feat: add multi-actions to presentation tool (#13357)

This commit is contained in:
jif-oai 2026-03-03 14:37:26 +00:00 committed by GitHub
parent ad393fa753
commit 1df040e62b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 262 additions and 55 deletions

View file

@ -85,6 +85,27 @@ pub struct PresentationArtifactRequest {
pub args: Value,
}
#[derive(Debug, Clone, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct PresentationArtifactToolRequest {
pub artifact_id: Option<String>,
pub actions: Vec<PresentationArtifactToolAction>,
}
#[derive(Debug, Clone)]
pub struct PresentationArtifactExecutionRequest {
pub artifact_id: Option<String>,
pub requests: Vec<PresentationArtifactRequest>,
}
#[derive(Debug, Clone, Deserialize)]
#[serde(deny_unknown_fields)]
pub struct PresentationArtifactToolAction {
pub action: String,
#[serde(default)]
pub args: Value,
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum PathAccessKind {
Read,
@ -99,6 +120,10 @@ pub struct PathAccessRequirement {
}
impl PresentationArtifactRequest {
pub fn is_mutating(&self) -> bool {
!is_read_only_action(&self.action)
}
pub fn required_path_accesses(
&self,
cwd: &Path,
@ -175,3 +200,50 @@ impl PresentationArtifactRequest {
Ok(access)
}
}
impl PresentationArtifactToolRequest {
pub fn is_mutating(&self) -> Result<bool, PresentationArtifactError> {
Ok(self.actions.iter().any(|request| !is_read_only_action(&request.action)))
}
pub fn into_execution_request(
self,
) -> Result<PresentationArtifactExecutionRequest, PresentationArtifactError> {
if self.actions.is_empty() {
return Err(PresentationArtifactError::InvalidArgs {
action: "presentation_artifact".to_string(),
message: "`actions` must contain at least one item".to_string(),
});
}
Ok(PresentationArtifactExecutionRequest {
artifact_id: self.artifact_id,
requests: self
.actions
.into_iter()
.map(|request| PresentationArtifactRequest {
artifact_id: None,
action: request.action,
args: request.args,
})
.collect(),
})
}
pub fn required_path_accesses(
&self,
cwd: &Path,
) -> Result<Vec<PathAccessRequirement>, PresentationArtifactError> {
let mut accesses = Vec::new();
for request in &self.actions {
accesses.extend(
PresentationArtifactRequest {
artifact_id: None,
action: request.action.clone(),
args: request.args.clone(),
}
.required_path_accesses(cwd)?,
);
}
Ok(accesses)
}
}

View file

@ -14,6 +14,47 @@ struct HistoryEntry {
}
impl PresentationArtifactManager {
pub fn execute_requests(
&mut self,
request: PresentationArtifactExecutionRequest,
cwd: &Path,
) -> Result<PresentationArtifactResponse, PresentationArtifactError> {
let PresentationArtifactExecutionRequest {
artifact_id,
requests,
} = request;
let request_count = requests.len();
let mut current_artifact_id = artifact_id;
let mut executed_actions = Vec::with_capacity(request_count);
let mut exported_paths = Vec::new();
let mut last_response = None;
for mut request in requests {
if request.artifact_id.is_none() {
request.artifact_id = current_artifact_id.clone();
}
let response = self.execute(request, cwd)?;
current_artifact_id = Some(response.artifact_id.clone());
exported_paths.extend(response.exported_paths.iter().cloned());
executed_actions.push(response.action.clone());
last_response = Some(response);
}
let mut response = last_response.ok_or_else(|| PresentationArtifactError::InvalidArgs {
action: "presentation_artifact".to_string(),
message: "request sequence must contain at least one action".to_string(),
})?;
if request_count > 1 {
let final_summary = response.summary.clone();
response.action = "batch".to_string();
response.summary =
format!("Executed {request_count} actions sequentially. {final_summary}");
response.executed_actions = Some(executed_actions);
response.exported_paths = exported_paths;
}
Ok(response)
}
pub fn execute(
&mut self,
request: PresentationArtifactRequest,
@ -2163,6 +2204,7 @@ impl PresentationArtifactManager {
removed.artifact_id,
removed.slides.len()
),
executed_actions: None,
exported_paths: Vec::new(),
artifact_snapshot: None,
slide_list: None,

View file

@ -3,6 +3,8 @@ pub struct PresentationArtifactResponse {
pub artifact_id: String,
pub action: String,
pub summary: String,
#[serde(skip_serializing_if = "Option::is_none")]
pub executed_actions: Option<Vec<String>>,
#[serde(skip_serializing_if = "Vec::is_empty")]
pub exported_paths: Vec<PathBuf>,
#[serde(skip_serializing_if = "Option::is_none")]
@ -38,6 +40,7 @@ impl PresentationArtifactResponse {
artifact_id,
action,
summary,
executed_actions: None,
exported_paths: Vec::new(),
artifact_snapshot: Some(artifact_snapshot),
slide_list: None,
@ -63,6 +66,7 @@ fn response_for_document_state(
artifact_id,
action,
summary,
executed_actions: None,
exported_paths: Vec::new(),
artifact_snapshot: document.map(snapshot_for_document),
slide_list: None,
@ -132,4 +136,3 @@ pub struct ThemeSnapshot {
pub major_font: Option<String>,
pub minor_font: Option<String>,
}

View file

@ -258,6 +258,85 @@ fn exported_images_are_real_pictures_with_media_parts() -> Result<(), Box<dyn st
Ok(())
}
#[test]
fn tool_request_accepts_sequential_actions() -> Result<(), Box<dyn std::error::Error>> {
let request: PresentationArtifactToolRequest = serde_json::from_value(serde_json::json!({
"actions": [
{
"action": "create",
"args": { "name": "Batch Deck" }
},
{
"action": "export_pptx",
"args": { "path": "deck.pptx" }
}
]
}))?;
let execution = request.into_execution_request()?;
assert_eq!(execution.artifact_id, None);
assert_eq!(execution.requests.len(), 2);
assert_eq!(execution.requests[0].action, "create");
assert_eq!(execution.requests[1].action, "export_pptx");
Ok(())
}
#[test]
fn manager_can_execute_sequential_actions() -> Result<(), Box<dyn std::error::Error>> {
let temp_dir = tempfile::tempdir()?;
let mut manager = PresentationArtifactManager::default();
let response = manager.execute_requests(
PresentationArtifactExecutionRequest {
artifact_id: None,
requests: vec![
PresentationArtifactRequest {
artifact_id: None,
action: "create".to_string(),
args: serde_json::json!({ "name": "Batch Deck" }),
},
PresentationArtifactRequest {
artifact_id: None,
action: "add_slide".to_string(),
args: serde_json::json!({}),
},
PresentationArtifactRequest {
artifact_id: None,
action: "add_text_shape".to_string(),
args: serde_json::json!({
"slide_index": 0,
"text": "hello",
"position": { "left": 40, "top": 40, "width": 200, "height": 80 }
}),
},
],
},
temp_dir.path(),
)?;
assert_eq!(response.action, "batch");
assert_eq!(
response.executed_actions,
Some(vec![
"create".to_string(),
"add_slide".to_string(),
"add_text_shape".to_string(),
])
);
assert_eq!(
response
.artifact_snapshot
.as_ref()
.map(|snapshot| snapshot.slide_count),
Some(1)
);
assert!(
response
.summary
.contains("Executed 3 actions sequentially.")
);
Ok(())
}
#[test]
fn imported_pptx_surfaces_image_elements() -> Result<(), Box<dyn std::error::Error>> {
let temp_dir = tempfile::tempdir()?;

View file

@ -56,7 +56,7 @@ use async_channel::Sender;
use chrono::Local;
use chrono::Utc;
use codex_artifact_presentation::PresentationArtifactError;
use codex_artifact_presentation::PresentationArtifactRequest;
use codex_artifact_presentation::PresentationArtifactExecutionRequest;
use codex_artifact_presentation::PresentationArtifactResponse;
use codex_hooks::HookEvent;
use codex_hooks::HookEventAfterAgent;
@ -1782,11 +1782,11 @@ impl Session {
pub(crate) async fn execute_presentation_artifact(
&self,
request: PresentationArtifactRequest,
request: PresentationArtifactExecutionRequest,
cwd: &Path,
) -> Result<PresentationArtifactResponse, PresentationArtifactError> {
let mut state = self.state.lock().await;
state.presentation_artifacts.execute(request, cwd)
state.presentation_artifacts.execute_requests(request, cwd)
}
async fn record_initial_history(&self, conversation_history: InitialHistory) {

View file

@ -2,7 +2,7 @@ use async_trait::async_trait;
use codex_artifact_presentation::PathAccessKind;
use codex_artifact_presentation::PathAccessRequirement;
use codex_artifact_presentation::PresentationArtifactError;
use codex_artifact_presentation::PresentationArtifactRequest;
use codex_artifact_presentation::PresentationArtifactToolRequest;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::ReviewDecision;
use serde_json::to_string;
@ -37,23 +37,10 @@ impl ToolHandler for PresentationArtifactHandler {
let ToolPayload::Function { arguments } = &invocation.payload else {
return true;
};
let Ok(request) = parse_arguments::<PresentationArtifactRequest>(arguments) else {
let Ok(request) = parse_arguments::<PresentationArtifactToolRequest>(arguments) else {
return true;
};
!matches!(
request.action.as_str(),
"get_summary"
| "list_slides"
| "list_layouts"
| "list_layout_placeholders"
| "list_slide_placeholders"
| "inspect"
| "resolve"
| "to_proto"
| "get_style"
| "describe_styles"
| "record_patch"
)
request.is_mutating().unwrap_or(true)
}
async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
@ -80,7 +67,7 @@ impl ToolHandler for PresentationArtifactHandler {
}
};
let request: PresentationArtifactRequest = parse_arguments(&arguments)?;
let request: PresentationArtifactToolRequest = parse_arguments(&arguments)?;
for access in request
.required_path_accesses(&turn.cwd)
.map_err(presentation_error)?
@ -89,7 +76,12 @@ impl ToolHandler for PresentationArtifactHandler {
}
let response = session
.execute_presentation_artifact(request, &turn.cwd)
.execute_presentation_artifact(
request
.into_execution_request()
.map_err(presentation_error)?,
&turn.cwd,
)
.await
.map_err(presentation_error)?;

View file

@ -567,6 +567,26 @@ fn create_view_image_tool() -> ToolSpec {
}
fn create_presentation_artifact_tool() -> ToolSpec {
let action_step_schema = JsonSchema::Object {
properties: BTreeMap::from([
(
"action".to_string(),
JsonSchema::String {
description: Some("Action name to run for this step.".to_string()),
},
),
(
"args".to_string(),
JsonSchema::Object {
properties: BTreeMap::new(),
required: None,
additional_properties: Some(true.into()),
},
),
]),
required: Some(vec!["action".to_string(), "args".to_string()]),
additional_properties: Some(false.into()),
};
let properties = BTreeMap::from([
(
"artifact_id".to_string(),
@ -577,17 +597,12 @@ fn create_presentation_artifact_tool() -> ToolSpec {
},
),
(
"action".to_string(),
JsonSchema::String {
description: Some("Action name to run against the artifact.".to_string()),
},
),
(
"args".to_string(),
JsonSchema::Object {
properties: BTreeMap::new(),
required: None,
additional_properties: Some(true.into()),
"actions".to_string(),
JsonSchema::Array {
items: Box::new(action_step_schema),
description: Some(
"Array of `(action, args)` steps to execute sequentially.".to_string(),
),
},
),
]);
@ -598,7 +613,7 @@ fn create_presentation_artifact_tool() -> ToolSpec {
strict: false,
parameters: JsonSchema::Object {
properties,
required: Some(vec!["action".to_string(), "args".to_string()]),
required: Some(vec!["actions".to_string()]),
additional_properties: Some(false.into()),
},
})

View file

@ -4,6 +4,7 @@ Create and edit PowerPoint presentation artifacts inside the current thread.
- Resume and fork do not restore live artifact state. Export files if you need a durable handoff.
- Relative paths resolve from the current working directory.
- Position and size values are in slide points.
- Every tool call uses a top-level `actions` array of sequential steps. Each call operates on a single top-level `artifact_id` when one is needed. If a call starts with `create` or `import_pptx`, later steps in the same call automatically reuse the returned artifact id.
Supported actions:
- `create`
@ -61,60 +62,63 @@ Supported actions:
- `delete_artifact`
Example create:
`{"action":"create","args":{"name":"Quarterly Update"}}`
`{"actions":[{"action":"create","args":{"name":"Quarterly Update"}}]}`
Example create with custom slide size:
`{"action":"create","args":{"name":"Quarterly Update","slide_size":{"width":960,"height":540}}}`
`{"actions":[{"action":"create","args":{"name":"Quarterly Update","slide_size":{"width":960,"height":540}}}]}`
Example edit:
`{"artifact_id":"presentation_x","action":"add_text_shape","args":{"slide_index":0,"text":"Revenue up 24%","position":{"left":48,"top":72,"width":260,"height":80}}}`
`{"artifact_id":"presentation_x","actions":[{"action":"add_text_shape","args":{"slide_index":0,"text":"Revenue up 24%","position":{"left":48,"top":72,"width":260,"height":80}}}]}`
Example sequential batch:
`{"actions":[{"action":"create","args":{"name":"Quarterly Update"}},{"action":"add_slide","args":{}},{"action":"add_text_shape","args":{"slide_index":0,"text":"Revenue up 24%","position":{"left":48,"top":72,"width":260,"height":80}}}]}`
Table creation also accepts optional `column_widths` and `row_heights` arrays in points when you need explicit table sizing instead of even splits.
Example export:
`{"artifact_id":"presentation_x","action":"export_pptx","args":{"path":"artifacts/q2-update.pptx"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"export_pptx","args":{"path":"artifacts/q2-update.pptx"}}]}`
Example layout flow:
`{"artifact_id":"presentation_x","action":"create_layout","args":{"name":"Title Slide"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"create_layout","args":{"name":"Title Slide"}}]}`
`{"artifact_id":"presentation_x","action":"add_layout_placeholder","args":{"layout_id":"layout_1","name":"title","placeholder_type":"title","text":"Click to add title","position":{"left":48,"top":48,"width":624,"height":72}}}`
`{"artifact_id":"presentation_x","actions":[{"action":"add_layout_placeholder","args":{"layout_id":"layout_1","name":"title","placeholder_type":"title","text":"Click to add title","position":{"left":48,"top":48,"width":624,"height":72}}}]}`
`{"artifact_id":"presentation_x","action":"set_slide_layout","args":{"slide_index":0,"layout_id":"layout_1"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"set_slide_layout","args":{"slide_index":0,"layout_id":"layout_1"}}]}`
`{"artifact_id":"presentation_x","action":"list_layout_placeholders","args":{"layout_id":"layout_1"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"list_layout_placeholders","args":{"layout_id":"layout_1"}}]}`
`{"artifact_id":"presentation_x","action":"list_slide_placeholders","args":{"slide_index":0}}`
`{"artifact_id":"presentation_x","actions":[{"action":"list_slide_placeholders","args":{"slide_index":0}}]}`
Layout references in `create_layout.parent_layout_id`, `add_layout_placeholder.layout_id`, `add_slide`, `insert_slide`, `set_slide_layout`, and `list_layout_placeholders` accept either a layout id or a layout name. Name matching prefers exact id, then exact name, then case-insensitive name.
`insert_slide` accepts `index` or `after_slide_index`. If neither is provided, the new slide is inserted immediately after the active slide, or appended if no active slide is set yet.
Example inspect:
`{"artifact_id":"presentation_x","action":"inspect","args":{"include":"deck,slide,textbox,shape,table,chart,image,notes,layoutList","exclude":"notes","search":"roadmap","max_chars":12000}}`
`{"artifact_id":"presentation_x","actions":[{"action":"inspect","args":{"include":"deck,slide,textbox,shape,table,chart,image,notes,layoutList","exclude":"notes","search":"roadmap","max_chars":12000}}]}`
Example inspect target window:
`{"artifact_id":"presentation_x","action":"inspect","args":{"include":"textbox","target":{"id":"sh/element_3","before_lines":1,"after_lines":1}}}`
`{"artifact_id":"presentation_x","actions":[{"action":"inspect","args":{"include":"textbox","target":{"id":"sh/element_3","before_lines":1,"after_lines":1}}}]}`
Example resolve:
`{"artifact_id":"presentation_x","action":"resolve","args":{"id":"sh/element_3"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"resolve","args":{"id":"sh/element_3"}}]}`
Example proto export:
`{"artifact_id":"presentation_x","action":"to_proto","args":{}}`
`{"artifact_id":"presentation_x","actions":[{"action":"to_proto","args":{}}]}`
`to_proto` returns a full JSON snapshot of the current in-memory presentation document, including slide/layout records, anchors, notes, theme state, and typed element payloads.
Example patch recording:
`{"artifact_id":"presentation_x","action":"record_patch","args":{"operations":[{"action":"add_text_shape","args":{"slide_index":0,"text":"Headline","position":{"left":48,"top":48,"width":320,"height":72}}},{"action":"set_slide_background","args":{"slide_index":0,"fill":"#F7F1E8"}}]}}`
`{"artifact_id":"presentation_x","actions":[{"action":"record_patch","args":{"operations":[{"action":"add_text_shape","args":{"slide_index":0,"text":"Headline","position":{"left":48,"top":48,"width":320,"height":72}}},{"action":"set_slide_background","args":{"slide_index":0,"fill":"#F7F1E8"}}]}}]}`
Example patch application:
`{"artifact_id":"presentation_x","action":"apply_patch","args":{"patch":{"version":1,"artifactId":"presentation_x","operations":[{"action":"add_text_shape","args":{"slide_index":0,"text":"Headline","position":{"left":48,"top":48,"width":320,"height":72}}},{"action":"set_slide_background","args":{"slide_index":0,"fill":"#F7F1E8"}}]}}}`
`{"artifact_id":"presentation_x","actions":[{"action":"apply_patch","args":{"patch":{"version":1,"artifactId":"presentation_x","operations":[{"action":"add_text_shape","args":{"slide_index":0,"text":"Headline","position":{"left":48,"top":48,"width":320,"height":72}}},{"action":"set_slide_background","args":{"slide_index":0,"fill":"#F7F1E8"}}]}}}]}`
Patch payloads are single-artifact and currently support existing in-memory editing actions like slide/element/layout/theme/text updates. Lifecycle, import/export, and nested history actions are intentionally excluded.
Example undo/redo:
`{"artifact_id":"presentation_x","action":"undo","args":{}}`
`{"artifact_id":"presentation_x","actions":[{"action":"undo","args":{}}]}`
`{"artifact_id":"presentation_x","action":"redo","args":{}}`
`{"artifact_id":"presentation_x","actions":[{"action":"redo","args":{}}]}`
Deck summaries, slide listings, `inspect`, and `resolve` now include active-slide metadata. Use `set_active_slide` to change it explicitly.
@ -123,10 +127,10 @@ Theme snapshots and `to_proto` both expose the deck theme hex color map via `hex
Named text styles are supported through `add_style`, `get_style`, and `describe_styles`. Built-in styles include `title`, `heading1`, `body`, `list`, and `numberedList`.
Example style creation:
`{"artifact_id":"presentation_x","action":"add_style","args":{"name":"callout","font_size":18,"color":"#336699","italic":true,"underline":true}}`
`{"artifact_id":"presentation_x","actions":[{"action":"add_style","args":{"name":"callout","font_size":18,"color":"#336699","italic":true,"underline":true}}]}`
Example style lookup:
`{"artifact_id":"presentation_x","action":"get_style","args":{"name":"title"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"get_style","args":{"name":"title"}}]}`
Text styling payloads on `add_text_shape`, `add_shape.text_style`, `update_text.styling`, and `update_table_cell.styling` accept `style` and `underline` in addition to the existing whole-element fields.
@ -155,9 +159,9 @@ Shape strokes accept an optional `style` field such as `solid`, `dashed`, `dotte
Connectors are supported via `add_connector`, with straight/elbow/curved types plus dash styles and arrow heads.
Example preview:
`{"artifact_id":"presentation_x","action":"export_preview","args":{"slide_index":0,"path":"artifacts/q2-update-slide1.png"}}`
`{"artifact_id":"presentation_x","actions":[{"action":"export_preview","args":{"slide_index":0,"path":"artifacts/q2-update-slide1.png"}}]}`
`export_preview` also accepts `format`, `scale`, and `quality` for rendered previews. `format` currently supports `png`, `jpeg`, and `svg`.
Example JPEG preview:
`{"artifact_id":"presentation_x","action":"export_preview","args":{"slide_index":0,"path":"artifacts/q2-update-slide1.jpg","format":"jpeg","scale":0.75,"quality":85}}`
`{"artifact_id":"presentation_x","actions":[{"action":"export_preview","args":{"slide_index":0,"path":"artifacts/q2-update-slide1.jpg","format":"jpeg","scale":0.75,"quality":85}}]}`