SDD skill: $(cmd) in redirect target triggers permission prompt

## Problem

The SDD skill's `SKILL.md` instructs Claude to use `$(git rev-parse --git-path sdd)` inline in commands, e.g.:

```
cat "$(git rev-parse --git-path sdd)/progress.md"
```

When Claude follows this pattern for the progress ledger *append* operation, it generates:

```bash
echo "Task N: complete ..." >> "$(git rev-parse --git-path sdd)/progress.md"
```

Claude Code flags this with **"Redirect target contains $(cmd) output — path is runtime-determined"** and prompts the user, because it can't statically verify the redirect destination. No permission glob pattern can suppress this — it's a separate safety check.

## Current behavior

Every ledger append triggers a permission prompt that can't be configured away.

## Expected behavior

The skill instructions should use a resolved path pattern so the redirect target is a plain variable, not a command substitution:

```bash
sdd_dir=$(git rev-parse --git-path sdd)
echo "Task N: complete ..." >> "$sdd_dir/progress.md"
```

The shell scripts (`task-brief`, `review-package`) already use this resolved-path pattern correctly — only the prose instructions in `SKILL.md` use the inline `$(cmd)` form.

## Affected file

`skills/subagent-driven-development/SKILL.md`, line ~254 (Durable Progress section)

Open skills/subagent-driven-development/SKILL.md around line 254 and read the Durable Progress section, then compare its ledger append example with the resolved-path pattern in task-brief and review-package. Done means the SKILL.md prose uses a resolved variable for the redirect target rather than inline $(git rev-parse --git-path sdd), with no other behavior changes.

Escrito por el modelo de indexación a partir del texto del issue.