The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
feat: normalize ecc2 profiles across harnesses
Affaan Mustafa03e52f49e86d
diff --git a/ecc2/src/session/manager.rs b/ecc2/src/session/manager.rsindex 5c6d4e30f..d63fdf192 100644--- a/ecc2/src/session/manager.rs+++ b/ecc2/src/session/manager.rs@@ -3195,12 +3195,7 @@ fn build_configured_harness_command(}}- let task = if runner.inline_system_prompt_for_task && runner.append_system_prompt_flag.is_none()- {- normalize_task_with_inline_system_prompt(task, profile)- } else {- task.to_string()- };+ let task = normalize_task_for_configured_runner(runner, task, profile);if let Some(flag) = runner.task_flag.as_deref() {command.arg(flag);@@ -3217,24 +3212,143 @@ fn normalize_task_for_harness(task: &str,profile: Option<&SessionAgentProfile>,) -> String {- let rendered = normalize_task_with_inline_system_prompt(task, profile);-match harness {HarnessKind::Claude => task.to_string(),- HarnessKind::Codex | HarnessKind::OpenCode | HarnessKind::Gemini => rendered,+ HarnessKind::Codex => render_task_with_profile_projection(+ task,+ profile,+ TaskProjectionSupport {+ supports_model: true,+ supports_add_dirs: true,+ ..TaskProjectionSupport::default()+ },+ ),+ HarnessKind::OpenCode => render_task_with_profile_projection(+ task,+ profile,+ TaskProjectionSupport {+ supports_model: true,+ ..TaskProjectionSupport::default()+ },+ ),+ HarnessKind::Gemini => render_task_with_profile_projection(+ task,+ profile,+ TaskProjectionSupport {+ supports_model: true,+ supports_add_dirs: true,+ ..TaskProjectionSupport::default()+ },+ ),_ => task.to_string(),}}-fn normalize_task_with_inline_system_prompt(+#[derive(Debug, Default, Clone, Copy)]+struct TaskProjectionSupport {+ supports_model: bool,+ supports_add_dirs: bool,+ supports_allowed_tools: bool,+ supports_disallowed_tools: bool,+ supports_permission_mode: bool,+ supports_max_budget_usd: bool,+ supports_append_system_prompt: bool,+}++fn normalize_task_for_configured_runner(+ runner: &crate::config::HarnessRunnerConfig,task: &str,profile: Option<&SessionAgentProfile>,) -> String {- let Some(system_prompt) = profile.and_then(|profile| profile.append_system_prompt.as_ref())- else {+ render_task_with_profile_projection(+ task,+ profile,+ TaskProjectionSupport {+ supports_model: runner.model_flag.is_some(),+ supports_add_dirs: runner.add_dir_flag.is_some()+ || runner.include_directories_flag.is_some(),+ supports_allowed_tools: runner.allowed_tools_flag.is_some(),+ supports_disallowed_tools: runner.disallowed_tools_flag.is_some(),+ supports_permission_mode: runner.permission_mode_flag.is_some(),+ supports_max_budget_usd: runner.max_budget_usd_flag.is_some(),+ supports_append_system_prompt: runner.append_system_prompt_flag.is_some()+ && !runner.inline_system_prompt_for_task,+ },+ )+}++fn render_task_with_profile_projection(+ task: &str,+ profile: Option<&SessionAgentProfile>,+ support: TaskProjectionSupport,+) -> String {+ let Some(profile) = profile else {return task.to_string();};- format!("System instructions:\n{system_prompt}\n\nTask:\n{task}")++ let mut sections = Vec::new();+ if !support.supports_append_system_prompt {+ if let Some(system_prompt) = profile.append_system_prompt.as_ref() {+ sections.push(format!("System instructions:\n{system_prompt}"));+ }+ }++ let mut directives = Vec::new();+ if !support.supports_model {+ if let Some(model) = profile.model.as_ref() {+ directives.push(format!("Preferred model: {model}"));+ }+ }+ if !support.supports_add_dirs && !profile.add_dirs.is_empty() {+ directives.push(format!(+ "Additional context dirs: {}",+ profile+ .add_dirs+ .iter()+ .map(|dir| dir.to_string_lossy().to_string())+ .collect::<Vec<_>>()+ .join(", ")+ ));+ }+ if !support.supports_allowed_tools && !profile.allowed_tools.is_empty() {+ directives.push(format!(+ "Allowed tools: {}",+ profile.allowed_tools.join(", ")+ ));+ }+ if !support.supports_disallowed_tools && !profile.disallowed_tools.is_empty() {+ directives.push(format!(+ "Disallowed tools: {}",+ profile.disallowed_tools.join(", ")+ ));+ }+ if !support.supports_permission_mode {+ if let Some(permission_mode) = profile.permission_mode.as_ref() {+ directives.push(format!("Permission mode: {permission_mode}"));+ }+ }+ if !support.supports_max_budget_usd {+ if let Some(max_budget_usd) = profile.max_budget_usd {+ directives.push(format!("Max budget USD: {max_budget_usd}"));+ }+ }+ if let Some(token_budget) = profile.token_budget {+ directives.push(format!("Token budget: {token_budget}"));+ }++ if !directives.is_empty() {+ sections.push(format!(+ "ECC execution profile:\n- {}",+ directives.join("\n- ")+ ));+ }++ if sections.is_empty() {+ return task.to_string();+ }++ sections.push(format!("Task:\n{task}"));+ sections.join("\n\n")}async fn spawn_claude_code(@@ -4125,7 +4239,7 @@ mod tests {"docs","--add-dir","specs",- "System instructions:\nReview thoroughly.\n\nTask:\nreview this change",+ "System instructions:\nReview thoroughly.\n\nECC execution profile:\n- Allowed tools: Read\n- Disallowed tools: Bash\n- Permission mode: plan\n- Max budget USD: 1.25\n- Token budget: 750\n\nTask:\nreview this change",]);}@@ -4171,7 +4285,7 @@ mod tests {"ecc-sess-9999","--model","anthropic/claude-sonnet-4",- "System instructions:\nBuild carefully.\n\nTask:\nstabilize callback flow",+ "System instructions:\nBuild carefully.\n\nECC execution profile:\n- Additional context dirs: docs\n\nTask:\nstabilize callback flow",]);}@@ -4215,7 +4329,7 @@ mod tests {"gemini-2.5-pro","--include-directories","docs,../shared",- "System instructions:\nUse repo context carefully.\n\nTask:\ninvestigate auth regression",+ "System instructions:\nUse repo context carefully.\n\nECC execution profile:\n- Allowed tools: Read\n- Disallowed tools: Bash\n- Permission mode: plan\n- Max budget USD: 1\n- Token budget: 500\n\nTask:\ninvestigate auth regression",]);}@@ -4343,6 +4457,59 @@ mod tests {);}+ #[test]+ fn build_agent_command_projects_unsupported_profile_fields_for_configured_runner() {+ let mut cfg = Config::default();+ cfg.harness_runners.insert(+ "cursor".to_string(),+ crate::config::HarnessRunnerConfig {+ program: "cursor-agent".to_string(),+ base_args: vec!["run".to_string()],+ task_flag: Some("--task".to_string()),+ model_flag: Some("--model".to_string()),+ ..Default::default()+ },+ );+ let profile = SessionAgentProfile {+ profile_name: "worker".to_string(),+ agent: None,+ model: Some("gpt-5.4".to_string()),+ allowed_tools: vec!["Read".to_string()],+ disallowed_tools: vec!["Bash".to_string()],+ permission_mode: Some("plan".to_string()),+ add_dirs: vec![PathBuf::from("docs"), PathBuf::from("specs")],+ max_budget_usd: Some(2.5),+ token_budget: Some(900),+ append_system_prompt: Some("Use repo context carefully.".to_string()),+ };++ let command = build_agent_command(+ &cfg,+ "cursor",+ Path::new("cursor-agent"),+ "fix callback regression",+ "sess-cur2",+ Path::new("/tmp/repo"),+ Some(&profile),+ );+ let args = command+ .as_std()+ .get_args()+ .map(|value| value.to_string_lossy().to_string())+ .collect::<Vec<_>>();++ assert_eq!(+ args,+ vec![+ "run",+ "--model",+ "gpt-5.4",+ "--task",+ "System instructions:\nUse repo context carefully.\n\nECC execution profile:\n- Additional context dirs: docs, specs\n- Allowed tools: Read\n- Disallowed tools: Bash\n- Permission mode: plan\n- Max budget USD: 2.5\n- Token budget: 900\n\nTask:\nfix callback regression",+ ]+ );+ }+#[test]fn build_session_record_canonicalizes_known_agent_aliases() -> Result<()> {let tempdir = TestDir::new("manager-canonical-agent-type")?;