand with
token cap
- context_seahorse.go: agent.ContextManager adapter, registered as "seahorse",
provider↔seahorse message type conversion (ToolCalls, tool_result)
* fix(seahorse): correct 3 adapter bugs in context management
- TokenCount: use full message (Content+ToolCalls+Media) instead of Content-only
- Empty Content: rebuild Content from tool_result Parts when stored empty
- Duplicate summaries: summaries only in Summary field, not in History messages
- Grep: fix SearchResult.Snippet→Content for summaries
- Schema: fix FTS5 SQL uses VIRTUAL TABLE not TEMP TABLE
- TestFTS5SQLConstants: verify FTS5 SQL syntax correctness
- Test: fix flaky TestCompactLeaf
* fix(agent): ingest steering messages into seahorse SQLite
Steering messages were only persisted to session JSONL but not ingested
into seahorse SQLite, causing them to be missing from context assembly.
Added `ts.ingestMessage(turnCtx, al, pm)` call in the steering message
injection block alongside the existing JSONL persistence.
Test: TestSeahorseSteeringMessageIngested verifies steering messages
appear in seahorse SQLite DB after being processed.
* fix(seahorse): address 3 blocking bugs from code review
- Fix resequenceContextItemsTx scan error handling (store.go:850)
Changed `return err` to `return scanErr` to properly propagate scan errors
instead of returning nil (which silently corrupts data)
- Fix sql.NullString for INTEGER column (store.go:847)
Changed `mid` from sql.NullString to sql.NullInt64 since message_id
is INTEGER in schema. Removed unnecessary strconv.ParseInt call.
- Fix compactCondensed fallback deleting non-candidate items
Added ReplaceContextItemsWithSummary method for per-item deletion
when candidates are not contiguous in ordinal space.
Optimized to use range deletion when candidates are consecutive.
* fix(seahorse): pass Budget to Compact for correct condensed threshold
Issue #4 from PR review: When Budget was not passed to seahorse.Compact,
it defaulted to `tokensBefore * 0.75`, making `tokensBefore > budget`
always true and causing condensed compaction to trigger unnecessarily.
Changes:
- context_seahorse.go: Forward Budget from CompactRequest to CompactInput
- loop.go: Pass Budget (ContextWindow) in all 3 Compact calls
- Add test verifying condensed is skipped when tokens < threshold
- Fix lint issues in store.go and store_test.go
* fix(seahorse): add mutex for assembler lazy initialization
Issue #5 from PR review: The check-then-create pattern for e.assembler
was a data race when multiple goroutines called Assemble() concurrently:
if e.assembler == nil {
e.assembler = &Assembler{...}
}
Changes:
- Add assemblerMu sync.Mutex to Engine struct
- Add initAssemblerOnce() using double-checked locking (same pattern as initCompactionOnce)
- Add TestAssemblerLazyInitRace to verify thread-safety
* fix(seahorse): handle non-consecutive depths in selectShallowestCondensationCandidate
Issue #8 from PR review: the loop iterated depth 0, 1, 2... assuming
consecutive keys, but break when key was missing caused deeper depths
to never be checked.
Fix: collect all existing depth keys, sort, then iterate in order.
* fix(seahorse): wrap DeleteMessagesAfterID and appendContextItems in transactions
- DeleteMessagesAfterID: wrap all DELETE operations in a transaction for
atomicity, remove redundant manual FTS delete (handled by trigger)
- appendContextItems: use transaction to fix read-then-write race condition
- Add GetMaxOrdinalTx and resolveItemTokenCountTx for transaction-scoped queries
- Remove unused resolveItemTokenCount function
Fixes PR review issues 6 and 7.
* fix(seahorse): derive readable content from Parts and cap CompactUntilUnder iterations
- Derive readable content from MessageParts in AddMessageWithParts so
FTS5 indexing and summary formatting can access tool call information
- formatMessagesForSummary and truncateSummary now fall back to Parts
when Content is empty, fixing blank summaries for Part-based messages
- Add MaxCompactIterations (20) to prevent CompactUntilUnder infinite
loops; exceeded iterations are logged as warnings
* feat(mcp): store oversized text results as artifacts
* feat(mcp): fix doc
* fix(mcp): preserve raw MCP payload in text artifacts
* fix(mcp): avoid leaking large text when artifact persistence fails
* chore(mcp): clarify inline text limit and cover artifact edge cases
