feat(acp): Discover session transcripts for parsing

.claude/CLAUDE.md

@@ -80,12 +80,12 @@ Found 23 skills:
 /home/clifford/Documents/source/nori/cli/.claude/skills/using-skills/SKILL.md
   Name: Getting Started with Abilities
   Description: Describes how to use abilities. Read before any conversation.
-/home/clifford/Documents/source/nori/cli/.claude/skills/using-screenshots/SKILL.md
-  Name: Taking and Analyzing Screenshots
-  Description: Use this to capture screen context.
 /home/clifford/Documents/source/nori/cli/.claude/skills/using-git-worktrees/SKILL.md
   Name: Using Git Worktrees
   Description: Use this whenever you need to create an isolated workspace.
+/home/clifford/Documents/source/nori/cli/.claude/skills/using-screenshots/SKILL.md
+  Name: Taking and Analyzing Screenshots
+  Description: Use this to capture screen context.
 /home/clifford/Documents/source/nori/cli/.claude/skills/updating-noridocs/SKILL.md
   Name: Updating Noridocs
   Description: Use this when you have finished making code changes and you are ready to update the documentation based on those changes.
@@ -107,39 +107,39 @@ Found 23 skills:
 /home/clifford/Documents/source/nori/cli/.claude/skills/write-noridoc/SKILL.md
   Name: Write Noridoc
   Description: Write or update documentation in the server-side noridocs system.
-/home/clifford/Documents/source/nori/cli/.claude/skills/sync-noridocs/SKILL.md
-  Name: Sync Noridocs
-  Description: Sync all local docs.md files to server-side noridocs system.
 /home/clifford/Documents/source/nori/cli/.claude/skills/recall/SKILL.md
   Name: Recall
   Description: Search the Nori knowledge base for relevant context, solutions, and documentation.
-/home/clifford/Documents/source/nori/cli/.claude/skills/read-noridoc/SKILL.md
-  Name: Read Noridoc
-  Description: Read documentation from the server-side noridocs system by file path.
-/home/clifford/Documents/source/nori/cli/.claude/skills/prompt-analysis/SKILL.md
-  Name: Prompt Analysis
-  Description: Analyze prompts for quality and best practices before sending them to Claude.
+/home/clifford/Documents/source/nori/cli/.claude/skills/sync-noridocs/SKILL.md
+  Name: Sync Noridocs
+  Description: Sync all local docs.md files to server-side noridocs system.
 /home/clifford/Documents/source/nori/cli/.claude/skills/memorize/SKILL.md
   Name: Memorize
   Description: Use this to save important implementation decisions, patterns, or context to the Nori knowledge base for future sessions.
-/home/clifford/Documents/source/nori/cli/.claude/skills/list-noridocs/SKILL.md
-  Name: List Noridocs
-  Description: List all server-side noridocs, optionally filtered by repository and/or path prefix.
+/home/clifford/Documents/source/nori/cli/.claude/skills/read-noridoc/SKILL.md
+  Name: Read Noridoc
+  Description: Read documentation from the server-side noridocs system by file path.
 /home/clifford/Documents/source/nori/cli/.claude/skills/handle-large-tasks/SKILL.md
   Name: Handle-Large-Tasks
   Description: Use this skill to split large plans into smaller chunks. This skill manages your context window for large tasks. Use it when a task will take a long time and cause context issues.
 /home/clifford/Documents/source/nori/cli/.claude/skills/finishing-a-development-branch/SKILL.md
   Name: Finishing a Development Branch
   Description: Use this when you have completed some feature implementation and have written passing tests, and you are ready to create a PR.
+/home/clifford/Documents/source/nori/cli/.claude/skills/prompt-analysis/SKILL.md
+  Name: Prompt Analysis
+  Description: Analyze prompts for quality and best practices before sending them to Claude.
 /home/clifford/Documents/source/nori/cli/.claude/skills/creating-skills/SKILL.md
   Name: Creating-Skills
   Description: Use when you need to create a new custom skill for a profile - guides through gathering requirements, creating directory structure, writing SKILL.md, and optionally adding bundled scripts
-/home/clifford/Documents/source/nori/cli/.claude/skills/building-ui-ux/SKILL.md
-  Name: Building UI/UX
-  Description: Use when implementing user interfaces or user experiences - guides through exploration of design variations, frontend setup, iteration, and proper integration
 /home/clifford/Documents/source/nori/cli/.claude/skills/brainstorming/SKILL.md
   Name: Brainstorming
   Description: IMMEDIATELY USE THIS SKILL when creating or develop anything and before writing code or implementation plans - refines rough ideas into fully-formed designs through structured Socratic questioning, alternative exploration, and incremental validation
+/home/clifford/Documents/source/nori/cli/.claude/skills/building-ui-ux/SKILL.md
+  Name: Building UI/UX
+  Description: Use when implementing user interfaces or user experiences - guides through exploration of design variations, frontend setup, iteration, and proper integration
+/home/clifford/Documents/source/nori/cli/.claude/skills/list-noridocs/SKILL.md
+  Name: List Noridocs
+  Description: List all server-side noridocs, optionally filtered by repository and/or path prefix.
 
 Check if any of these skills are relevant to the user's task. If relevant, use the Read tool to load the skill before proceeding.
 

IMPLEMENTATION_PLAN.md

@@ -0,0 +1,233 @@
+# Agent Session ID Discovery and Token Usage Implementation Plan
+
+**Goal:** Enable the `/status` command to display token usage by discovering the current ACP session ID, locating the corresponding session transcript, and parsing token usage data.
+
+**Architecture:** The ACP protocol provides session IDs at session creation. We will add a session discovery module to locate transcript files based on agent type and session ID, then integrate parsed token usage into the Nori session header displayed by `/status`.
+
+**Tech Stack:** Rust, codex-acp, codex-tui, session_parser module, tokio async
+
+---
+
+## Testing Plan
+
+I will add unit tests that ensure:
+1. Session transcript path building works correctly for each agent type (Claude, Codex, Gemini)
+2. The session discovery module can find transcripts given session ID and agent kind
+3. Token usage integration in session_header displays correctly
+
+I will add integration tests that ensure:
+1. End-to-end flow: session ID from ACP → transcript discovery → token usage parsing → display
+2. Edge cases: missing transcripts, malformed session IDs, no token data available
+
+NOTE: I will write *all* tests before I add any implementation behavior.
+
+---
+
+## Phase 1: Session Discovery Module
+
+### Step 1.1: Create session_discovery.rs module skeleton
+- **File**: `/home/user/nori-cli/codex-rs/acp/src/session_discovery.rs`
+- **Actions**:
+  1. Create the file with module documentation
+  2. Define `discover_transcript_path(agent_kind: AgentKind, session_id: &str, cwd: &Path) -> Option<PathBuf>`
+  3. Export from `/home/user/nori-cli/codex-rs/acp/src/lib.rs`
+
+### Step 1.2: Write failing tests for Claude transcript discovery
+- **File**: `/home/user/nori-cli/codex-rs/acp/tests/session_discovery_test.rs`
+- **Test cases**:
+  1. `test_claude_transcript_path_format` - verify path format `~/.claude/projects/<PROJECT_PATH>/<SESSIONID>.jsonl`
+  2. `test_claude_transcript_discovery_with_valid_session` - given session ID, find transcript
+  3. `test_claude_transcript_discovery_missing_file` - return None when file not found
+
+### Step 1.3: Write failing tests for Codex transcript discovery
+- **Test cases**:
+  1. `test_codex_transcript_path_discovery` - search by session GUID in filename
+  2. `test_codex_transcript_discovery_by_date` - find in date-organized directory
+
+### Step 1.4: Write failing tests for Gemini transcript discovery
+- **Test cases**:
+  1. `test_gemini_transcript_path_format` - verify path format with hashed paths
+  2. `test_gemini_transcript_discovery_with_session_id`
+
+### Step 1.5: Run tests, verify they fail
+```bash
+cargo test -p codex-acp session_discovery
+```
+
+### Step 1.6: Implement Claude transcript discovery
+- **File**: `/home/user/nori-cli/codex-rs/acp/src/session_discovery.rs`
+- **Logic**:
+  1. Expand `~/.claude/projects/` base path
+  2. Build relative project path from `cwd` (hash or relativize)
+  3. Check for `<SESSION_ID>.jsonl` file existence
+  4. Return path if found
+
+### Step 1.7: Implement Codex transcript discovery
+- **Logic**:
+  1. Expand `~/.codex/sessions/` base path
+  2. Search recursively for files matching `*-<SESSION_GUID>.jsonl`
+  3. Return first match or None
+
+### Step 1.8: Implement Gemini transcript discovery
+- **Logic**:
+  1. Expand `~/.gemini/tmp/` base path
+  2. Hash the cwd to get the hashed path component
+  3. Search for `session-*-<SESSION_ID>.json`
+  4. Return path if found
+
+### Step 1.9: Run tests, verify they pass
+```bash
+cargo test -p codex-acp session_discovery
+```
+
+### Step 1.10: Commit Phase 1
+```bash
+git add -A && git commit -m "feat(acp): Add session transcript discovery module"
+```
+
+---
+
+## Phase 2: Expose Session ID from ACP Backend
+
+### Step 2.1: Write failing test for session_id accessor
+- **File**: `/home/user/nori-cli/codex-rs/acp/tests/backend_test.rs` (or add to existing)
+- **Test**: Verify `AcpBackend::session_id()` returns the session ID string
+
+### Step 2.2: Run test, verify it fails
+
+### Step 2.3: Add session_id() method to AcpBackend
+- **File**: `/home/user/nori-cli/codex-rs/acp/src/backend.rs`
+- **Current**: `AcpBackend` already has `session_id: Arc<acp::SessionId>` (backend.rs:102-111)
+- **Add**: Public accessor `pub fn session_id(&self) -> &str`
+
+### Step 2.4: Run test, verify it passes
+
+### Step 2.5: Add agent_kind() method to AcpBackend
+- **File**: `/home/user/nori-cli/codex-rs/acp/src/backend.rs`
+- **Add**: Store and expose `AgentKind` from the `AcpAgentConfig`
+
+### Step 2.6: Commit Phase 2
+```bash
+git add -A && git commit -m "feat(acp): Expose session_id and agent_kind from AcpBackend"
+```
+
+---
+
+## Phase 3: Extend TUI to Track ACP Session Info
+
+### Step 3.1: Write failing test for session info in ChatWidget
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/chatwidget/tests.rs`
+- **Test**: Verify ChatWidget can access session ID when in ACP mode
+
+### Step 3.2: Run test, verify it fails
+
+### Step 3.3: Add session info channel/handle to ACP agent spawning
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/chatwidget/agent.rs`
+- **Modify**: `SpawnAgentResult` to include optional `AcpSessionInfo { session_id: String, agent_kind: AgentKind }`
+- **Modify**: `spawn_acp_agent` to extract and return session info
+
+### Step 3.4: Store session info in ChatWidget
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/chatwidget.rs`
+- **Add**: Optional field `acp_session_info: Option<AcpSessionInfo>`
+- **Update**: On agent spawn, store the session info
+
+### Step 3.5: Run test, verify it passes
+
+### Step 3.6: Commit Phase 3
+```bash
+git add -A && git commit -m "feat(tui): Track ACP session info for token usage display"
+```
+
+---
+
+## Phase 4: Integrate Token Usage into /status Command
+
+### Step 4.1: Write failing snapshot test for status with token usage
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/nori/session_header.rs`
+- **Test**: Verify token usage section appears in `/status` output when available
+
+### Step 4.2: Run test, verify it fails
+
+### Step 4.3: Modify new_nori_status_output signature
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/nori/session_header.rs`
+- **Change**: Add optional `TokenUsageReport` parameter
+- **Display**: Add "Token Usage" section showing:
+  - Total tokens (input + output)
+  - Input tokens (with cached breakdown if available)
+  - Output tokens (with reasoning breakdown if available)
+  - Context window usage percentage if available
+
+### Step 4.4: Update add_status_output in ChatWidget
+- **File**: `/home/user/nori-cli/codex-rs/tui/src/chatwidget.rs`
+- **Modify**: `add_status_output()` to:
+  1. Check if ACP session info is available
+  2. If yes, call session discovery to find transcript path
+  3. Parse token usage using `parse_session_transcript()`
+  4. Pass token usage to `new_nori_status_output()`
+
+### Step 4.5: Run tests, verify they pass
+```bash
+cargo test -p codex-tui nori_session_header
+```
+
+### Step 4.6: Update snapshot tests
+```bash
+cargo test -p codex-tui -- --update-snapshots
+```
+
+### Step 4.7: Commit Phase 4
+```bash
+git add -A && git commit -m "feat(tui): Display token usage in /status command for ACP sessions"
+```
+
+---
+
+## Phase 5: Handle Edge Cases
+
+### Step 5.1: Add test for missing transcript file
+- Verify graceful handling when transcript file doesn't exist yet (new session)
+
+### Step 5.2: Add test for malformed/empty transcript
+- Verify graceful handling when transcript is empty or malformed
+
+### Step 5.3: Add test for HTTP mode (non-ACP)
+- Verify `/status` still works when not using ACP
+
+### Step 5.4: Implement error handling
+- Return None for token usage if discovery or parsing fails
+- Log warnings for debugging
+- Display "Token usage unavailable" or similar when not available
+
+### Step 5.5: Commit Phase 5
+```bash
+git add -A && git commit -m "fix(tui): Handle edge cases for token usage display"
+```
+
+---
+
+## Testing Details
+
+Tests will verify:
+1. **Session discovery behavior**: Given agent type + session ID + cwd, correct transcript path is found
+2. **Token usage display**: Snapshot tests verify the formatting of token usage in /status output
+3. **Edge case handling**: Graceful degradation when transcripts are missing/malformed
+
+## Implementation Details
+
+- Session ID comes from `AcpBackend::session_id()` which wraps `acp::SessionId` from the protocol
+- Transcript paths differ per agent:
+  - Claude: `~/.claude/projects/<PROJECT_PATH>/<SESSIONID>.jsonl`
+  - Codex: `~/.codex/sessions/<YEAR>/<MM>/<DD>/rollout-*-<SESSION_GUID>.jsonl`
+  - Gemini: `~/.gemini/tmp/<HASHED_PATHS>/chats/session-*-<SESSIONID>.json`
+- `parse_session_transcript()` from `session_parser.rs` already handles token aggregation
+- Token usage display reuses existing formatting from `status/card.rs` patterns
+
+## Questions
+
+1. **Claude project path hashing**: Claude stores projects in hashed paths. Need to verify the exact hashing algorithm used (likely SHA-256 or similar). Should check Claude Code source or existing transcript fixtures.
+
+2. **Async vs sync transcript reading**: Should transcript discovery and parsing be async? Currently `parse_session_transcript` uses sync I/O. For responsiveness, might want async with timeout.
+
+3. **Caching**: Should we cache the parsed token usage to avoid re-parsing on every `/status` call? The transcript file grows during the session, so fresh reads may be needed.
+
+---

codex-rs/acp/src/backend.rs

@@ -34,6 +34,7 @@ use crate::connection::AcpConnection;
 use crate::connection::AcpModelState;
 use crate::connection::ApprovalEventType;
 use crate::connection::ApprovalRequest;
+use crate::registry::AgentKind;
 use crate::registry::get_agent_config;
 use crate::translator;
 use crate::translator::is_patch_operation;
@@ -63,6 +64,7 @@ pub struct AcpBackendConfig {
 pub struct AcpBackend {
     connection: Arc<AcpConnection>,
     session_id: acp::SessionId,
+    agent_kind: AgentKind,
     event_tx: mpsc::Sender<Event>,
     #[allow(dead_code)]
     cwd: PathBuf,
@@ -109,6 +111,7 @@ impl AcpBackend {
         let backend = Self {
             connection,
             session_id,
+            agent_kind: agent_config.agent,
             event_tx: event_tx.clone(),
             cwd: cwd.clone(),
             pending_approvals: Arc::clone(&pending_approvals),
@@ -404,6 +407,11 @@ impl AcpBackend {
         &self.session_id
     }
 
+    /// Get the agent kind (Claude, Codex, or Gemini).
+    pub fn agent_kind(&self) -> AgentKind {
+        self.agent_kind
+    }
+
     /// Get a reference to the underlying ACP connection.
     ///
     /// This provides access to low-level ACP operations like model switching.

codex-rs/acp/src/lib.rs

@@ -10,6 +10,7 @@ pub mod backend;
 pub mod config;
 pub mod connection;
 pub mod registry;
+pub mod session_discovery;
 pub mod session_parser;
 pub mod tracing_setup;
 pub mod translator;
@@ -46,6 +47,12 @@ pub use tracing_setup::init_rolling_file_tracing;
 pub use translator::TranslatedEvent;
 pub use translator::translate_session_update;
 
+// Session discovery exports
+pub use session_discovery::DiscoveryError;
+pub use session_discovery::cwd_to_claude_project_path;
+pub use session_discovery::discover_transcript_path;
+pub use session_discovery::discover_transcript_path_with_home;
+
 // Re-export commonly used types from agent-client-protocol
 pub use agent_client_protocol::Agent;
 pub use agent_client_protocol::Client;