Auto-number reviewers, verb-form roles, and rally reset

crates/rally-workflow-build/src/lib.rs

@@ -87,7 +87,7 @@ fn default_state_version() -> u32 {
 }
 
 fn default_implementer() -> String {
-    "implementer".to_string()
+    "implement".to_string()
 }
 
 impl BuildWorkflowState {
@@ -580,7 +580,7 @@ pub fn process_review_state(state: &mut SessionState) -> Result<Option<String>>
 pub fn is_implementer(state: &SessionState, agent: &str) -> bool {
     match read_workflow_state(state) {
         Ok(workflow_state) => is_implementer_with_state(&workflow_state, agent),
-        Err(_) => agent == "implementer",
+        Err(_) => agent == "implement",
     }
 }
 

prompts/skill-wrapper.md

@@ -6,10 +6,71 @@ rally-managed: true
 
 # Rally
 
-Run this command first with the user's arguments:
+Rally coordinates multi-agent coding sessions. Your job is to figure out what the user wants, construct the right `{{RALLY_BIN}} skill run` command, execute it, and follow the instruction loop.
+
+## Step 1: Determine the command
+
+The user said: `$ARGUMENTS`
+
+### If the arguments clearly match a known command syntax
+
+Run it directly. Examples of clear, structured arguments:
+
+| User says | Run |
+|-----------|-----|
+| `implement todos/foo.md implement` | `{{RALLY_BIN}} skill run implement todos/foo.md implement` |
+| `implement todos/foo.md review --reviewers 2` | `{{RALLY_BIN}} skill run implement todos/foo.md review --reviewers 2` |
+| `negotiate "auth design" agent-alpha --agents 3` | `{{RALLY_BIN}} skill run negotiate "auth design" agent-alpha --agents 3` |
+| (empty / no arguments) | `{{RALLY_BIN}} skill run` (resumes last session) |
+
+### If the arguments are natural language or ambiguous
+
+Interpret the user's intent to construct a `{{RALLY_BIN}} skill run` command. Use these rules:
+
+**Available workflows:**
+
+- `{{RALLY_BIN}} skill run implement <todo-file> <role> [--reviewers N]` — implement a todo with code review
+- `{{RALLY_BIN}} skill run negotiate <topic> <role> [--agents N]` — multi-agent planning/negotiation
+
+**Roles for implement sessions:**
+
+- `implement` — writes the code (use when the user says "implement", "build", "code", etc.)
+- `review` — reviews checkpoints (use when the user says "review", "check", etc.)
+
+**Rally auto-assigns reviewer names.** Always pass `review` as the role — rally will automatically assign unique names (review-1, review-2, etc.). Never pick your own reviewer number.
+
+**Mapping examples:**
+
+| User says | Interpretation |
+|-----------|---------------|
+| `impl todos/foo.md 2 reviewers` | `{{RALLY_BIN}} skill run implement todos/foo.md implement --reviewers 2` |
+| `impl todos/foo.md you implement with 2 reviewers` | `{{RALLY_BIN}} skill run implement todos/foo.md implement --reviewers 2` |
+| `review todos/foo.md` | `{{RALLY_BIN}} skill run implement todos/foo.md review` |
+| `implement the microvm todo as reviewer` | Find the todo file, run as review |
+| `impl todos/foo.md` | If role is ambiguous, ask the user |
+
+### If you need more workspace context
+
+Run this to discover todo files, active sessions, and saved context:
 
 ```bash
-{{RALLY_BIN}} skill agent-howto $ARGUMENTS
+{{RALLY_BIN}} skill context
 ```
 
-Then follow the printed instructions exactly.
+## Step 2: Run the command
+
+Run the `{{RALLY_BIN}} skill run ...` command you determined above.
+
+## Step 3: Follow the instruction loop
+
+1. **READ** the instruction printed by the command.
+2. **DO** exactly what it says — write files, run tests, git commit, etc.
+3. **RUN** the follow-up commands listed at the end of your instruction (checkpoint, review, etc). Run them ALL, in order.
+4. **RUN** `{{RALLY_BIN}} skill run` again with no arguments to get your next instruction.
+5. **REPEAT** from step 1. Do NOT stop until the output says "session complete; stop working".
+
+**IMPORTANT: You are ONE agent in a multi-agent session. Other agents are running in separate terminals.**
+
+- If the command prints "Waiting for ..." — just wait and poll. Do NOT try to spawn or recruit other agents.
+- If the command exits with code 2 or prints "session complete" — you are done. Stop.
+- Rally tells you your assigned name (e.g. "registered 'review-2'"). Use that name for all subsequent commands.

src/cli.rs

@@ -62,6 +62,8 @@ pub enum Command {
     Status(StatusArgs),
     #[command(about = "List all sessions")]
     Sessions,
+    #[command(about = "Interactively select and delete a session")]
+    Reset,
     #[command(about = "File a divergence issue (plan sessions)")]
     FileIssue(FileIssueArgs),
     #[command(about = "Challenge another agent's position (plan sessions)")]
@@ -114,7 +116,7 @@ pub struct CreateArgs {
     pub workflow: Option<String>,
     #[arg(
         long,
-        default_value = "implementer",
+        default_value = "implement",
         help = "Implementer agent name for implement sessions"
     )]
     pub implementer: String,
@@ -346,6 +348,11 @@ pub enum SkillSubcommand {
         long_about = "Print the latest Rally agent instructions for `/rally` wrappers.\n\nPass the same wrapper args so this command can print the exact follow-up `rally skill run ...` command to execute."
     )]
     AgentHowto(SkillAgentHowtoArgs),
+    #[command(
+        about = "Print workspace context for intelligent command resolution",
+        long_about = "Print workspace context for intelligent command resolution.\n\nOutputs available workflows, todo files in the workspace, active sessions, and saved session context. Designed to be read by a coding agent to determine the right `rally skill run` command."
+    )]
+    Context(SkillContextArgs),
     #[command(about = "Validate wrapper installation and target paths")]
     Doctor(SkillDoctorArgs),
     #[command(about = "Uninstall Rally-managed wrapper artifacts")]
@@ -399,6 +406,15 @@ pub struct SkillAgentHowtoArgs {
     pub args: Vec<String>,
 }
 
+#[derive(Args, Debug)]
+pub struct SkillContextArgs {
+    #[arg(
+        long,
+        help = "Workspace directory to scan for todo files (defaults to current directory)"
+    )]
+    pub workspace: Option<PathBuf>,
+}
+
 #[derive(Args, Debug)]
 pub struct SkillDoctorArgs {
     #[arg(