Struct TaskStateInput 

pub struct TaskStateInput {

    pub user_content: String,
    pub intents: Vec<String>,
    pub authority: String,
    pub retrieval_metrics: Option<RetrievalMetrics>,
    pub tool_search_stats: Option<ToolSearchStats>,
    pub mcp_tools_available: bool,
    pub taskable_agent_count: usize,
    pub fit_agent_count: usize,
    pub fit_agent_names: Vec<String>,
    pub enabled_skill_count: usize,
    pub matching_skill_count: usize,
    pub missing_skills: Vec<String>,
    pub remaining_budget_tokens: usize,
    pub provider_breaker_open: bool,
    pub inference_mode: String,
    pub decomposition_proposal: Option<DecompositionProposal>,
    pub explicit_specialist_workflow: bool,
    pub named_tool_match: bool,
    pub recent_response_skeletons: Vec<String>,
    pub recent_user_message_lengths: Vec<usize>,
    pub self_echo_fragments: Vec<String>,
    pub declared_action: Option<DeclaredAction>,
    pub previous_turn_had_protocol_issues: bool,
    pub normalization_retry_streak: u8,
}

Raw facts assembled by roboticus-api from pipeline subsystems. This is the sole input to synthesize — the API layer collects these values but does NOT interpret them.

Fields

user_content: String

The user’s message text.

intents: Vec<String>

Serialized intent names (from IntentRegistry).

authority: String

Serialized InputAuthority variant name.

retrieval_metrics: Option<RetrievalMetrics>

tool_search_stats: Option<ToolSearchStats>

mcp_tools_available: bool

Whether any MCP-sourced tools are available.

taskable_agent_count: usize

Enabled, non-proxy subagents.

fit_agent_count: usize

Agents whose skills overlap with the task.

fit_agent_names: Vec<String>

Names of fitting agents.

enabled_skill_count: usize

matching_skill_count: usize

Skills whose triggers match the user input.

missing_skills: Vec<String>

Skills that would be useful but are not registered.

remaining_budget_tokens: usize

Tokens remaining after system prompt.

provider_breaker_open: bool

Whether any LLM provider circuit breaker is open.

inference_mode: String

“standard” or “streaming”.

decomposition_proposal: Option<DecompositionProposal>

explicit_specialist_workflow: bool

Whether the user explicitly requested specialist/delegation workflow.

named_tool_match: bool

Whether the user’s message references an existing registered tool by name. When true, the named capability already exists in the tool registry and will be available during inference — the planner should NOT propose specialist creation for it.

recent_response_skeletons: Vec<String>

Structural fingerprints of the last N assistant responses. Each string is a compact skeleton like “narrative+question+options”.

recent_user_message_lengths: Vec<usize>

Word counts of the last N user messages (engagement signal).

self_echo_fragments: Vec<String>

Notable phrases (8+ words) extracted from recent assistant turns. Used to detect self-echoing — the agent reusing its own prior phrasing.

declared_action: Option<DeclaredAction>

Whether the user’s input contains a declared physical action (verb + target pattern, e.g. “I attack the goblin”).

previous_turn_had_protocol_issues: bool

Whether the previous assistant turn contained protocol normalization issues (malformed tool calls, narrated next steps, unexecuted streaming markers). When true, the planner may select NormalizationRetry.

normalization_retry_streak: u8

Number of consecutive turns with protocol normalization issues. Zero when previous_turn_had_protocol_issues is false; allows the planner to escalate retry strategy for persistent normalization failures.

Trait Implementations

impl Clone for TaskStateInput

fn clone(&self) -> TaskStateInput

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TaskStateInput

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.