1use gh_workflow::{Event, Expression, Job, Run, Step, Strategy, Use, Workflow, WorkflowDispatch};
2use serde_json::json;
3
4use crate::tasks::workflows::{
5 runners::{self, Platform},
6 steps::{self, FluentBuilder as _, NamedJob, named, setup_cargo_config},
7 vars::{self, WorkflowInput},
8};
9
10pub(crate) fn run_agent_evals() -> Workflow {
11 let agent_evals = agent_evals();
12 let model_name = WorkflowInput::string("model_name", None);
13
14 named::workflow()
15 .on(Event::default().workflow_dispatch(
16 WorkflowDispatch::default().add_input(model_name.name, model_name.input()),
17 ))
18 .concurrency(vars::one_workflow_per_non_main_branch())
19 .add_env(("CARGO_TERM_COLOR", "always"))
20 .add_env(("CARGO_INCREMENTAL", 0))
21 .add_env(("RUST_BACKTRACE", 1))
22 .add_env(("ANTHROPIC_API_KEY", vars::ANTHROPIC_API_KEY))
23 .add_env(("OPENAI_API_KEY", vars::OPENAI_API_KEY))
24 .add_env(("GOOGLE_AI_API_KEY", vars::GOOGLE_AI_API_KEY))
25 .add_env(("GOOGLE_CLOUD_PROJECT", vars::GOOGLE_CLOUD_PROJECT))
26 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
27 .add_env(("ZED_EVAL_TELEMETRY", 1))
28 .add_env(("MODEL_NAME", model_name.to_string()))
29 .add_job(agent_evals.name, agent_evals.job)
30}
31
32pub(crate) fn run_unit_evals() -> Workflow {
33 let model_name = WorkflowInput::string("model_name", None);
34 let commit_sha = WorkflowInput::string("commit_sha", None);
35
36 let unit_evals = named::job(unit_evals(Some(&commit_sha)));
37
38 named::workflow()
39 .name("run_unit_evals")
40 .on(Event::default().workflow_dispatch(
41 WorkflowDispatch::default()
42 .add_input(model_name.name, model_name.input())
43 .add_input(commit_sha.name, commit_sha.input()),
44 ))
45 .concurrency(vars::allow_concurrent_runs())
46 .add_env(("CARGO_TERM_COLOR", "always"))
47 .add_env(("CARGO_INCREMENTAL", 0))
48 .add_env(("RUST_BACKTRACE", 1))
49 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
50 .add_env(("ZED_EVAL_TELEMETRY", 1))
51 .add_env(("MODEL_NAME", model_name.to_string()))
52 .add_job(unit_evals.name, unit_evals.job)
53}
54
55fn add_api_keys(step: Step<Run>) -> Step<Run> {
56 step.add_env(("ANTHROPIC_API_KEY", vars::ANTHROPIC_API_KEY))
57 .add_env(("OPENAI_API_KEY", vars::OPENAI_API_KEY))
58 .add_env(("GOOGLE_AI_API_KEY", vars::GOOGLE_AI_API_KEY))
59 .add_env(("GOOGLE_CLOUD_PROJECT", vars::GOOGLE_CLOUD_PROJECT))
60}
61
62fn agent_evals() -> NamedJob {
63 fn run_eval() -> Step<Run> {
64 named::bash(
65 "cargo run --package=eval -- --repetitions=8 --concurrency=1 --model \"${MODEL_NAME}\"",
66 )
67 }
68
69 named::job(
70 Job::default()
71 .runs_on(runners::LINUX_DEFAULT)
72 .timeout_minutes(60_u32 * 10)
73 .add_step(steps::checkout_repo())
74 .add_step(steps::cache_rust_dependencies_namespace())
75 .map(steps::install_linux_dependencies)
76 .add_step(setup_cargo_config(Platform::Linux))
77 .add_step(steps::script("cargo build --package=eval"))
78 .add_step(add_api_keys(run_eval()))
79 .add_step(steps::cleanup_cargo_config(Platform::Linux)),
80 )
81}
82
83pub(crate) fn run_cron_unit_evals() -> Workflow {
84 let unit_evals = cron_unit_evals();
85
86 named::workflow()
87 .name("run_cron_unit_evals")
88 .on(Event::default()
89 // .schedule([
90 // // GitHub might drop jobs at busy times, so we choose a random time in the middle of the night.
91 // Schedule::default().cron("47 1 * * 2"),
92 // ])
93 .workflow_dispatch(WorkflowDispatch::default()))
94 .concurrency(vars::one_workflow_per_non_main_branch())
95 .add_env(("CARGO_TERM_COLOR", "always"))
96 .add_env(("CARGO_INCREMENTAL", 0))
97 .add_env(("RUST_BACKTRACE", 1))
98 .add_env(("ZED_CLIENT_CHECKSUM_SEED", vars::ZED_CLIENT_CHECKSUM_SEED))
99 .add_job(unit_evals.name, unit_evals.job)
100}
101
102fn cron_unit_evals() -> NamedJob {
103 fn send_failure_to_slack() -> Step<Use> {
104 named::uses(
105 "slackapi",
106 "slack-github-action",
107 "b0fa283ad8fea605de13dc3f449259339835fc52",
108 )
109 .if_condition(Expression::new("${{ failure() }}"))
110 .add_with(("method", "chat.postMessage"))
111 .add_with(("token", vars::SLACK_APP_ZED_UNIT_EVALS_BOT_TOKEN))
112 .add_with(("payload", indoc::indoc!{r#"
113 channel: C04UDRNNJFQ
114 text: "Unit Evals Failed: https://github.com/zed-industries/zed/actions/runs/${{ github.run_id }}"
115 "#}))
116 }
117
118 named::job(cron_unit_evals_job().add_step(send_failure_to_slack()))
119}
120
121const UNIT_EVAL_MODELS: &[&str] = &[
122 "anthropic/claude-sonnet-4-5-latest",
123 "anthropic/claude-opus-4-5-latest",
124 "google/gemini-3-pro",
125 "openai/gpt-5",
126];
127
128fn cron_unit_evals_job() -> Job {
129 let script_step = add_api_keys(steps::script("./script/run-unit-evals"))
130 .add_env(("ZED_AGENT_MODEL", "${{ matrix.model }}"));
131
132 Job::default()
133 .runs_on(runners::LINUX_DEFAULT)
134 .strategy(Strategy::default().fail_fast(false).matrix(json!({
135 "model": UNIT_EVAL_MODELS
136 })))
137 .add_step(steps::checkout_repo())
138 .add_step(steps::setup_cargo_config(Platform::Linux))
139 .add_step(steps::cache_rust_dependencies_namespace())
140 .map(steps::install_linux_dependencies)
141 .add_step(steps::cargo_install_nextest())
142 .add_step(steps::clear_target_dir_if_large(Platform::Linux))
143 .add_step(script_step)
144 .add_step(steps::cleanup_cargo_config(Platform::Linux))
145}
146
147fn unit_evals(commit: Option<&WorkflowInput>) -> Job {
148 let script_step = add_api_keys(steps::script("./script/run-unit-evals"));
149
150 Job::default()
151 .runs_on(runners::LINUX_DEFAULT)
152 .add_step(steps::checkout_repo())
153 .add_step(steps::setup_cargo_config(Platform::Linux))
154 .add_step(steps::cache_rust_dependencies_namespace())
155 .map(steps::install_linux_dependencies)
156 .add_step(steps::cargo_install_nextest())
157 .add_step(steps::clear_target_dir_if_large(Platform::Linux))
158 .add_step(match commit {
159 Some(commit) => script_step.add_env(("UNIT_EVAL_COMMIT", commit)),
160 None => script_step,
161 })
162 .add_step(steps::cleanup_cargo_config(Platform::Linux))
163}