Skip to main content

ironflow_core/
provider.rs

1//! Provider trait and configuration types for agent invocations.
2//!
3//! The [`AgentProvider`] trait is the primary extension point in ironflow: implement it
4//! to plug in any AI backend (local model, HTTP API, mock, etc.) without changing
5//! your workflow code.
6//!
7//! Built-in implementations:
8//!
9//! * [`ClaudeCodeProvider`](crate::providers::claude::ClaudeCodeProvider) - local `claude` CLI.
10//! * `SshProvider` - remote via SSH (requires `transport-ssh` feature).
11//! * `DockerProvider` - Docker container (requires `transport-docker` feature).
12//! * `K8sEphemeralProvider` - ephemeral K8s pod (requires `transport-k8s` feature).
13//! * `K8sPersistentProvider` - persistent K8s pod (requires `transport-k8s` feature).
14//! * [`RecordReplayProvider`](crate::providers::record_replay::RecordReplayProvider) -
15//!   records and replays fixtures for deterministic testing.
16
17use std::collections::BTreeMap;
18use std::fmt;
19use std::future::Future;
20use std::marker::PhantomData;
21use std::pin::Pin;
22use std::sync::Arc;
23
24use schemars::JsonSchema;
25use serde::{Deserialize, Serialize};
26use serde_json::Value;
27
28use crate::error::AgentError;
29use crate::operations::agent::{Model, PermissionMode};
30
31/// Boxed future returned by [`AgentProvider::invoke`].
32pub type InvokeFuture<'a> =
33    Pin<Box<dyn Future<Output = Result<AgentOutput, AgentError>> + Send + 'a>>;
34
35// ── Typestate markers ──────────────────────────────────────────────
36
37/// Marker: no tools have been added via the builder.
38#[derive(Debug, Clone, Copy)]
39pub struct NoTools;
40
41/// Marker: at least one tool has been added via [`AgentConfig::allow_tool`].
42#[derive(Debug, Clone, Copy)]
43pub struct WithTools;
44
45/// Marker: no JSON schema has been set via the builder.
46#[derive(Debug, Clone, Copy)]
47pub struct NoSchema;
48
49/// Marker: a JSON schema has been set via [`AgentConfig::output`] or
50/// [`AgentConfig::output_schema_raw`].
51#[derive(Debug, Clone, Copy)]
52pub struct WithSchema;
53
54// ── AgentInput ─────────────────────────────────────────────────────
55
56/// Declarative external input fetched into the agent's filesystem before invocation.
57///
58/// Each input is a URL that the provider must download and materialize at
59/// `mount_path` so the agent can read it via the `Read` tool.
60///
61/// Provider behavior:
62///
63/// * [`ClaudeCodeProvider`](crate::providers::claude::ClaudeCodeProvider) (local) -
64///   downloads via reqwest into a per-invocation temp directory and rewrites
65///   `mount_path` to the resolved local path.
66/// * `K8sEphemeralProvider` - injects a `curlimages/curl` initContainer that
67///   downloads each URL into a shared `emptyDir`, mounted on the main container
68///   at the parent directory of `mount_path`.
69///
70/// The `mount_path` must be an absolute path. Intermediate directories are
71/// created automatically.
72#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
73pub struct AgentInput {
74    /// Source URL to download (HTTP/HTTPS, including signed S3/R2 URLs).
75    pub url: String,
76
77    /// Absolute filesystem path where the file must be available inside the
78    /// agent's filesystem.
79    pub mount_path: String,
80}
81
82impl AgentInput {
83    /// Create a new input descriptor.
84    pub fn new(url: &str, mount_path: &str) -> Self {
85        Self {
86            url: url.to_string(),
87            mount_path: mount_path.to_string(),
88        }
89    }
90}
91
92// ── AgentConfig ────────────────────────────────────────────────────
93
94/// Serializable configuration passed to an [`AgentProvider`] for a single invocation.
95///
96/// Built by [`Agent::run`](crate::operations::agent::Agent::run) from the builder state.
97/// Provider implementations translate these fields into whatever format the underlying
98/// backend expects.
99///
100/// # Typestate: tools vs structured output
101///
102/// Claude CLI has a [known bug](https://github.com/anthropics/claude-code/issues/18536)
103/// where combining `--json-schema` with `--allowedTools` always returns
104/// `structured_output: null`. To prevent this at compile time, [`allow_tool`](Self::allow_tool)
105/// and [`output`](Self::output) / [`output_schema_raw`](Self::output_schema_raw) are mutually
106/// exclusive: using one removes the other from the available API.
107///
108/// ```
109/// use ironflow_core::provider::AgentConfig;
110///
111/// // OK: tools only
112/// let _ = AgentConfig::new("search").allow_tool("WebSearch");
113///
114/// // OK: structured output only
115/// let _ = AgentConfig::new("classify").output_schema_raw(r#"{"type":"object"}"#);
116/// ```
117///
118/// ```compile_fail
119/// use ironflow_core::provider::AgentConfig;
120/// // COMPILE ERROR: cannot add tools after setting structured output
121/// let _ = AgentConfig::new("x").output_schema_raw("{}").allow_tool("Read");
122/// ```
123///
124/// ```compile_fail
125/// use ironflow_core::provider::AgentConfig;
126/// // COMPILE ERROR: cannot set structured output after adding tools
127/// let _ = AgentConfig::new("x").allow_tool("Read").output_schema_raw("{}");
128/// ```
129///
130/// **Workaround**: split the work into two steps -- one agent with tools to
131/// gather data, then a second agent with `.output::<T>()` to structure the result.
132#[derive(Debug, Clone, Serialize, Deserialize)]
133#[serde(bound(serialize = "", deserialize = ""))]
134#[non_exhaustive]
135pub struct AgentConfig<Tools = NoTools, Schema = NoSchema> {
136    /// Optional system prompt that sets the agent's persona or constraints.
137    pub system_prompt: Option<String>,
138
139    /// The user prompt - the main instruction to the agent.
140    pub prompt: String,
141
142    /// Which model to use for this invocation.
143    ///
144    /// Accepts any string. Use [`Model`] constants for well-known Claude models
145    /// (e.g. `Model::SONNET`), or pass a custom identifier for other providers.
146    #[serde(default = "default_model")]
147    pub model: String,
148
149    /// Allowlist of tool names the agent may invoke (empty = provider default).
150    #[serde(default)]
151    pub allowed_tools: Vec<String>,
152
153    /// Denylist of tool names the agent MUST NOT invoke.
154    ///
155    /// Maps to `--disallowedTools` on the Claude CLI. Unlike
156    /// [`allowed_tools`](Self::allowed_tools), this does **not** activate any
157    /// tools; it only filters out tools that would otherwise be loaded by
158    /// default. As such, it is safe to combine with structured output
159    /// ([`output`](Self::output)) without triggering the Claude CLI bug that
160    /// affects `--json-schema` + `--allowedTools`.
161    #[serde(default)]
162    pub disallowed_tools: Vec<String>,
163
164    /// Maximum number of agentic turns before the provider should stop.
165    pub max_turns: Option<u32>,
166
167    /// Maximum spend in USD for this single invocation.
168    pub max_budget_usd: Option<f64>,
169
170    /// Working directory for the agent process.
171    pub working_dir: Option<String>,
172
173    /// Path to an MCP server configuration file.
174    pub mcp_config: Option<String>,
175
176    /// When `true`, pass `--strict-mcp-config` to the Claude CLI so it only
177    /// loads MCP servers from [`mcp_config`](Self::mcp_config) and ignores
178    /// any global/user MCP configuration (e.g. `~/.claude.json`).
179    ///
180    /// Useful to prevent global MCP servers from leaking tools into steps
181    /// that request `structured_output`, which triggers the Claude CLI bug
182    /// where `--json-schema` combined with any active tool returns
183    /// `structured_output: null`. See
184    /// <https://github.com/anthropics/claude-code/issues/18536>.
185    ///
186    /// Combine with `mcp_config` set to a file containing
187    /// `{"mcpServers":{}}` to disable every MCP server for the invocation.
188    #[serde(default)]
189    pub strict_mcp_config: bool,
190
191    /// When `true`, pass `--bare` to Claude CLI. Bare mode disables:
192    /// - auto-memory (automatic creation of `~/.claude/.../memory/*.md` files)
193    /// - `CLAUDE.md` auto-discovery (no global/project `CLAUDE.md` loaded)
194    /// - hooks, LSP, plugin sync, attribution, background prefetches
195    ///
196    /// Recommended for orchestrator agents that should not have any implicit
197    /// side effects on the user's filesystem or inherit user-level context.
198    ///
199    /// # Authentication requirement
200    ///
201    /// `--bare` is **only compatible with an Anthropic API key**
202    /// (`ANTHROPIC_API_KEY` environment variable). It does **not** work with
203    /// OAuth authentication (`claude /login` / keychain-stored credentials),
204    /// because bare mode disables keychain reads.
205    #[serde(default)]
206    pub bare: bool,
207
208    /// Permission mode controlling how the agent handles tool-use approvals.
209    #[serde(default)]
210    pub permission_mode: PermissionMode,
211
212    /// Optional JSON Schema string. When set, the provider should request
213    /// structured (typed) output from the model.
214    #[serde(alias = "output_schema")]
215    pub json_schema: Option<String>,
216
217    /// Optional session ID to resume a previous conversation.
218    ///
219    /// When set, the provider should continue the conversation from the
220    /// specified session rather than starting a new one.
221    pub resume_session_id: Option<String>,
222
223    /// Enable verbose/debug mode to capture the full conversation trace.
224    ///
225    /// When `true`, the provider uses streaming output (`stream-json`) to
226    /// record every assistant message and tool call. The resulting
227    /// [`AgentOutput::debug_messages`] field will contain the conversation
228    /// trace for inspection.
229    #[serde(default)]
230    pub verbose: bool,
231
232    /// Custom labels applied to the pod (K8s providers only).
233    ///
234    /// Non-K8s providers ignore this field. Labels are merged with the
235    /// provider-level pod labels and the hardcoded ironflow labels. In case
236    /// of conflict, hardcoded labels always win, then invocation-level labels,
237    /// then provider-level defaults.
238    #[serde(default, skip_serializing_if = "BTreeMap::is_empty")]
239    pub pod_labels: BTreeMap<String, String>,
240
241    /// External inputs to materialize on the agent's filesystem before invocation.
242    ///
243    /// See [`AgentInput`] for the semantics. The provider is responsible for
244    /// fetching each URL and placing it at `mount_path` before the agent runs.
245    /// Add inputs with [`AgentConfig::input_file`].
246    #[serde(default, skip_serializing_if = "Vec::is_empty")]
247    pub inputs: Vec<AgentInput>,
248
249    /// Zero-sized typestate marker (not serialized).
250    #[serde(skip)]
251    pub(crate) _marker: PhantomData<(Tools, Schema)>,
252}
253
254fn default_model() -> String {
255    Model::SONNET.to_string()
256}
257
258// ── Constructor (base type only) ───────────────────────────────────
259
260impl AgentConfig {
261    /// Create an `AgentConfig` with required fields and defaults for the rest.
262    pub fn new(prompt: &str) -> Self {
263        Self {
264            system_prompt: None,
265            prompt: prompt.to_string(),
266            model: Model::SONNET.to_string(),
267            allowed_tools: Vec::new(),
268            disallowed_tools: Vec::new(),
269            max_turns: None,
270            max_budget_usd: None,
271            working_dir: None,
272            mcp_config: None,
273            strict_mcp_config: false,
274            bare: false,
275            permission_mode: PermissionMode::Default,
276            json_schema: None,
277            resume_session_id: None,
278            verbose: false,
279            pod_labels: BTreeMap::new(),
280            inputs: Vec::new(),
281            _marker: PhantomData,
282        }
283    }
284}
285
286// ── Methods available on ALL typestate variants ────────────────────
287
288impl<Tools, Schema> AgentConfig<Tools, Schema> {
289    /// Set the system prompt.
290    pub fn system_prompt(mut self, prompt: &str) -> Self {
291        self.system_prompt = Some(prompt.to_string());
292        self
293    }
294
295    /// Set the model name.
296    pub fn model(mut self, model: &str) -> Self {
297        self.model = model.to_string();
298        self
299    }
300
301    /// Set the maximum budget in USD.
302    pub fn max_budget_usd(mut self, budget: f64) -> Self {
303        self.max_budget_usd = Some(budget);
304        self
305    }
306
307    /// Set the maximum number of turns.
308    pub fn max_turns(mut self, turns: u32) -> Self {
309        self.max_turns = Some(turns);
310        self
311    }
312
313    /// Set the working directory.
314    pub fn working_dir(mut self, dir: &str) -> Self {
315        self.working_dir = Some(dir.to_string());
316        self
317    }
318
319    /// Set the permission mode.
320    pub fn permission_mode(mut self, mode: PermissionMode) -> Self {
321        self.permission_mode = mode;
322        self
323    }
324
325    /// Enable verbose/debug mode.
326    pub fn verbose(mut self, enabled: bool) -> Self {
327        self.verbose = enabled;
328        self
329    }
330
331    /// Set the MCP server configuration file path.
332    pub fn mcp_config(mut self, config: &str) -> Self {
333        self.mcp_config = Some(config.to_string());
334        self
335    }
336
337    /// Enable strict MCP config mode.
338    ///
339    /// When `true`, the Claude CLI is invoked with `--strict-mcp-config`,
340    /// which disables loading of any MCP server defined outside the
341    /// [`mcp_config`](Self::mcp_config) file (the global `~/.claude.json`
342    /// and user-level configs are ignored).
343    ///
344    /// This is the recommended way to prevent global MCP servers from
345    /// silently injecting tools into a structured-output step and
346    /// triggering the Claude CLI bug that returns `structured_output: null`
347    /// whenever any tool is active. See
348    /// <https://github.com/anthropics/claude-code/issues/18536>.
349    ///
350    /// # Examples
351    ///
352    /// ```
353    /// use ironflow_core::provider::AgentConfig;
354    /// use schemars::JsonSchema;
355    ///
356    /// #[derive(serde::Deserialize, JsonSchema)]
357    /// struct Out { ok: bool }
358    ///
359    /// // Isolate the step from any global MCP server so structured output works.
360    /// let config = AgentConfig::new("classify this")
361    ///     .strict_mcp_config(true)
362    ///     .mcp_config(r#"{"mcpServers":{}}"#)
363    ///     .output::<Out>();
364    /// ```
365    pub fn strict_mcp_config(mut self, strict: bool) -> Self {
366        self.strict_mcp_config = strict;
367        self
368    }
369
370    /// Enable bare mode (minimal Claude Code environment, see `--bare`).
371    ///
372    /// When `true`, the Claude CLI is invoked with `--bare`, which disables:
373    /// - auto-memory (no automatic `~/.claude/.../memory/*.md` file creation)
374    /// - `CLAUDE.md` auto-discovery (neither global nor project-level)
375    /// - hooks, LSP, plugin sync, attribution, background prefetches,
376    ///   keychain reads
377    ///
378    /// Sets `CLAUDE_CODE_SIMPLE=1` in the child process.
379    ///
380    /// Recommended for orchestrator steps that should not have any implicit
381    /// side effects on the user's filesystem or inherit user-level context
382    /// (email, preferences, etc.).
383    ///
384    /// # Authentication requirement
385    ///
386    /// `--bare` is **only compatible with an Anthropic API key**
387    /// (`ANTHROPIC_API_KEY` environment variable). It does **not** work with
388    /// OAuth authentication (`claude /login` / keychain-stored credentials),
389    /// because bare mode disables keychain reads. Invoking a bare agent on an
390    /// OAuth-only host will fail with an authentication error.
391    ///
392    /// # Examples
393    ///
394    /// ```
395    /// use ironflow_core::provider::AgentConfig;
396    ///
397    /// let config = AgentConfig::new("classify this")
398    ///     .bare(true);
399    /// ```
400    pub fn bare(mut self, enabled: bool) -> Self {
401        self.bare = enabled;
402        self
403    }
404
405    /// Replace the entire disallowed-tools list.
406    ///
407    /// Maps to `--disallowedTools` on the Claude CLI. This method is available
408    /// on **every** typestate variant (including
409    /// [`AgentConfig<NoTools, WithSchema>`]) because, unlike
410    /// [`allow_tool`](AgentConfig::allow_tool), `disallowed_tools` does not
411    /// activate any tool -- it only filters out tools that would otherwise be
412    /// loaded by default.
413    ///
414    /// As such, it is safe to combine with structured output:
415    ///
416    /// # Examples
417    ///
418    /// ```
419    /// use ironflow_core::provider::AgentConfig;
420    /// use schemars::JsonSchema;
421    ///
422    /// #[derive(serde::Deserialize, JsonSchema)]
423    /// struct Out { ok: bool }
424    ///
425    /// let config = AgentConfig::new("classify this")
426    ///     .disallowed_tools(["Write", "Edit"])
427    ///     .output::<Out>();
428    /// ```
429    pub fn disallowed_tools<I, S>(mut self, tools: I) -> Self
430    where
431        I: IntoIterator<Item = S>,
432        S: Into<String>,
433    {
434        self.disallowed_tools = tools.into_iter().map(Into::into).collect();
435        self
436    }
437
438    /// Add a single custom pod label (K8s providers only).
439    ///
440    /// Can be called multiple times. Non-K8s providers ignore this field.
441    ///
442    /// # Examples
443    ///
444    /// ```
445    /// use ironflow_core::provider::AgentConfig;
446    ///
447    /// let config = AgentConfig::new("analyze")
448    ///     .pod_label("ironflow.io/network-profile", "grafana-only")
449    ///     .pod_label("team", "observability");
450    /// ```
451    pub fn pod_label(mut self, key: &str, value: &str) -> Self {
452        self.pod_labels.insert(key.to_string(), value.to_string());
453        self
454    }
455
456    /// Replace the entire custom pod labels map (K8s providers only).
457    ///
458    /// Non-K8s providers ignore this field.
459    ///
460    /// # Examples
461    ///
462    /// ```
463    /// use std::collections::BTreeMap;
464    /// use ironflow_core::provider::AgentConfig;
465    ///
466    /// let mut labels = BTreeMap::new();
467    /// labels.insert("env".to_string(), "staging".to_string());
468    /// let config = AgentConfig::new("deploy").pod_labels(labels);
469    /// ```
470    pub fn pod_labels(mut self, labels: BTreeMap<String, String>) -> Self {
471        self.pod_labels = labels;
472        self
473    }
474
475    /// Set a session ID to resume a previous conversation.
476    pub fn resume(mut self, session_id: &str) -> Self {
477        self.resume_session_id = Some(session_id.to_string());
478        self
479    }
480
481    /// Declare an external input that the provider must materialize on the
482    /// agent's filesystem before invocation.
483    ///
484    /// `url` is fetched (HTTP/HTTPS) and written to `mount_path` (absolute
485    /// path) inside the agent's runtime. Each provider materializes inputs
486    /// in its own way:
487    ///
488    /// * Local provider: downloads to a temp dir on the host.
489    /// * K8s providers: spawn a `curlimages/curl` initContainer that downloads
490    ///   into a shared `emptyDir` mounted on the main container.
491    ///
492    /// Can be called multiple times to declare several inputs.
493    ///
494    /// # Examples
495    ///
496    /// ```
497    /// use ironflow_core::provider::AgentConfig;
498    ///
499    /// let config = AgentConfig::new("Read /work/dossier.pdf and summarize")
500    ///     .allow_tool("Read")
501    ///     .input_file("https://r2.example.com/dossier.pdf", "/work/dossier.pdf");
502    /// ```
503    pub fn input_file(mut self, url: &str, mount_path: &str) -> Self {
504        self.inputs.push(AgentInput::new(url, mount_path));
505        self
506    }
507
508    /// Convert to a different typestate by moving all fields.
509    ///
510    /// Safe because the marker is a zero-sized [`PhantomData`] -- no
511    /// runtime data changes.
512    fn change_state<T2, S2>(self) -> AgentConfig<T2, S2> {
513        AgentConfig {
514            system_prompt: self.system_prompt,
515            prompt: self.prompt,
516            model: self.model,
517            allowed_tools: self.allowed_tools,
518            disallowed_tools: self.disallowed_tools,
519            max_turns: self.max_turns,
520            max_budget_usd: self.max_budget_usd,
521            working_dir: self.working_dir,
522            mcp_config: self.mcp_config,
523            strict_mcp_config: self.strict_mcp_config,
524            bare: self.bare,
525            permission_mode: self.permission_mode,
526            json_schema: self.json_schema,
527            resume_session_id: self.resume_session_id,
528            verbose: self.verbose,
529            pod_labels: self.pod_labels,
530            inputs: self.inputs,
531            _marker: PhantomData,
532        }
533    }
534}
535
536// ── allow_tool: only when no schema is set ─────────────────────────
537
538impl<Tools> AgentConfig<Tools, NoSchema> {
539    /// Add an allowed tool.
540    ///
541    /// Can be called multiple times to allow several tools. Returns an
542    /// [`AgentConfig<WithTools, NoSchema>`], which **cannot** call
543    /// [`output`](AgentConfig::output) or [`output_schema_raw`](AgentConfig::output_schema_raw).
544    ///
545    /// This restriction exists because Claude CLI has a
546    /// [known bug](https://github.com/anthropics/claude-code/issues/18536)
547    /// where `--json-schema` combined with `--allowedTools` always returns
548    /// `structured_output: null`.
549    ///
550    /// **Workaround**: use two sequential agent steps -- one with tools to
551    /// gather data, then one with `.output::<T>()` to structure the result.
552    ///
553    /// # Examples
554    ///
555    /// ```
556    /// use ironflow_core::provider::AgentConfig;
557    ///
558    /// let config = AgentConfig::new("search the web")
559    ///     .allow_tool("WebSearch")
560    ///     .allow_tool("WebFetch");
561    /// ```
562    ///
563    /// ```compile_fail
564    /// use ironflow_core::provider::AgentConfig;
565    /// // ERROR: cannot set structured output after adding tools
566    /// let _ = AgentConfig::new("x")
567    ///     .allow_tool("Read")
568    ///     .output_schema_raw(r#"{"type":"object"}"#);
569    /// ```
570    pub fn allow_tool(mut self, tool: &str) -> AgentConfig<WithTools, NoSchema> {
571        self.allowed_tools.push(tool.to_string());
572        self.change_state()
573    }
574}
575
576// ── output: only when no tools are set ─────────────────────────────
577
578impl<Schema> AgentConfig<NoTools, Schema> {
579    /// Set structured output from a Rust type implementing [`JsonSchema`].
580    ///
581    /// The schema is serialized once at build time. When set, the provider
582    /// will request typed output conforming to this schema.
583    ///
584    /// **Important:** structured output requires `max_turns >= 2`.
585    ///
586    /// Returns an [`AgentConfig<NoTools, WithSchema>`], which **cannot**
587    /// call [`allow_tool`](AgentConfig::allow_tool).
588    ///
589    /// This restriction exists because Claude CLI has a
590    /// [known bug](https://github.com/anthropics/claude-code/issues/18536)
591    /// where `--json-schema` combined with `--allowedTools` always returns
592    /// `structured_output: null`.
593    ///
594    /// **Workaround**: use two sequential agent steps -- one with tools to
595    /// gather data, then one with `.output::<T>()` to structure the result.
596    ///
597    /// # Known limitations of Claude CLI structured output
598    ///
599    /// The Claude CLI does not guarantee strict schema conformance for
600    /// structured output. The following upstream bugs affect the behavior:
601    ///
602    /// - **Schema flattening** ([anthropics/claude-agent-sdk-python#502]):
603    ///   a schema like `{"type":"object","properties":{"items":{"type":"array",...}}}`
604    ///   may return a bare array instead of the wrapper object. The CLI
605    ///   non-deterministically flattens schemas with a single array field.
606    /// - **Non-deterministic wrapping** ([anthropics/claude-agent-sdk-python#374]):
607    ///   the same prompt can produce differently wrapped output across runs.
608    /// - **No conformance guarantee** ([anthropics/claude-code#9058]):
609    ///   the CLI does not validate output against the provided JSON schema.
610    ///
611    /// Because of these bugs, ironflow's provider layer applies multiple
612    /// fallback strategies when extracting the structured value (see
613    /// [`extract_structured_value`](crate::providers::claude::common::extract_structured_value)).
614    ///
615    /// [anthropics/claude-agent-sdk-python#502]: https://github.com/anthropics/claude-agent-sdk-python/issues/502
616    /// [anthropics/claude-agent-sdk-python#374]: https://github.com/anthropics/claude-agent-sdk-python/issues/374
617    /// [anthropics/claude-code#9058]: https://github.com/anthropics/claude-code/issues/9058
618    ///
619    /// # Examples
620    ///
621    /// ```
622    /// use ironflow_core::provider::AgentConfig;
623    /// use schemars::JsonSchema;
624    ///
625    /// #[derive(serde::Deserialize, JsonSchema)]
626    /// struct Labels { labels: Vec<String> }
627    ///
628    /// let config = AgentConfig::new("classify this text")
629    ///     .output::<Labels>();
630    /// ```
631    ///
632    /// ```compile_fail
633    /// use ironflow_core::provider::AgentConfig;
634    /// use schemars::JsonSchema;
635    /// #[derive(serde::Deserialize, JsonSchema)]
636    /// struct Out { x: i32 }
637    /// // ERROR: cannot add tools after setting structured output
638    /// let _ = AgentConfig::new("x").output::<Out>().allow_tool("Read");
639    /// ```
640    /// # Panics
641    ///
642    /// Panics if the schema generated by `schemars` cannot be serialized
643    /// to JSON. This indicates a bug in the type's `JsonSchema` derive,
644    /// not a recoverable runtime error.
645    pub fn output<T: JsonSchema>(mut self) -> AgentConfig<NoTools, WithSchema> {
646        let schema = schemars::schema_for!(T);
647        let serialized = serde_json::to_string(&schema).unwrap_or_else(|e| {
648            panic!(
649                "failed to serialize JSON schema for {}: {e}",
650                std::any::type_name::<T>()
651            )
652        });
653        self.json_schema = Some(serialized);
654        self.change_state()
655    }
656
657    /// Set structured output from a pre-serialized JSON Schema string.
658    ///
659    /// Returns an [`AgentConfig<NoTools, WithSchema>`], which **cannot**
660    /// call [`allow_tool`](AgentConfig::allow_tool). See [`output`](Self::output)
661    /// for the rationale and workaround.
662    pub fn output_schema_raw(mut self, schema: &str) -> AgentConfig<NoTools, WithSchema> {
663        self.json_schema = Some(schema.to_string());
664        self.change_state()
665    }
666}
667
668// ── From conversions to base type ──────────────────────────────────
669
670impl From<AgentConfig<WithTools, NoSchema>> for AgentConfig {
671    fn from(config: AgentConfig<WithTools, NoSchema>) -> Self {
672        config.change_state()
673    }
674}
675
676impl From<AgentConfig<NoTools, WithSchema>> for AgentConfig {
677    fn from(config: AgentConfig<NoTools, WithSchema>) -> Self {
678        config.change_state()
679    }
680}
681
682// ── AgentOutput ────────────────────────────────────────────────────
683
684/// Raw output returned by an [`AgentProvider`] after a successful invocation.
685///
686/// Carries the agent's response value together with usage and billing metadata.
687#[derive(Clone, Debug, Serialize, Deserialize)]
688#[non_exhaustive]
689pub struct AgentOutput {
690    /// The agent's response. A plain [`Value::String`] for text mode, or an
691    /// arbitrary JSON value when a JSON schema was requested.
692    pub value: Value,
693
694    /// Provider-assigned session identifier, useful for resuming conversations.
695    pub session_id: Option<String>,
696
697    /// Total cost in USD for this invocation, if reported by the provider.
698    pub cost_usd: Option<f64>,
699
700    /// Number of input tokens consumed, if reported.
701    pub input_tokens: Option<u64>,
702
703    /// Number of output tokens generated, if reported.
704    pub output_tokens: Option<u64>,
705
706    /// The concrete model identifier used (e.g. `"claude-sonnet-4-20250514"`).
707    pub model: Option<String>,
708
709    /// Wall-clock duration of the invocation in milliseconds.
710    pub duration_ms: u64,
711
712    /// Conversation trace captured when [`AgentConfig::verbose`] is `true`.
713    ///
714    /// Contains every assistant message and tool call made during the
715    /// invocation, in chronological order. `None` when verbose mode is off.
716    pub debug_messages: Option<Vec<DebugMessage>>,
717}
718
719/// A single assistant turn captured during a verbose invocation.
720///
721/// Each `DebugMessage` represents one assistant response, which may contain
722/// free-form text, tool calls, or both.
723///
724/// # Examples
725///
726/// ```no_run
727/// use ironflow_core::prelude::*;
728///
729/// # async fn example() -> Result<(), OperationError> {
730/// let provider = ClaudeCodeProvider::new();
731/// let result = Agent::new()
732///     .prompt("List files in src/")
733///     .verbose()
734///     .run(&provider)
735///     .await?;
736///
737/// if let Some(messages) = result.debug_messages() {
738///     for msg in messages {
739///         println!("{msg}");
740///     }
741/// }
742/// # Ok(())
743/// # }
744/// ```
745#[derive(Debug, Clone, Serialize, Deserialize)]
746#[non_exhaustive]
747pub struct DebugMessage {
748    /// Free-form text produced by the assistant in this turn, if any.
749    pub text: Option<String>,
750
751    /// Extended thinking blocks produced by the model in this turn.
752    ///
753    /// Available only when the model emits `thinking` content blocks
754    /// (Opus 4.7 adaptive thinking, Claude 3.7+ extended thinking, etc.).
755    /// The blocks are joined in arrival order.
756    #[serde(default, skip_serializing_if = "Option::is_none")]
757    pub thinking: Option<String>,
758
759    /// `true` when the model emitted a `thinking` content block but the
760    /// text was redacted (only a signature is provided).
761    ///
762    /// Opus 4.7 adaptive thinking and the `display: "omitted"` setting both
763    /// produce signature-only thinking blocks: the model proves it reasoned
764    /// without exposing the chain of thought. The UI should still show a
765    /// badge so the user knows thinking happened.
766    #[serde(default, skip_serializing_if = "std::ops::Not::not")]
767    pub thinking_redacted: bool,
768
769    /// Tool calls made by the assistant in this turn.
770    pub tool_calls: Vec<DebugToolCall>,
771
772    /// Tool results received from the user/runtime for the preceding tool calls.
773    ///
774    /// In the Claude stream-json format, tool results come as `"type":"user"`
775    /// messages whose content is a list of `tool_result` blocks. We attach
776    /// them to the turn that emitted the matching `tool_use` so the timeline
777    /// stays compact.
778    #[serde(default, skip_serializing_if = "Vec::is_empty")]
779    pub tool_results: Vec<DebugToolResult>,
780
781    /// The model's stop reason for this turn (e.g. `"end_turn"`, `"tool_use"`).
782    pub stop_reason: Option<String>,
783
784    /// Input tokens consumed by this turn, if reported.
785    #[serde(default, skip_serializing_if = "Option::is_none")]
786    pub input_tokens: Option<u64>,
787
788    /// Output tokens generated by this turn, if reported.
789    #[serde(default, skip_serializing_if = "Option::is_none")]
790    pub output_tokens: Option<u64>,
791}
792
793impl fmt::Display for DebugMessage {
794    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
795        if let Some(ref thinking) = self.thinking {
796            writeln!(f, "[thinking] {thinking}")?;
797        } else if self.thinking_redacted {
798            writeln!(f, "[thinking redacted]")?;
799        }
800        if let Some(ref text) = self.text {
801            writeln!(f, "[assistant] {text}")?;
802        }
803        for tc in &self.tool_calls {
804            write!(f, "{tc}")?;
805        }
806        for tr in &self.tool_results {
807            write!(f, "{tr}")?;
808        }
809        Ok(())
810    }
811}
812
813/// A single tool call captured during a verbose invocation.
814///
815/// Records the tool name and its input arguments as a raw JSON value.
816#[derive(Debug, Clone, Serialize, Deserialize)]
817#[non_exhaustive]
818pub struct DebugToolCall {
819    /// Stable identifier assigned by the model (`tool_use_id`).
820    ///
821    /// Used to correlate a call with its subsequent [`DebugToolResult`].
822    #[serde(default, skip_serializing_if = "Option::is_none")]
823    pub id: Option<String>,
824
825    /// Name of the tool invoked (e.g. `"Read"`, `"Bash"`, `"Grep"`).
826    pub name: String,
827
828    /// Input arguments passed to the tool, as raw JSON.
829    pub input: Value,
830}
831
832impl fmt::Display for DebugToolCall {
833    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
834        writeln!(f, "  [tool_use] {} -> {}", self.name, self.input)
835    }
836}
837
838/// A tool result returned to the model after a tool call.
839///
840/// Carries the tool output (any JSON value: string, object, array) and
841/// an error flag if the tool failed.
842#[derive(Debug, Clone, Serialize, Deserialize)]
843#[non_exhaustive]
844pub struct DebugToolResult {
845    /// The `tool_use_id` this result answers, matching [`DebugToolCall::id`].
846    #[serde(default, skip_serializing_if = "Option::is_none")]
847    pub tool_use_id: Option<String>,
848
849    /// Raw content returned by the tool.
850    pub content: Value,
851
852    /// Whether the tool reported an error.
853    #[serde(default)]
854    pub is_error: bool,
855}
856
857impl fmt::Display for DebugToolResult {
858    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
859        let kind = if self.is_error {
860            "tool_error"
861        } else {
862            "tool_result"
863        };
864        writeln!(f, "  [{kind}] {}", self.content)
865    }
866}
867
868impl AgentOutput {
869    /// Create an `AgentOutput` with the given value and sensible defaults.
870    pub fn new(value: Value) -> Self {
871        Self {
872            value,
873            session_id: None,
874            cost_usd: None,
875            input_tokens: None,
876            output_tokens: None,
877            model: None,
878            duration_ms: 0,
879            debug_messages: None,
880        }
881    }
882}
883
884// ── Log sink ──────────────────────────────────────────────────────
885
886/// Sink for streaming log lines from provider invocations in real time.
887///
888/// Providers that support live log streaming (e.g. K8s ephemeral) call
889/// [`log`](LogSink::log) for each output line as it is produced, enabling
890/// downstream consumers (SSE endpoints, log pushers) to display progress
891/// before the invocation completes.
892///
893/// This trait lives in `ironflow-core` so providers can emit logs without
894/// depending on higher-level crates.
895///
896/// # Examples
897///
898/// ```
899/// use std::sync::{Arc, Mutex};
900/// use ironflow_core::provider::LogSink;
901///
902/// struct VecSink(Mutex<Vec<(String, String)>>);
903///
904/// impl LogSink for VecSink {
905///     fn log(&self, stream: &str, line: &str) {
906///         self.0.lock().unwrap().push((stream.to_string(), line.to_string()));
907///     }
908/// }
909///
910/// let sink = Arc::new(VecSink(Mutex::new(Vec::new())));
911/// sink.log("stdout", "hello world");
912/// assert_eq!(sink.0.lock().unwrap().len(), 1);
913/// ```
914pub trait LogSink: Send + Sync {
915    /// Emit a single log line on the given stream.
916    ///
917    /// `stream` is one of `"stdout"`, `"stderr"`, or `"system"`.
918    /// Implementations should silently drop lines if the receiver is closed.
919    fn log(&self, stream: &str, line: &str);
920}
921
922// ── Provider trait ─────────────────────────────────────────────────
923
924/// Trait for AI agent backends.
925///
926/// Implement this trait to provide a custom AI backend for [`Agent`](crate::operations::agent::Agent).
927/// The only required method is [`invoke`](AgentProvider::invoke), which takes an
928/// [`AgentConfig`] and returns an [`AgentOutput`] (or an [`AgentError`]).
929///
930/// # Examples
931///
932/// ```no_run
933/// use ironflow_core::provider::{AgentConfig, AgentOutput, AgentProvider, InvokeFuture};
934///
935/// struct MyProvider;
936///
937/// impl AgentProvider for MyProvider {
938///     fn invoke<'a>(&'a self, config: &'a AgentConfig) -> InvokeFuture<'a> {
939///         Box::pin(async move {
940///             // Call your custom backend here...
941///             todo!()
942///         })
943///     }
944/// }
945/// ```
946pub trait AgentProvider: Send + Sync {
947    /// Execute a single agent invocation with the given configuration.
948    ///
949    /// # Errors
950    ///
951    /// Returns [`AgentError`] if the underlying backend process fails,
952    /// times out, or produces output that does not match the requested schema.
953    fn invoke<'a>(&'a self, config: &'a AgentConfig) -> InvokeFuture<'a>;
954
955    /// Execute an agent invocation with real-time log streaming.
956    ///
957    /// Providers that support live output streaming should override this
958    /// method to pipe each output line to the [`LogSink`] as it arrives.
959    /// The default implementation ignores the sink and delegates to
960    /// [`invoke`](AgentProvider::invoke).
961    ///
962    /// # Errors
963    ///
964    /// Returns [`AgentError`] if the underlying backend process fails,
965    /// times out, or produces output that does not match the requested schema.
966    fn invoke_with_logs<'a>(
967        &'a self,
968        config: &'a AgentConfig,
969        log_sink: Arc<dyn LogSink>,
970    ) -> InvokeFuture<'a> {
971        let _ = log_sink;
972        self.invoke(config)
973    }
974}
975
976#[cfg(test)]
977mod tests {
978    use super::*;
979    use serde_json::json;
980
981    fn full_config() -> AgentConfig {
982        AgentConfig {
983            system_prompt: Some("you are helpful".to_string()),
984            prompt: "do stuff".to_string(),
985            model: Model::OPUS.to_string(),
986            allowed_tools: vec!["Read".to_string(), "Write".to_string()],
987            disallowed_tools: vec!["Bash".to_string()],
988            max_turns: Some(10),
989            max_budget_usd: Some(2.5),
990            working_dir: Some("/tmp".to_string()),
991            mcp_config: Some("{}".to_string()),
992            strict_mcp_config: true,
993            bare: true,
994            permission_mode: PermissionMode::Auto,
995            json_schema: Some(r#"{"type":"object"}"#.to_string()),
996            resume_session_id: None,
997            verbose: false,
998            pod_labels: BTreeMap::new(),
999            inputs: Vec::new(),
1000            _marker: PhantomData,
1001        }
1002    }
1003
1004    #[test]
1005    fn agent_config_serialize_deserialize_roundtrip() {
1006        let config = full_config();
1007        let json = serde_json::to_string(&config).unwrap();
1008        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1009
1010        assert_eq!(back.system_prompt, Some("you are helpful".to_string()));
1011        assert_eq!(back.prompt, "do stuff");
1012        assert_eq!(back.allowed_tools, vec!["Read", "Write"]);
1013        assert_eq!(back.max_turns, Some(10));
1014        assert_eq!(back.max_budget_usd, Some(2.5));
1015        assert_eq!(back.working_dir, Some("/tmp".to_string()));
1016        assert_eq!(back.mcp_config, Some("{}".to_string()));
1017        assert_eq!(back.json_schema, Some(r#"{"type":"object"}"#.to_string()));
1018    }
1019
1020    #[test]
1021    fn agent_config_with_all_optional_fields_none() {
1022        let config: AgentConfig = AgentConfig {
1023            system_prompt: None,
1024            prompt: "hello".to_string(),
1025            model: Model::HAIKU.to_string(),
1026            allowed_tools: vec![],
1027            disallowed_tools: vec![],
1028            max_turns: None,
1029            max_budget_usd: None,
1030            working_dir: None,
1031            mcp_config: None,
1032            strict_mcp_config: false,
1033            bare: false,
1034            permission_mode: PermissionMode::Default,
1035            json_schema: None,
1036            resume_session_id: None,
1037            verbose: false,
1038            pod_labels: BTreeMap::new(),
1039            inputs: Vec::new(),
1040            _marker: PhantomData,
1041        };
1042        let json = serde_json::to_string(&config).unwrap();
1043        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1044
1045        assert_eq!(back.system_prompt, None);
1046        assert_eq!(back.prompt, "hello");
1047        assert!(back.allowed_tools.is_empty());
1048        assert_eq!(back.max_turns, None);
1049        assert_eq!(back.max_budget_usd, None);
1050        assert_eq!(back.working_dir, None);
1051        assert_eq!(back.mcp_config, None);
1052        assert_eq!(back.json_schema, None);
1053    }
1054
1055    #[test]
1056    fn agent_output_serialize_deserialize_roundtrip() {
1057        let output = AgentOutput {
1058            value: json!({"key": "value"}),
1059            session_id: Some("sess-abc".to_string()),
1060            cost_usd: Some(0.01),
1061            input_tokens: Some(500),
1062            output_tokens: Some(200),
1063            model: Some("claude-sonnet".to_string()),
1064            duration_ms: 3000,
1065            debug_messages: None,
1066        };
1067        let json = serde_json::to_string(&output).unwrap();
1068        let back: AgentOutput = serde_json::from_str(&json).unwrap();
1069
1070        assert_eq!(back.value, json!({"key": "value"}));
1071        assert_eq!(back.session_id, Some("sess-abc".to_string()));
1072        assert_eq!(back.cost_usd, Some(0.01));
1073        assert_eq!(back.input_tokens, Some(500));
1074        assert_eq!(back.output_tokens, Some(200));
1075        assert_eq!(back.model, Some("claude-sonnet".to_string()));
1076        assert_eq!(back.duration_ms, 3000);
1077    }
1078
1079    #[test]
1080    fn agent_config_new_has_correct_defaults() {
1081        let config = AgentConfig::new("test prompt");
1082        assert_eq!(config.prompt, "test prompt");
1083        assert_eq!(config.system_prompt, None);
1084        assert_eq!(config.model, Model::SONNET);
1085        assert!(config.allowed_tools.is_empty());
1086        assert_eq!(config.max_turns, None);
1087        assert_eq!(config.max_budget_usd, None);
1088        assert_eq!(config.working_dir, None);
1089        assert_eq!(config.mcp_config, None);
1090        assert!(matches!(config.permission_mode, PermissionMode::Default));
1091        assert_eq!(config.json_schema, None);
1092        assert_eq!(config.resume_session_id, None);
1093        assert!(!config.verbose);
1094    }
1095
1096    #[test]
1097    fn agent_output_new_has_correct_defaults() {
1098        let output = AgentOutput::new(json!("test"));
1099        assert_eq!(output.value, json!("test"));
1100        assert_eq!(output.session_id, None);
1101        assert_eq!(output.cost_usd, None);
1102        assert_eq!(output.input_tokens, None);
1103        assert_eq!(output.output_tokens, None);
1104        assert_eq!(output.model, None);
1105        assert_eq!(output.duration_ms, 0);
1106        assert!(output.debug_messages.is_none());
1107    }
1108
1109    #[test]
1110    fn agent_config_resume_session_roundtrip() {
1111        let mut config = AgentConfig::new("test");
1112        config.resume_session_id = Some("sess-xyz".to_string());
1113        let json = serde_json::to_string(&config).unwrap();
1114        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1115        assert_eq!(back.resume_session_id, Some("sess-xyz".to_string()));
1116    }
1117
1118    #[test]
1119    fn agent_output_debug_does_not_panic() {
1120        let output = AgentOutput {
1121            value: json!(null),
1122            session_id: None,
1123            cost_usd: None,
1124            input_tokens: None,
1125            output_tokens: None,
1126            model: None,
1127            duration_ms: 0,
1128            debug_messages: None,
1129        };
1130        let debug_str = format!("{:?}", output);
1131        assert!(!debug_str.is_empty());
1132    }
1133
1134    #[test]
1135    fn allow_tool_transitions_to_with_tools() {
1136        let config = AgentConfig::new("test").allow_tool("Read");
1137        assert_eq!(config.allowed_tools, vec!["Read"]);
1138
1139        // Can add more tools
1140        let config = config.allow_tool("Write");
1141        assert_eq!(config.allowed_tools, vec!["Read", "Write"]);
1142    }
1143
1144    #[test]
1145    fn output_schema_raw_transitions_to_with_schema() {
1146        let config = AgentConfig::new("test").output_schema_raw(r#"{"type":"object"}"#);
1147        assert_eq!(config.json_schema.as_deref(), Some(r#"{"type":"object"}"#));
1148    }
1149
1150    #[test]
1151    fn with_tools_converts_to_base_type() {
1152        let typed = AgentConfig::new("test").allow_tool("Read");
1153        let base: AgentConfig = typed.into();
1154        assert_eq!(base.allowed_tools, vec!["Read"]);
1155    }
1156
1157    #[test]
1158    fn with_schema_converts_to_base_type() {
1159        let typed = AgentConfig::new("test").output_schema_raw(r#"{"type":"object"}"#);
1160        let base: AgentConfig = typed.into();
1161        assert_eq!(base.json_schema.as_deref(), Some(r#"{"type":"object"}"#));
1162    }
1163
1164    #[test]
1165    fn serde_roundtrip_ignores_marker() {
1166        let config = AgentConfig::new("test").allow_tool("Read");
1167        let json = serde_json::to_string(&config).unwrap();
1168        assert!(!json.contains("marker"));
1169
1170        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1171        assert_eq!(back.allowed_tools, vec!["Read"]);
1172    }
1173
1174    #[test]
1175    fn bare_defaults_to_false() {
1176        let config = AgentConfig::new("hello");
1177        assert!(!config.bare, "bare must default to false");
1178    }
1179
1180    #[test]
1181    fn bare_builder_sets_flag() {
1182        let config = AgentConfig::new("hello").bare(true);
1183        assert!(config.bare, "bare(true) must enable the flag");
1184
1185        let config = config.bare(false);
1186        assert!(!config.bare, "bare(false) must disable the flag");
1187    }
1188
1189    #[test]
1190    fn bare_serde_default_when_missing() {
1191        let raw = r#"{"prompt":"hello","model":"sonnet"}"#;
1192        let config: AgentConfig = serde_json::from_str(raw).unwrap();
1193        assert!(
1194            !config.bare,
1195            "bare must default to false when absent from serialized payload"
1196        );
1197    }
1198
1199    #[test]
1200    fn bare_serde_roundtrip() {
1201        let mut config = AgentConfig::new("hello");
1202        config.bare = true;
1203        let json = serde_json::to_string(&config).unwrap();
1204        assert!(
1205            json.contains("\"bare\":true"),
1206            "serialized form must contain bare:true, got: {json}"
1207        );
1208
1209        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1210        assert!(back.bare, "bare must survive a serde roundtrip");
1211    }
1212
1213    #[test]
1214    fn disallowed_tools_defaults_to_empty() {
1215        let config = AgentConfig::new("hello");
1216        assert!(
1217            config.disallowed_tools.is_empty(),
1218            "disallowed_tools must default to empty"
1219        );
1220    }
1221
1222    #[test]
1223    fn disallowed_tools_builder_replaces_list() {
1224        let config = AgentConfig::new("hello").disallowed_tools(["Write", "Edit"]);
1225        assert_eq!(config.disallowed_tools, vec!["Write", "Edit"]);
1226
1227        // Subsequent call fully replaces the list.
1228        let config = config.disallowed_tools(["Bash"]);
1229        assert_eq!(config.disallowed_tools, vec!["Bash"]);
1230
1231        // Empty input clears the list.
1232        let config = config.disallowed_tools(std::iter::empty::<String>());
1233        assert!(config.disallowed_tools.is_empty());
1234    }
1235
1236    #[test]
1237    fn disallowed_tools_compatible_with_output() {
1238        #[derive(serde::Deserialize, JsonSchema)]
1239        #[allow(dead_code)]
1240        struct Out {
1241            ok: bool,
1242        }
1243
1244        // Typestate compile check: .disallowed_tools(...) must be callable
1245        // before AND after .output::<T>() because it lives on
1246        // impl<Tools, Schema>, not impl<Tools, NoSchema>.
1247        let before: AgentConfig<NoTools, WithSchema> = AgentConfig::new("classify")
1248            .disallowed_tools(["Write", "Edit"])
1249            .output::<Out>();
1250        assert_eq!(before.disallowed_tools, vec!["Write", "Edit"]);
1251        assert!(before.json_schema.is_some());
1252
1253        let after: AgentConfig<NoTools, WithSchema> = AgentConfig::new("classify")
1254            .output::<Out>()
1255            .disallowed_tools(["Write"]);
1256        assert_eq!(after.disallowed_tools, vec!["Write"]);
1257        assert!(after.json_schema.is_some());
1258    }
1259
1260    #[test]
1261    fn disallowed_tools_serde_default_when_missing() {
1262        let raw = r#"{"prompt":"hello","model":"sonnet"}"#;
1263        let config: AgentConfig = serde_json::from_str(raw).unwrap();
1264        assert!(
1265            config.disallowed_tools.is_empty(),
1266            "disallowed_tools must default to empty when absent from serialized payload"
1267        );
1268    }
1269
1270    #[test]
1271    fn disallowed_tools_serde_roundtrip() {
1272        let config = AgentConfig::new("hello").disallowed_tools(["Write", "Edit"]);
1273        let json = serde_json::to_string(&config).unwrap();
1274        assert!(
1275            json.contains("\"disallowed_tools\":[\"Write\",\"Edit\"]"),
1276            "serialized form must contain the disallowed_tools array, got: {json}"
1277        );
1278
1279        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1280        assert_eq!(back.disallowed_tools, vec!["Write", "Edit"]);
1281    }
1282
1283    #[test]
1284    fn pod_labels_defaults_to_empty() {
1285        let config = AgentConfig::new("test");
1286        assert!(config.pod_labels.is_empty());
1287    }
1288
1289    #[test]
1290    fn pod_label_builder_adds_entry() {
1291        let config = AgentConfig::new("test").pod_label("k", "v");
1292        assert_eq!(config.pod_labels.len(), 1);
1293        assert_eq!(config.pod_labels["k"], "v");
1294    }
1295
1296    #[test]
1297    fn pod_labels_builder_replaces_map() {
1298        let config = AgentConfig::new("test").pod_label("old", "value");
1299        let mut new_map = BTreeMap::new();
1300        new_map.insert("new".to_string(), "value".to_string());
1301        let config = config.pod_labels(new_map);
1302        assert_eq!(config.pod_labels.len(), 1);
1303        assert_eq!(config.pod_labels["new"], "value");
1304        assert!(!config.pod_labels.contains_key("old"));
1305    }
1306
1307    #[test]
1308    fn pod_labels_serde_default_when_missing() {
1309        let raw = r#"{"prompt":"hello","model":"sonnet"}"#;
1310        let config: AgentConfig = serde_json::from_str(raw).unwrap();
1311        assert!(
1312            config.pod_labels.is_empty(),
1313            "pod_labels must default to empty when absent from serialized payload"
1314        );
1315    }
1316
1317    #[test]
1318    fn pod_labels_serde_skip_when_empty() {
1319        let config = AgentConfig::new("hello");
1320        let json = serde_json::to_string(&config).unwrap();
1321        assert!(
1322            !json.contains("pod_labels"),
1323            "empty pod_labels must be skipped during serialization, got: {json}"
1324        );
1325    }
1326
1327    #[test]
1328    fn pod_labels_serde_roundtrip() {
1329        let config = AgentConfig::new("hello")
1330            .pod_label("ironflow.io/network-profile", "grafana-only")
1331            .pod_label("team", "observability");
1332        let json = serde_json::to_string(&config).unwrap();
1333        assert!(
1334            json.contains("pod_labels"),
1335            "non-empty pod_labels must be present in serialized form, got: {json}"
1336        );
1337
1338        let back: AgentConfig = serde_json::from_str(&json).unwrap();
1339        assert_eq!(back.pod_labels.len(), 2);
1340        assert_eq!(
1341            back.pod_labels["ironflow.io/network-profile"],
1342            "grafana-only"
1343        );
1344        assert_eq!(back.pod_labels["team"], "observability");
1345    }
1346
1347    // ── LogSink tests ─────────────────────────────────────────────
1348
1349    use crate::test_support::VecSink;
1350
1351    #[test]
1352    fn log_sink_collects_lines() {
1353        let sink = VecSink::new();
1354        sink.log("stdout", "line 1");
1355        sink.log("stderr", "err!");
1356        sink.log("system", "done");
1357
1358        let lines = sink.0.lock().unwrap();
1359        assert_eq!(lines.len(), 3);
1360        assert_eq!(lines[0], ("stdout".to_string(), "line 1".to_string()));
1361        assert_eq!(lines[1], ("stderr".to_string(), "err!".to_string()));
1362        assert_eq!(lines[2], ("system".to_string(), "done".to_string()));
1363    }
1364
1365    #[test]
1366    fn log_sink_arc_is_clone_and_send() {
1367        let sink: Arc<dyn LogSink> = VecSink::new();
1368        let cloned = sink.clone();
1369        sink.log("stdout", "from original");
1370        cloned.log("stdout", "from clone");
1371    }
1372
1373    // ── invoke_with_logs default impl ─────────────────────────────
1374
1375    struct FixedProvider {
1376        output: AgentOutput,
1377    }
1378
1379    impl AgentProvider for FixedProvider {
1380        fn invoke<'a>(&'a self, _config: &'a AgentConfig) -> InvokeFuture<'a> {
1381            Box::pin(async {
1382                Ok(AgentOutput {
1383                    value: self.output.value.clone(),
1384                    session_id: self.output.session_id.clone(),
1385                    cost_usd: self.output.cost_usd,
1386                    input_tokens: self.output.input_tokens,
1387                    output_tokens: self.output.output_tokens,
1388                    model: self.output.model.clone(),
1389                    duration_ms: self.output.duration_ms,
1390                    debug_messages: None,
1391                })
1392            })
1393        }
1394    }
1395
1396    #[tokio::test]
1397    async fn invoke_with_logs_default_delegates_to_invoke() {
1398        let provider = FixedProvider {
1399            output: AgentOutput::new(json!("ok")),
1400        };
1401        let config = AgentConfig::new("test");
1402        let sink: Arc<dyn LogSink> = VecSink::new();
1403
1404        let result = provider.invoke_with_logs(&config, sink.clone()).await;
1405        assert!(result.is_ok());
1406        assert_eq!(result.unwrap().value, json!("ok"));
1407    }
1408
1409    #[tokio::test]
1410    async fn invoke_with_logs_default_ignores_sink() {
1411        let provider = FixedProvider {
1412            output: AgentOutput::new(json!("ok")),
1413        };
1414        let config = AgentConfig::new("test");
1415        let sink = VecSink::new();
1416
1417        let _ = provider
1418            .invoke_with_logs(&config, sink.clone() as Arc<dyn LogSink>)
1419            .await;
1420
1421        let lines = sink.0.lock().unwrap();
1422        assert!(lines.is_empty(), "default impl should not emit any logs");
1423    }
1424}