From f04c07da810789bf5cd431d5ca60e112913c7adc Mon Sep 17 00:00:00 2001 From: Jussi Elo Date: Thu, 5 Nov 2026 08:38:30 +0000 Subject: [PATCH 1/3] ROCMAI-864: rename SKILL_NAMES to TELEMETRY_TOOL_NAMES After the dash-tui skills subsystem was removed, SKILL_NAMES held only the four telemetry tools (gpu_status, list_instances, bench_summary, tokens_per_watt), so the name described something that no longer exists. Rename the constant, its re-export in agent/mod.rs and every user in agent/tools.rs and agent/clients.rs (including the wire-level registry test), and fix its doc comment. The tool name strings themselves are untouched, so the names sent to the model are unchanged. Signed-off-by: Jussi Elo --- crates/rocm-dash-tui/src/agent/clients.rs | 14 +++++++++----- crates/rocm-dash-tui/src/agent/mod.rs | 4 ++-- crates/rocm-dash-tui/src/agent/tools.rs | 23 +++++++++++++---------- 3 files changed, 24 insertions(+), 17 deletions(-) diff --git a/crates/rocm-dash-tui/src/agent/clients.rs b/crates/rocm-dash-tui/src/agent/clients.rs index 949f8ce048..27801c94e8 100644 --- a/crates/rocm-dash-tui/src/agent/clients.rs +++ b/crates/rocm-dash-tui/src/agent/clients.rs @@ -849,7 +849,7 @@ mod tests { use serde_json::{Value, json}; use crate::agent::{ - ROCM_MUTATING_TOOL_NAMES, ROCM_READ_TOOL_NAMES, SKILL_NAMES, fixture_snapshot, + ROCM_MUTATING_TOOL_NAMES, ROCM_READ_TOOL_NAMES, TELEMETRY_TOOL_NAMES, fixture_snapshot, }; /// A throwaway, never-pre-existing directory for @@ -1704,11 +1704,15 @@ mod tests { // asserts those tools are actually sent on the wire from that one site. // // Telemetry set: register_telemetry_tools registers exactly the - // SKILL_NAMES tools (GpuStatus, ListInstances, BenchSummary, + // TELEMETRY_TOOL_NAMES tools (GpuStatus, ListInstances, BenchSummary, // TokensPerWatt). Pinning the size here means the telemetry // registration can't silently diverge from the canonical set. - assert_eq!(SKILL_NAMES.len(), 4, "canonical telemetry set size"); - for n in SKILL_NAMES { + assert_eq!( + TELEMETRY_TOOL_NAMES.len(), + 4, + "canonical telemetry set size" + ); + for n in TELEMETRY_TOOL_NAMES { assert!(!n.is_empty(), "empty telemetry/skill tool name"); // Telemetry tools are disjoint from both ROCm sets. assert!( @@ -1807,7 +1811,7 @@ mod tests { // outside the three name tables (the retired `list_skills` and // `skill_plan` included), and nothing in them left unregistered. A new // tool is added to its table in the same change, which keeps this true. - let expected_tool_names: std::collections::BTreeSet<&str> = SKILL_NAMES + let expected_tool_names: std::collections::BTreeSet<&str> = TELEMETRY_TOOL_NAMES .iter() .chain(ROCM_READ_TOOL_NAMES.iter()) .chain(ROCM_MUTATING_TOOL_NAMES.iter()) diff --git a/crates/rocm-dash-tui/src/agent/mod.rs b/crates/rocm-dash-tui/src/agent/mod.rs index d64b20c14e..e79a0a90b8 100644 --- a/crates/rocm-dash-tui/src/agent/mod.rs +++ b/crates/rocm-dash-tui/src/agent/mod.rs @@ -57,8 +57,8 @@ pub use tools::{ ExamineRocmTool, GpuSnapshotRocmTool, GpuStatusArgs, GpuStatusTool, InstallEngineRocmTool, InstallSdkDryRunRocmTool, InstallSdkRocmTool, LaunchServerRocmTool, ListInstancesTool, NaturalLanguagePlanRocmTool, NoArgs, PathExistsRocmTool, PortStatusRocmTool, - ROCM_MUTATING_TOOL_NAMES, ROCM_READ_TOOL_NAMES, RocmCommandRocmTool, SKILL_NAMES, - ServiceLogsRocmTool, ServicesRocmTool, StopServerRocmTool, TokensPerWattTool, ToolError, + ROCM_MUTATING_TOOL_NAMES, ROCM_READ_TOOL_NAMES, RocmCommandRocmTool, ServiceLogsRocmTool, + ServicesRocmTool, StopServerRocmTool, TELEMETRY_TOOL_NAMES, TokensPerWattTool, ToolError, UpdateCheckRocmTool, WatcherDisableRocmTool, WatcherEnableRocmTool, }; diff --git a/crates/rocm-dash-tui/src/agent/tools.rs b/crates/rocm-dash-tui/src/agent/tools.rs index 113ece0518..5e38310cf0 100644 --- a/crates/rocm-dash-tui/src/agent/tools.rs +++ b/crates/rocm-dash-tui/src/agent/tools.rs @@ -178,8 +178,8 @@ impl Tool for TokensPerWattTool { } } -/// All Skill names, for definition/uniqueness checks and docs. -pub const SKILL_NAMES: [&str; 4] = [ +/// All telemetry tool names, for definition/uniqueness checks and docs. +pub const TELEMETRY_TOOL_NAMES: [&str; 4] = [ GpuStatusTool::NAME, ListInstancesTool::NAME, BenchSummaryTool::NAME, @@ -427,7 +427,7 @@ rocm_read_tool!( } ); -/// All read-only ROCm tool names (mirrors [`SKILL_NAMES`]). +/// All read-only ROCm tool names (mirrors [`TELEMETRY_TOOL_NAMES`]). /// /// Used for uniqueness/registration checks and the parity map. Mutating tools /// are intentionally absent. `natural_language_plan` is read-only: it plans but @@ -787,15 +787,15 @@ mod tests { #[test] fn skill_names_are_unique_and_non_empty() { - for n in SKILL_NAMES { + for n in TELEMETRY_TOOL_NAMES { assert!(!n.is_empty(), "skill name must be non-empty"); } - let mut sorted = SKILL_NAMES.to_vec(); + let mut sorted = TELEMETRY_TOOL_NAMES.to_vec(); sorted.sort_unstable(); sorted.dedup(); assert_eq!( sorted.len(), - SKILL_NAMES.len(), + TELEMETRY_TOOL_NAMES.len(), "skill names must be unique" ); } @@ -860,7 +860,7 @@ mod tests { // installs an engine, and no executor ever applied their plans. They // must not come back in any tool set the model is offered. for gone in ["list_skills", "skill_plan"] { - assert!(!SKILL_NAMES.contains(&gone), "{gone} is retired"); + assert!(!TELEMETRY_TOOL_NAMES.contains(&gone), "{gone} is retired"); assert!(!ROCM_READ_TOOL_NAMES.contains(&gone), "{gone} is retired"); assert!( !ROCM_MUTATING_TOOL_NAMES.contains(&gone), @@ -1015,7 +1015,10 @@ mod tests { } // Disjoint from read-only + skill names (iterates the registry itself). for n in ROCM_MUTATING_TOOL_NAMES { - assert!(!SKILL_NAMES.contains(&n), "collision with skill: {n}"); + assert!( + !TELEMETRY_TOOL_NAMES.contains(&n), + "collision with skill: {n}" + ); // install_sdk_dry_run (read-only) must not clash with install_sdk. assert!( !ROCM_READ_TOOL_NAMES.contains(&n), @@ -1232,9 +1235,9 @@ mod tests { // natural_language_plan is read-only (plans, never executes) — Phase 7. assert!(ROCM_READ_TOOL_NAMES.contains(&"natural_language_plan")); // The telemetry/skill tools are still present and disjoint from the new set. - assert!(SKILL_NAMES.contains(&"gpu_status")); + assert!(TELEMETRY_TOOL_NAMES.contains(&"gpu_status")); for n in ROCM_READ_TOOL_NAMES { - assert!(!SKILL_NAMES.contains(&n), "name collision: {n}"); + assert!(!TELEMETRY_TOOL_NAMES.contains(&n), "name collision: {n}"); } } } From 4e6de32cc0b9ece713f4fbb0d7412ec0cbe1e1db Mon Sep 17 00:00:00 2001 From: Jussi Elo Date: Thu, 5 Nov 2026 08:43:53 +0000 Subject: [PATCH 2/3] ROCMAI-864: drop the stale "Skill" wording for the telemetry tools The four telemetry tools (gpu_status, list_instances, bench_summary, tokens_per_watt) were called skills in comments, test names and messages, and in docs/architecture.md. Reword those to plain "tool" wording, name the four tools and the TELEMETRY_TOOL_NAMES table in the architecture entry for agent/tools.rs, and point the lemonade port comment at the chat probe in rocm_dash_tui::llm instead of the removed rocm_dash_tui::skills module. Renamed test leaf names, all in rocm-dash-tui: - skill_names_are_unique_and_non_empty -> telemetry_tool_names_are_unique_and_non_empty - rocm_read_tool_names_are_complete_and_disjoint_from_skills -> rocm_read_tool_names_are_complete_and_disjoint_from_telemetry_tools - annotate_reply_appends_deduped_skills -> annotate_reply_appends_deduped_tools - mock_tool_calling_answer_cites_skill -> mock_tool_calling_answer_cites_tool - transcript_renders_skill_annotation_on_its_own_line -> transcript_renders_tool_annotation_on_its_own_line The tracing field skills_fired on the "chat request complete" event is now tools_fired. The tool name strings sent to the model, the "via:" annotation text and every other behaviour are unchanged. Signed-off-by: Jussi Elo --- crates/rocm-dash-collectors/src/lemonade.rs | 5 ++- crates/rocm-dash-tui/src/agent/clients.rs | 42 ++++++++++----------- crates/rocm-dash-tui/src/agent/tools.rs | 24 ++++++------ crates/rocm-dash-tui/src/ui/tabs/chat.rs | 6 +-- docs/architecture.md | 2 +- 5 files changed, 40 insertions(+), 39 deletions(-) diff --git a/crates/rocm-dash-collectors/src/lemonade.rs b/crates/rocm-dash-collectors/src/lemonade.rs index 65e58a0559..d556aafc13 100644 --- a/crates/rocm-dash-collectors/src/lemonade.rs +++ b/crates/rocm-dash-collectors/src/lemonade.rs @@ -25,8 +25,9 @@ use rocm_dash_core::traits::{ }; use serde::Deserialize; -/// Lemonade's default OpenAI-compatible port. Mirrors `rocm_dash_tui::skills` -/// (defined locally to avoid a cross-crate dep from collectors → tui). +/// Lemonade's default OpenAI-compatible port. Mirrors the endpoint the +/// dash-tui chat probe uses (`rocm_dash_tui::llm`), defined locally to avoid a +/// cross-crate dep from collectors → tui. pub const LEMONADE_PORT: u16 = 13305; /// The runtime-stats path on the canonical `/api/v1` base. pub const LEMONADE_STATS_PATH: &str = "/api/v1/stats"; diff --git a/crates/rocm-dash-tui/src/agent/clients.rs b/crates/rocm-dash-tui/src/agent/clients.rs index 27801c94e8..0237c706e9 100644 --- a/crates/rocm-dash-tui/src/agent/clients.rs +++ b/crates/rocm-dash-tui/src/agent/clients.rs @@ -150,14 +150,14 @@ pub fn build_messages(turns: &[ChatTurn]) -> Vec { .collect() } -/// Append a "via: tool, tool" annotation so the operator can see which Skills +/// Append a "via: tool, tool" annotation so the operator can see which tools /// fired. Deduplicates, preserving first-seen order. Pure. -pub fn annotate_reply(reply: String, skills: &[String]) -> String { - if skills.is_empty() { +pub fn annotate_reply(reply: String, fired: &[String]) -> String { + if fired.is_empty() { return reply; } let mut seen: Vec = Vec::new(); - for s in skills { + for s in fired { if !seen.contains(s) { seen.push(s.clone()); } @@ -166,7 +166,7 @@ pub fn annotate_reply(reply: String, skills: &[String]) -> String { } /// Drive a built rig prompt request to completion under the shared -/// [`REQUEST_TIMEOUT`], then annotate the reply with the Skills that fired. +/// [`REQUEST_TIMEOUT`], then annotate the reply with the tools that fired. /// /// The shared `complete()` tail, called from [`run_agent_request`] — the one /// call site all three backends (RigAgentClient, ChatGptAgentClient, @@ -174,7 +174,7 @@ pub fn annotate_reply(reply: String, skills: &[String]) -> String { /// typestate), so this is generic over `IntoFuture>` /// with a `Display` error. A timeout maps to [`AgentError::Timeout`]; a backend /// error maps to [`AgentError::Request`]; success is annotated with the fired -/// Skills. ONE definition of the tail. +/// tools. ONE definition of the tail. /// /// `backend` tags the lifecycle trace events (request-start / completion / /// error / timeout) so a hung or failing chat is traceable to a specific @@ -214,20 +214,20 @@ where } Ok(Ok(reply)) => reply, }; - let skills = fired.lock().map(|g| g.clone()).unwrap_or_default(); + let fired_tools = fired.lock().map(|g| g.clone()).unwrap_or_default(); info!( backend, elapsed_ms = started.elapsed().as_millis() as u64, reply_len = reply.len(), - skills_fired = skills.len(), + tools_fired = fired_tools.len(), "chat request complete" ); - Ok(annotate_reply(reply, &skills)) + Ok(annotate_reply(reply, &fired_tools)) } /// Drive one `complete()` call from a preamble'd `AgentBuilder` through to a /// reply string: apply the sampling knobs, register the SAME three tool sets -/// (telemetry/skill, ROCm read, ROCm mutating) in the SAME order, build the +/// (telemetry, ROCm read, ROCm mutating) in the SAME order, build the /// prompt request with the shared history/turn-limit, and hand off to /// [`finish_agent_request`]. /// @@ -715,7 +715,7 @@ where // Tool registration is delegated to run_agent_request, the single // shared site all three backends (RigAgentClient, ChatGptAgentClient, - // AnthropicAgentClient) now call: the SAME telemetry/skill tools + + // AnthropicAgentClient) now call: the SAME telemetry tools + // every ROCm read + mutating tool, so capability and approval // behavior are uniform across backends. let agent = self.client.agent(&self.model).preamble(&self.preamble); @@ -745,7 +745,7 @@ struct MockApprovalTrigger { } /// Deterministic in-memory client for tests and the offline demo. Never touches -/// the network. Can emit a canned tool-calling-style answer (cites a Skill). +/// the network. Can emit a canned tool-calling-style answer (cites a tool). pub struct MockAgentClient { reply: String, fail: bool, @@ -956,16 +956,16 @@ mod tests { } #[test] - fn annotate_reply_appends_deduped_skills() { + fn annotate_reply_appends_deduped_tools() { let r = annotate_reply("hi".into(), &["gpu_status".into(), "gpu_status".into()]); assert!(r.contains("hi")); assert!(r.contains("via: gpu_status")); - // No skills → unchanged. + // No fired tools → unchanged. assert_eq!(annotate_reply("hi".into(), &[]), "hi"); } #[tokio::test] - async fn mock_tool_calling_answer_cites_skill() { + async fn mock_tool_calling_answer_cites_tool() { // The offline "what's GPU-2 doing?" demo path — no live LLM. let agent = MockAgentClient::with_tool_call("GPU-2 is at 87% util, 71°C, 250 W.", "gpu_status"); @@ -977,7 +977,7 @@ mod tests { assert!(reply.contains("87% util")); assert!( reply.contains("gpu_status"), - "reply cites the Skill that fired" + "reply cites the tool that fired" ); } @@ -1693,7 +1693,7 @@ mod tests { #[test] fn all_backends_register_same_rocm_tools() { // Contract: every backend's complete() registers the SAME three tool - // sets — telemetry/skill (register_telemetry_tools), read ROCm + // sets — telemetry (register_telemetry_tools), read ROCm // (register_rocm_read_tools), and mutating ROCm // (register_rocm_mutating_tools) — so capability + approval parity holds // across local/openai/anthropic. This enumerates the canonical sets that @@ -1704,7 +1704,7 @@ mod tests { // asserts those tools are actually sent on the wire from that one site. // // Telemetry set: register_telemetry_tools registers exactly the - // TELEMETRY_TOOL_NAMES tools (GpuStatus, ListInstances, BenchSummary, + // tools in TELEMETRY_TOOL_NAMES (GpuStatus, ListInstances, BenchSummary, // TokensPerWatt). Pinning the size here means the telemetry // registration can't silently diverge from the canonical set. assert_eq!( @@ -1713,7 +1713,7 @@ mod tests { "canonical telemetry set size" ); for n in TELEMETRY_TOOL_NAMES { - assert!(!n.is_empty(), "empty telemetry/skill tool name"); + assert!(!n.is_empty(), "empty telemetry tool name"); // Telemetry tools are disjoint from both ROCm sets. assert!( !ROCM_READ_TOOL_NAMES.contains(&n), @@ -1888,7 +1888,7 @@ mod tests { assert!( reply.contains("⚙ via: gpu_status"), - "reply should carry the fired-skill annotation: {reply}" + "reply should carry the fired-tool annotation: {reply}" ); let requests = recorded.requests(); @@ -1925,7 +1925,7 @@ mod tests { )); // The snapshot `run_agent_request` hands to `register_telemetry_tools` - // reaches `gpu_status`'s own output, not just the fired-skill log: the + // reaches `gpu_status`'s own output, not just the fired-tool log: the // second request (after the tool call) must carry the tool result // derived from `fixture_snapshot()`'s GPUs, not an empty/default one. let second = requests diff --git a/crates/rocm-dash-tui/src/agent/tools.rs b/crates/rocm-dash-tui/src/agent/tools.rs index 5e38310cf0..424c3cf8f3 100644 --- a/crates/rocm-dash-tui/src/agent/tools.rs +++ b/crates/rocm-dash-tui/src/agent/tools.rs @@ -2,7 +2,7 @@ // // SPDX-License-Identifier: MIT -//! Rig `Tool` ("Skill") wrappers over the cached telemetry snapshot, plus the +//! Rig `Tool` wrappers over the cached telemetry snapshot, plus the //! read-only and mutating ROCm machine-inspection tools and their registration //! onto a Rig `AgentBuilder`. //! @@ -30,12 +30,12 @@ use super::snapshot::{ }; // --------------------------------------------------------------------------- -// Rig Tool ("Skill") wrappers. Each holds an `Arc` (read-only) -// and a shared `fired` log so the reply can cite which Skills ran. `call` only -// reads the snapshot — no mutation, no network, no file I/O. +// Rig Tool wrappers for the telemetry tools. Each holds an `Arc` +// (read-only) and a shared `fired` log so the reply can cite which tools ran. +// `call` only reads the snapshot — no mutation, no network, no file I/O. // --------------------------------------------------------------------------- -/// Shared "which Skills fired" log, threaded into every tool struct below and +/// Shared "which tools fired" log, threaded into every tool struct below and /// into the three real `AgentClient` backends via `agent::clients`. pub(super) type FiredLog = Arc>>; @@ -786,9 +786,9 @@ mod tests { use crate::agent::fixture_snapshot; #[test] - fn skill_names_are_unique_and_non_empty() { + fn telemetry_tool_names_are_unique_and_non_empty() { for n in TELEMETRY_TOOL_NAMES { - assert!(!n.is_empty(), "skill name must be non-empty"); + assert!(!n.is_empty(), "telemetry tool name must be non-empty"); } let mut sorted = TELEMETRY_TOOL_NAMES.to_vec(); sorted.sort_unstable(); @@ -796,7 +796,7 @@ mod tests { assert_eq!( sorted.len(), TELEMETRY_TOOL_NAMES.len(), - "skill names must be unique" + "telemetry tool names must be unique" ); } @@ -1013,11 +1013,11 @@ mod tests { "missing mutating tool: {expected}" ); } - // Disjoint from read-only + skill names (iterates the registry itself). + // Disjoint from read-only + telemetry tool names (iterates the registry itself). for n in ROCM_MUTATING_TOOL_NAMES { assert!( !TELEMETRY_TOOL_NAMES.contains(&n), - "collision with skill: {n}" + "collision with telemetry tool: {n}" ); // install_sdk_dry_run (read-only) must not clash with install_sdk. assert!( @@ -1207,7 +1207,7 @@ mod tests { } #[test] - fn rocm_read_tool_names_are_complete_and_disjoint_from_skills() { + fn rocm_read_tool_names_are_complete_and_disjoint_from_telemetry_tools() { // Every expected read-only tool is registered… for expected in [ "doctor", @@ -1234,7 +1234,7 @@ mod tests { } // natural_language_plan is read-only (plans, never executes) — Phase 7. assert!(ROCM_READ_TOOL_NAMES.contains(&"natural_language_plan")); - // The telemetry/skill tools are still present and disjoint from the new set. + // The telemetry tools are still present and disjoint from the new set. assert!(TELEMETRY_TOOL_NAMES.contains(&"gpu_status")); for n in ROCM_READ_TOOL_NAMES { assert!(!TELEMETRY_TOOL_NAMES.contains(&n), "name collision: {n}"); diff --git a/crates/rocm-dash-tui/src/ui/tabs/chat.rs b/crates/rocm-dash-tui/src/ui/tabs/chat.rs index 0e641eb2e9..daefd2533e 100644 --- a/crates/rocm-dash-tui/src/ui/tabs/chat.rs +++ b/crates/rocm-dash-tui/src/ui/tabs/chat.rs @@ -328,7 +328,7 @@ fn transcript_lines_from_turns<'a>(turns: &'a [ChatTurn], theme: &Theme) -> Vec< ChatRole::Error => ("err ", theme.err), ChatRole::System => ("·· ", theme.muted), }; - // Multi-line content (e.g. an answer plus a "⚙ via: …" Skill annotation) + // Multi-line content (e.g. an answer plus a "⚙ via: …" fired-tools annotation) // renders one terminal line per segment; continuation lines are indented // to align under the first line's content. for (i, seg) in turn.content.split('\n').enumerate() { @@ -445,7 +445,7 @@ mod tests { } #[test] - fn transcript_renders_skill_annotation_on_its_own_line() { + fn transcript_renders_tool_annotation_on_its_own_line() { let mut s = AppState::new("t".into(), "default-dark".into()); // An agent reply carrying a tool-call surfacing annotation. s.chat.push(ChatTurn::agent( @@ -457,7 +457,7 @@ mod tests { assert_eq!(lines.len(), 2, "answer + annotation render as two lines"); assert!( format!("{lines:?}").contains("gpu_status"), - "the fired Skill is surfaced in the transcript" + "the fired tool is surfaced in the transcript" ); } diff --git a/docs/architecture.md b/docs/architecture.md index edba1fc0ed..3713bae28e 100644 --- a/docs/architecture.md +++ b/docs/architecture.md @@ -209,7 +209,7 @@ The old agent.rs was split (mechanical relocation — the shared seam types stay - [`agent/mod.rs`](../crates/rocm-dash-tui/src/agent/mod.rs) — the `AgentClient` seam, `AgentError`, `StateSnapshot`, `InferenceParams`, `REQUEST_TIMEOUT`. - [`agent/snapshot.rs`](../crates/rocm-dash-tui/src/agent/snapshot.rs) — pure JSON telemetry helpers with no `rig` dependency. -- [`agent/tools.rs`](../crates/rocm-dash-tui/src/agent/tools.rs) — the `rig::tool::Tool` "Skill" wrappers and ROCm read/mutating tool dispatch. +- [`agent/tools.rs`](../crates/rocm-dash-tui/src/agent/tools.rs) — the `rig::tool::Tool` wrappers for the four telemetry tools (`gpu_status`, `list_instances`, `bench_summary`, `tokens_per_watt`), their name table `TELEMETRY_TOOL_NAMES`, and the ROCm read/mutating tool dispatch. - [`agent/clients.rs`](../crates/rocm-dash-tui/src/agent/clients.rs) — the `RigAgentClient`/`ChatGptAgentClient`/`AnthropicAgentClient`/`MockAgentClient` backends. [`app/mod.rs`](../crates/rocm-dash-tui/src/app/mod.rs) is modularized (ROCMAI-84, Phase 4 of ROCMAI-27's sequencing) into an `app/` directory. [`app/types.rs`](../crates/rocm-dash-tui/src/app/types.rs), [`app/event_loop.rs`](../crates/rocm-dash-tui/src/app/event_loop.rs), [`app/scrollbar.rs`](../crates/rocm-dash-tui/src/app/scrollbar.rs), and [`app/actions.rs`](../crates/rocm-dash-tui/src/app/actions.rs) follow full domain extraction. [`app/chat.rs`](../crates/rocm-dash-tui/src/app/chat.rs), [`app/slash.rs`](../crates/rocm-dash-tui/src/app/slash.rs), and [`app/summary.rs`](../crates/rocm-dash-tui/src/app/summary.rs) were extracted earlier via mechanical relocation, the same convention the `agent/` split above follows. `quit_confirm`, `tool_approval`, `overlays`, `scroll` and `chat_state` each hold a further `impl AppState` block for one cohesive group of reducer methods: From 5f110536b8443147e96201505d52b88d30612439 Mon Sep 17 00:00:00 2001 From: Jussi Elo Date: Thu, 5 Nov 2026 13:11:50 +0000 Subject: [PATCH 3/3] ROCMAI-864: keep the skills_fired tracing field name The "chat request complete" tracing field is a log key that external filters may match on, so the rename is dropped to keep this PR free of observable changes. Nothing else changes. Signed-off-by: Jussi Elo --- crates/rocm-dash-tui/src/agent/clients.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/crates/rocm-dash-tui/src/agent/clients.rs b/crates/rocm-dash-tui/src/agent/clients.rs index 0237c706e9..ae42afd3e1 100644 --- a/crates/rocm-dash-tui/src/agent/clients.rs +++ b/crates/rocm-dash-tui/src/agent/clients.rs @@ -219,7 +219,7 @@ where backend, elapsed_ms = started.elapsed().as_millis() as u64, reply_len = reply.len(), - tools_fired = fired_tools.len(), + skills_fired = fired_tools.len(), "chat request complete" ); Ok(annotate_reply(reply, &fired_tools))