Functions — github.com/canwhite/AgentEval

hub / github.com/canwhite/AgentEval / functions

Functions112 in github.com/canwhite/AgentEval

⨍ Functions 112◇ Types & classes 45

↓ 7 callersFunctionis_safe_session_id

(s: &str)

src/web/mod.rs:200

↓ 7 callersFunctionresolve_path

Resolve a tool argument path relative to the source project directory. Returns an error if the path contains `..` or resolves outside the source dir.

src/probe/tools.rs:15

↓ 5 callersMethodbuild

(&self)

src/eval/mod.rs:216

↓ 5 callersFunctionrun

( mut rx: UnboundedReceiver<TurnRecord>, log_dir: String, jsonl_stem: String, grader_config: G

src/eval/mod.rs:15

↓ 4 callersMethodregister

(&mut self, tool: Box<dyn Tool>)

src/probe/tool.rs:23

↓ 3 callersFunctionextract_json_block

从 LLM 输出中提取 JSON（可能被 ```json ... ``` 包裹）

src/grader/judge.rs:66

↓ 3 callersFunctionextract_json_block

Extract a JSON block from text that may be wrapped in markdown fences.

src/probe/mod.rs:153

↓ 3 callersFunctionparse_probe_output

Parse the LLM's raw output into a ProbeReport. Tries strict JSON parsing first, then falls back to extracting a JSON block from markdown-wrapped outp

src/probe/mod.rs:118

↓ 3 callersFunctionread_existing

Read an existing .probe.json.

src/probe/mod.rs:106

↓ 3 callersFunctionrun_pipeline

运行完整的 grading pipeline: 1. 规则统计 → MetricsSnapshot 2. LLM 评审 → task_completion + response_quality 3. 汇总加权 → GradeReport

src/grader/mod.rs:16

↓ 3 callersFunctionwrite_json_file

(path: &str, value: &impl serde::Serialize)

src/eval/mod.rs:284

↓ 2 callersFunctionbody_to_json

(body: &[u8])

src/proxy.rs:220

↓ 2 callersFunctionbuild_with_id

(builder: &SessionBuilder, session_id: &str)

src/eval/mod.rs:256

↓ 2 callersFunctioncommon_prefix_len

(prev: &[Value], current: &[Value])

src/eval/mod.rs:238

↓ 2 callersFunctionextract_text_content

从 message value 提取可读文本（content 可能是 string 或 array）

src/format/openai.rs:249

↓ 2 callersFunctionheaders_to_value

(headers: &HeaderMap)

src/proxy.rs:210

↓ 2 callersFunctionparse_request_messages

从 request body 提取 messages 数组

src/format/openai.rs:6

↓ 2 callersFunctionread_diagnose_summary

(log_dir: &str, session_id: &str)

src/web/mod.rs:204

↓ 2 callersFunctionread_probe_summary

(log_dir: &str, session_id: &str)

src/web/mod.rs:216

↓ 2 callersFunctionwrite_grade_json

(report: &grader::types::GradeReport, log_dir: &str)

src/eval/mod.rs:279

↓ 2 callersFunctionwrite_view_final

(view: &SessionView, log_dir: &str)

src/eval/mod.rs:274

↓ 1 callersMethodas_openai_tools

Return the tool definitions in OpenAI function-calling JSON format.

src/probe/tool.rs:36

↓ 1 callersFunctionbuild_judge_prompt

构造完整的 LLM 评审 prompt

src/grader/prompt.rs:67

↓ 1 callersFunctionbuild_system_prompt

Build the system prompt that defines the probe agent's role and methodology.

src/probe/prompt.rs:10

↓ 1 callersFunctionbuild_user_prompt

Build the user prompt: diagnose issues (JSON) + session summary (compact text).

src/probe/prompt.rs:15

↓ 1 callersFunctioncalc_performance

性能评分（0.0~1.0）

src/grader/rules.rs:110

↓ 1 callersFunctioncalc_tool_efficiency

工具效率评分（0.0~1.0）

src/grader/rules.rs:82

↓ 1 callersMethodcall

(&self, args: Value)

src/probe/tools.rs:87

↓ 1 callersMethodchat

( &self, messages: &[Message], tools: &Value, )

src/probe/backend.rs:31

↓ 1 callersFunctiondiff_messages

(prev: &[Value], current: &'a [Value])

src/eval/mod.rs:229

↓ 1 callersMethoddispatch

(&self, name: &str, args: Value)

src/probe/tool.rs:27

↓ 1 callersFunctionextract_content

Extract text content from a message value (content may be string or array of blocks).

src/diagnose/rules.rs:617

↓ 1 callersFunctionextract_metrics

(view: &SessionView)

src/grader/rules.rs:4

↓ 1 callersFunctionextract_model

(body: &Value)

src/eval/mod.rs:249

↓ 1 callersFunctionformat_session

把 SessionView 格式化成 LLM 易读的会话摘要

src/grader/prompt.rs:5

↓ 1 callersFunctionformat_session_summary

Build a compact session summary from SessionView. - diagnose-marked turns are expanded with detail - normal turns get a one-line summary - large gaps

src/probe/prompt.rs:32

↓ 1 callersMethodhas_tool_calls

(&self)

src/probe/types.rs:123

↓ 1 callersFunctionis_new_session

(prev: &[Value], current: &[Value])

src/eval/mod.rs:245

↓ 1 callersFunctionis_streaming_request

(body: &[u8])

src/proxy.rs:230

↓ 1 callersFunctionjudge

调用评测 LLM，返回 task_completion 和 response_quality 评分

src/grader/judge.rs:8

↓ 1 callersFunctionjudge_llm

( view: &SessionView, metrics: &types::MetricsSnapshot, config: &GraderConfig, )

src/grader/mod.rs:106

↓ 1 callersMethodname

(&self)

src/probe/tools.rs:65

↓ 1 callersFunctionnormalize_args

Normalize JSON arguments for comparison: parse + re-serialize with sorted keys.

src/diagnose/rules.rs:590

↓ 1 callersFunctionparse_json_response

(body: &str)

src/probe/types.rs:178

↓ 1 callersFunctionparse_llm_response

Convert an OpenAI chat completion response body into an LlmResponse. Handles both streaming (SSE) and non-streaming (JSON) responses.

src/probe/types.rs:170

↓ 1 callersFunctionparse_non_streaming

(body: &Value)

src/format/openai.rs:61

↓ 1 callersFunctionparse_response_steps

从 response body 提取 model 步骤自动区分非流式（JSON）和流式（SSE 文本）响应。

src/format/openai.rs:16

↓ 1 callersFunctionparse_response_usage

从 response body 提取 usage 信息

src/format/openai.rs:25

↓ 1 callersFunctionparse_sse

(sse_text: &str)

src/format/openai.rs:143

↓ 1 callersFunctionparse_sse_response

(sse_text: &str)

src/probe/types.rs:235

↓ 1 callersMethodprocess

( &mut self, request_body: &Value, response_body: &Value, duration_ms: u64,

src/eval/mod.rs:128

↓ 1 callersFunctionprompt_bloat

System prompt content > 3000 chars.

src/diagnose/rules.rs:249

↓ 1 callersFunctionprompt_context_overflow

Signs of history truncation: tool role message with no matching tool_calls in preceding assistant messages within the same request.

src/diagnose/rules.rs:296

↓ 1 callersFunctionread_jsonl_entries

Read specific JSONL entries by their `id` fields.

src/diagnose/mod.rs:173

↓ 1 callersFunctionread_raw_jsonl

Read raw JSONL lines by IDs (for the raw JSONL API).

src/diagnose/mod.rs:209

↓ 1 callersMethodrun

Run the agent loop. Returns the final text content from the LLM (which should be the JSON probe report).

src/probe/agent.rs:48

↓ 1 callersFunctionrun_all

(view: &SessionView, jsonl_entries: &[JsonlEntry])

src/diagnose/rules.rs:18

↓ 1 callersFunctionrun_diagnose_cli

(args: &[String])

src/main.rs:82

↓ 1 callersFunctionrun_probe_cli

(args: &[String])

src/main.rs:158

↓ 1 callersFunctionseal_and_grade_bg

封口当前 session：立即写 view.json，后台 spawn 评分任务（不阻塞主循环）

src/eval/mod.rs:293

↓ 1 callersFunctionsse_has_content

Check if an SSE stream has any meaningful content (text or tool calls).

src/diagnose/rules.rs:430

↓ 1 callersFunctionsummarize_issues

Call the judge LLM to generate a 2-3 sentence summary of diagnose issues. Returns None if no API key is configured, there are no issues, or the LLM ca

src/diagnose/mod.rs:75

↓ 1 callersFunctiontimestamp_ms

()

src/proxy.rs:202

↓ 1 callersFunctiontoken_empty_response

response_body.choices empty or content empty (handles both SSE string and JSON object).

src/diagnose/rules.rs:363

↓ 1 callersFunctiontoken_excessive_input

input > 30000 AND output < 1000.

src/diagnose/rules.rs:519

↓ 1 callersFunctiontoken_waste

output/input < 2% AND input > 5000 tokens.

src/diagnose/rules.rs:474

↓ 1 callersFunctiontool_duplicate_3plus

Same name + same arguments called ≥ 3 times.

src/diagnose/rules.rs:137

↓ 1 callersFunctiontool_result_empty

tool_result.content is empty or whitespace-only.

src/diagnose/rules.rs:190

↓ 1 callersFunctiontool_result_error

tool_result.is_error == true.

src/diagnose/rules.rs:86

↓ 1 callersFunctiontool_result_missing

tool_call.result is None (not backfilled cross-turn).

src/diagnose/rules.rs:45

↓ 1 callersFunctionview_mismatch

Data integrity: turns.len() != jsonl_ids.len().

src/diagnose/rules.rs:558

↓ 1 callersFunctionwrite_view_json

(builder: &SessionBuilder, log_dir: &str)

src/eval/mod.rs:268

Functionaccepts_normal_path

()

src/probe/tools.rs:393

Methodassistant_with_tool_calls

(tool_calls: Vec<ToolCall>)

src/probe/types.rs:101

src/probe/tools.rs:69

Functiondiagnose_session

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:231

Methodfrom_issues

(issues: &[DiagnoseIssue])

src/diagnose/types.rs:22

Methodfrom_report

(report: &ProbeReport)

src/probe/types.rs:46

Functionget_diagnose

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:246

Functionget_probe

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:305

Functionget_raw_jsonl

( State(state): State<Arc<AppState>>, Path(jsonl_stem): Path<String>, axum::extract::Query(query):

src/web/mod.rs:265

Functionget_session

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:142

Functiongrade_session

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:168

Functionhandler

( State(state): State<std::sync::Arc<AppState>>, method: Method, uri: Uri, headers: HeaderMap,

src/proxy.rs:62

Functionlist_sessions

( State(state): State<Arc<AppState>>, )

src/web/mod.rs:64

Methodload

加载配置，优先从环境变量获取，如果没有则使用默认值

(config: &Config, eval_tx: tokio::sync::mpsc::UnboundedSender<TurnRecord>, grader_config: GraderConfig, probe_

src/proxy.rs:34

Methodnew

(session_id: String, model: String)

src/eval/mod.rs:116

Methodnew

(source_dir: &str)

src/probe/tools.rs:57

Methodnew

( backend: OpenAiBackend, tools: Registry, system_prompt: &str, user_prompt: &

src/probe/agent.rs:26

Methodnew

(api_base: &str, model: &str, api_key: &str)

src/probe/backend.rs:18

Methodnew

()

src/probe/tool.rs:19

Functionprobe_session

( State(state): State<Arc<AppState>>, Path(session_id): Path<String>, )

src/web/mod.rs:290

Functionread_existing

Read an existing .diagnose.json.

src/diagnose/mod.rs:65

Functionrejects_dotdot_in_middle

()

src/probe/tools.rs:386

Functionrejects_parent_traversal

()

src/probe/tools.rs:379

Functionrun

Run all diagnostic rules against a session. Reads `{log_dir}/{session_id}.view.json` and the corresponding `.jsonl`, runs all 10 rules, writes `{log_

src/diagnose/mod.rs:18

Functionrun

Run the probe agent against a session. 1. Read diagnose issues from .diagnose.json 2. Read session view from .view.json 3. Build system + user prompt

src/probe/mod.rs:29

next →1–100 of 112, ranked by callers