feat: migrate Codex from deprecated custom prompts to Skills format

Codex CLI 0.117+ deprecated ~/.codex/prompts/ custom prompts.
Migrate to the new Skills system at ~/.codex/skills/squad/SKILL.md
with proper name/description frontmatter for $squad invocation.

Author: tt-a1i

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "squad"
-version = "0.7.5"
+version = "0.7.6"
 edition = "2021"
 license = "MIT"
 description = "Multi-AI-agent terminal collaboration via simple CLI commands"

Cargo.toml

@@ -27,8 +27,8 @@ pub const PLATFORMS: &[Platform] = &[
     Platform {
         name: "codex",
         binary: "codex",
-        command_path: ".codex/prompts/squad.md",
-        content: SQUAD_MD_CONTENT,
+        command_path: ".codex/skills/squad/SKILL.md",
+        content: SQUAD_CODEX_CONTENT,
     },
     Platform {
         name: "opencode",
@@ -38,7 +38,73 @@ pub const PLATFORMS: &[Platform] = &[
     },
 ];
 
-/// Markdown format for Claude Code, Codex, OpenCode (uses $ARGUMENTS)
+/// Codex Skills format (uses $ARGUMENTS, placed in ~/.agents/skills/squad/SKILL.md)
+pub const SQUAD_CODEX_CONTENT: &str = r#"---
+name: squad
+description: Join squad multi-agent collaboration. Usage: $squad <role> [custom-id]
+---
+
+You are joining a squad multi-agent collaboration team.
+
+Your join arguments: $ARGUMENTS
+
+**IMPORTANT:** Do NOT run `squad $ARGUMENTS` or treat the arguments as a CLI subcommand. Instead, follow the setup steps below.
+
+## Phase 1: Setup (do this once)
+
+1. Parse your join arguments above.
+
+   **If arguments are empty or missing:**
+   Run `squad roles` to list available roles, then ask the user which role they want to join as. Do NOT proceed until the user picks a role.
+
+   **If arguments look like a role name** (1-2 words, e.g. "cto", "worker worker-2"):
+   - First word is your role — this can be ANY string: "cto", "ceo", "manager", "reviewer", etc. It does NOT need to appear in `squad roles` (that list only shows predefined templates).
+   - Optional second word is a custom agent ID
+   - If no custom ID provided, use the role name as your ID
+   - Examples: "manager" → id=manager, role=manager | "worker worker-2" → id=worker-2, role=worker | "cto" → id=cto, role=cto
+
+   **If arguments look like natural language** (e.g. "加入团队，作为管理员", "join as tech lead and review PRs"):
+   - Extract the intended role from the text. Pick a short English role name (e.g. "manager", "reviewer", "cto").
+   - Use that as your role and ID.
+   - If no role can be inferred, ask the user to clarify.
+
+2. Run `squad init` (safe to run — won't overwrite existing workspace).
+
+3. **Clean up stale agents from previous sessions:**
+   Run `squad agents` and check the output.
+   - If ALL agents show "stale" (no active agents), tell the user stale squad state was detected and ask the user whether they want to reset squad state with `squad clean` followed by `squad init`. Do NOT clean automatically.
+   - If some agents are active (a team is already running), skip cleanup and proceed.
+
+4. Run `squad join <id> --role <role> --client __SQUAD_CLIENT__ --protocol-version __SQUAD_PROTOCOL_VERSION__` to register yourself.
+   - Read the output line that says "Joined as ..." — that confirms your actual agent ID.
+   - If the ID was taken, squad auto-assigns a suffixed ID (e.g. worker-2). Use that ID for all commands.
+   - If role instructions are printed (=== Role Instructions ===), follow them.
+   - If no predefined template exists, interpret the role using your own knowledge.
+
+5. Run `squad agents` to see who else is on the team.
+
+6. **If any squad command returns "Session replaced":** another terminal took your ID. Re-join with a different ID (e.g. `squad join worker-2 --role worker --client __SQUAD_CLIENT__ --protocol-version __SQUAD_PROTOCOL_VERSION__`).
+
+## Phase 2: Enter Receive Mode (MANDATORY)
+
+**Immediately after setup, run `squad receive <your-id> --wait` to start listening for messages.** Do NOT wait for the user to tell you — enter receive mode now.
+
+After receiving a message:
+1. Execute the task or respond as appropriate for your role.
+2. Report results using `squad send` or `squad task` commands.
+3. Run `squad receive <your-id> --wait` again to wait for the next message.
+
+If receive times out with no messages, run it again immediately.
+
+Other useful commands:
+- `squad send <your-id> <to> "<message>"` — send a message (use @all to broadcast)
+- `squad task create <your-id> <to> --title "<title>"` — create a structured task
+- `squad agents` — see who is online
+- `squad pending` — check all unread messages
+- `squad history` — view message history
+"#;
+
+/// Markdown format for Claude Code, OpenCode (uses $ARGUMENTS)
 pub const SQUAD_MD_CONTENT: &str = r#"---
 description: Join squad multi-agent collaboration. Usage: /squad <role> [custom-id]
 ---