vtcode 0.137.0

A Rust-based terminal coding agent with modular architecture supporting multiple LLM providers
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160

//! Turn-snapshot capture and per-turn resolution.
//!
//! Captures [`TurnRequestSnapshot`] once per turn from `ctx`/`vt_cfg`
//! (provider name, prompt-cache shaping mode, active model, deferred-tool
//! policy, reasoning effort, etc.) so the rest of request assembly
//! (`prompt_assembly`, `tool_shaping`, `context_management`,
//! `response_chain`, and the `request_builder` orchestrator) consumes a
//! stable, already-resolved view of turn state instead of re-deriving it at
//! each call site. Invariant: the snapshot is read-only to its consumers;
//! nothing outside this module mutates a captured `TurnRequestSnapshot`.

use vtcode_core::ActivePrimaryAgent;
use vtcode_core::config::{OpenAIPromptCacheKeyMode, PromptCachingConfig};
use vtcode_core::core::agent::features::FeatureSet;
use vtcode_core::llm::provider::{self as uni};

use crate::agent::runloop::unified::incremental_system_prompt::PromptCacheShapingMode;
use crate::agent::runloop::unified::run_loop_context::TurnExecutionSnapshot;
use crate::agent::runloop::unified::session_setup::active_deferred_tool_policy;
use crate::agent::runloop::unified::turn::context::TurnProcessingContext;

/// Default turn timeout when no explicit configuration is set (seconds).
const DEFAULT_TURN_TIMEOUT_SECS: u64 = 300;

pub(super) fn is_openai_prompt_cache_enabled(
    provider_name: &str,
    global_prompt_cache_enabled: bool,
    openai_prompt_cache_enabled: bool,
) -> bool {
    provider_name.eq_ignore_ascii_case("openai") && global_prompt_cache_enabled && openai_prompt_cache_enabled
}

pub(super) fn resolve_prompt_cache_shaping_mode(
    provider_name: &str,
    prompt_cache: &PromptCachingConfig,
) -> PromptCacheShapingMode {
    debug_assert_eq!(provider_name, provider_name.to_ascii_lowercase());

    if !prompt_cache.cache_friendly_prompt_shaping || !prompt_cache.is_provider_enabled(provider_name) {
        return PromptCacheShapingMode::Disabled;
    }

    if matches!(provider_name, "anthropic" | "minimax") {
        PromptCacheShapingMode::AnthropicBlockRuntimeContext
    } else {
        PromptCacheShapingMode::TrailingRuntimeContext
    }
}

#[derive(Clone)]
pub(super) struct TurnRequestSnapshot {
    pub provider_name: String,
    pub planning_active: bool,
    pub full_auto: bool,
    pub auto_permission: bool,
    pub tool_free_recovery: bool,
    pub recovery_reason: Option<String>,
    pub request_user_input_enabled: bool,
    pub context_window_size: usize,
    pub turn_timeout_secs: u64,
    pub active_model: String,
    pub active_primary_agent: ActivePrimaryAgent,
    pub openai_prompt_cache_enabled: bool,
    pub openai_prompt_cache_key_mode: OpenAIPromptCacheKeyMode,
    pub prompt_cache_shaping_mode: PromptCacheShapingMode,
    pub capabilities: uni::ProviderCapabilities,
    pub execution: TurnExecutionSnapshot,
    /// Whether the active [`DeferredToolPolicy`] for this turn is
    /// client-local (`client_tool_search` enabled, no provider-hosted tool
    /// search available). When true, deferred tool definitions are omitted
    /// from the wire payload in [`build_turn_request`] and a `[Deferred
    /// Tools]` summary is appended to the system prompt in
    /// [`build_prompt_output`]. Always `false` for Anthropic/OpenAI hosted
    /// policies, which must keep every deferred tool on the wire.
    ///
    /// [`DeferredToolPolicy`]: vtcode_core::tools::handlers::DeferredToolPolicy
    /// [`build_turn_request`]: super::request_builder::build_turn_request
    /// [`build_prompt_output`]: super::prompt_assembly::build_prompt_output
    pub client_local_tool_deferral: bool,
}

pub(super) fn capture_turn_request_snapshot(
    ctx: &mut TurnProcessingContext<'_>,
    active_model: &str,
    tool_free_recovery: bool,
) -> TurnRequestSnapshot {
    let prompt_cache_config = &ctx.config.prompt_cache;
    let planning_active = ctx.is_planning_active();
    let auto_permission = ctx.full_auto && !planning_active;
    let provider_name = ctx.provider_client.name().to_ascii_lowercase();
    let openai_prompt_cache_enabled = is_openai_prompt_cache_enabled(
        &provider_name,
        prompt_cache_config.enabled,
        prompt_cache_config.providers.openai.enabled,
    );
    let prompt_cache_shaping_mode = resolve_prompt_cache_shaping_mode(&provider_name, prompt_cache_config);
    let request_user_input_enabled = FeatureSet::from_config(ctx.vt_cfg)
        .request_user_input_enabled(planning_active, ctx.renderer.supports_inline_ui())
        // After a permanent `request_user_input` denial (e.g. non-interactive
        // runtime), suppress the tool from the catalog so the model stops
        // seeing and retrying it. `mark_interview_denied()` is set from both
        // `handle_failure` and the permission-flow denial path.
        && !ctx.plan_session.is_interview_denied();
    let active_primary_agent = ctx.active_primary_agent.active().clone();
    let active_model = resolve_effective_request_model(active_model, &active_primary_agent);
    let context_window_size = ctx.provider_client.effective_context_size(&active_model);
    let turn_timeout_secs = ctx
        .vt_cfg
        .map(|cfg| cfg.optimization.agent_execution.max_execution_time_secs)
        .unwrap_or(DEFAULT_TURN_TIMEOUT_SECS);
    let openai_prompt_cache_key_mode = prompt_cache_config.providers.openai.prompt_cache_key_mode.clone();
    let full_auto = ctx.full_auto;
    let capabilities = uni::get_cached_capabilities(&**ctx.provider_client, &active_model);
    let client_local_tool_deferral =
        active_deferred_tool_policy(&*ctx.config, ctx.vt_cfg, &**ctx.provider_client).is_client_local();

    TurnRequestSnapshot {
        provider_name,
        planning_active,
        full_auto,
        auto_permission,
        tool_free_recovery,
        recovery_reason: ctx.recovery_reason().map(str::to_string),
        request_user_input_enabled,
        context_window_size,
        turn_timeout_secs,
        active_model,
        active_primary_agent,
        openai_prompt_cache_enabled,
        openai_prompt_cache_key_mode,
        prompt_cache_shaping_mode,
        capabilities,
        execution: ctx.harness_state.execution_snapshot(),
        client_local_tool_deferral,
    }
}

fn resolve_effective_request_model(base_model: &str, active_primary_agent: &ActivePrimaryAgent) -> String {
    active_primary_agent
        .model
        .as_deref()
        .map(str::trim)
        .filter(|model| !model.is_empty() && !model.eq_ignore_ascii_case("inherit"))
        .unwrap_or(base_model)
        .to_string()
}

pub(super) fn resolve_effective_reasoning_effort(
    cfg: Option<&vtcode_core::config::loader::VTCodeConfig>,
    turn_snapshot: &TurnRequestSnapshot,
) -> Option<vtcode_core::config::types::ReasoningEffortLevel> {
    if !turn_snapshot.capabilities.reasoning_effort || turn_snapshot.tool_free_recovery {
        return None;
    }

    turn_snapshot
        .active_primary_agent
        .reasoning_effort
        .or_else(|| cfg.map(|cfg| cfg.agent.reasoning_effort))
}