ai-dispatch 7.2.2

## v1.5 — reliability fixes and dead code cleanup

[[task]]
name = "remove-topo-levels"
agent = "opencode"
skills = ["implementer"]
prompt = """
Remove the dead `topo_levels()` function from src/batch.rs.

It was replaced by dependency-based scheduling in v1.4 and now triggers a dead_code warning.

STEPS:
1. Delete the `topo_levels()` function (lines 69-104 approximately)
2. Delete any helper types it uses that are now also dead (VisitState is still used by validate_dag, keep it)
3. cargo check must pass with NO warnings about dead code in batch.rs

CONSTRAINTS:
- Only modify src/batch.rs
- Do NOT delete validate_dag(), dependency_indices(), or VisitState — those are still used
- Do NOT reformat existing code
- cargo check and cargo test must pass
"""
dir = "."
worktree = "fix/remove-topo-levels"
verify = "auto"

[[task]]
name = "fix-truncate-utf8"
agent = "opencode"
skills = ["implementer"]
prompt = """
Fix unsafe byte-level string slicing in src/cmd/show.rs truncate() function.

BUG: Line ~413 does `&s[..max.saturating_sub(3)]` which panics on multi-byte UTF-8 characters (e.g., Chinese, emoji, curly quotes).

FIX: Use `s.floor_char_boundary(max.saturating_sub(3))` for safe truncation, matching the pattern already used in src/agent/truncate.rs.

Change from:
```rust
fn truncate(s: &str, max: usize) -> String {
    if s.len() <= max {
        s.to_string()
    } else {
        format!("{}...", &s[..max.saturating_sub(3)])
    }
}
```

To:
```rust
fn truncate(s: &str, max: usize) -> String {
    if s.len() <= max {
        s.to_string()
    } else {
        let end = s.floor_char_boundary(max.saturating_sub(3));
        format!("{}...", &s[..end])
    }
}
```

CONSTRAINTS:
- Only modify src/cmd/show.rs
- Only change the truncate() function
- Do NOT reformat existing code
- cargo check and cargo test must pass
"""
dir = "."
worktree = "fix/truncate-utf8"
verify = "auto"

[[task]]
name = "max-task-duration"
agent = "opencode"
skills = ["implementer"]
prompt = """
Add max task duration auto-fail to zombie detection.

GOAL: Tasks running longer than a configurable maximum are automatically marked as failed during zombie checks.

EXISTING CODE:
- src/background.rs has check_zombie_tasks_with() at ~line 207
- It checks if worker process is alive and marks dead workers as zombies
- crate::paths::aid_dir() returns ~/.aid

IMPLEMENTATION:

1. Add a constant DEFAULT_MAX_TASK_DURATION_MINS at top of background.rs:
   const DEFAULT_MAX_TASK_DURATION_MINS: i64 = 60;

2. In check_zombie_tasks_with(), after the existing worker-alive check, add a duration check:
   For each running task, compute elapsed time:
   let elapsed_mins = (Local::now() - task.created_at).num_minutes();
   If elapsed_mins > DEFAULT_MAX_TASK_DURATION_MINS:
   - Kill the worker process if it exists (kill SIGTERM)
   - Record failure with detail "Task exceeded max duration ({elapsed_mins}m > {DEFAULT_MAX_TASK_DURATION_MINS}m)"
   - Add to cleaned list

3. The check should happen AFTER the zombie process check (so zombie processes are caught first, then long-running processes).

CONSTRAINTS:
- Only modify src/background.rs
- ~15 lines of new code
- Do NOT reformat existing code
- cargo check and cargo test must pass
"""
dir = "."
worktree = "fix/max-duration"
verify = "auto"