feat(adapters): Superpowers skill evaluation adapter

.gitignore

@@ -64,3 +64,4 @@ docs/让*
 tests/run_*.sh
 tests/launch_*.py
 *.launch.log
+uv.lock

docs/superpowers/SECURITY.md

@@ -0,0 +1,48 @@
+# Security Considerations for Superpowers Adapter
+
+## Execution Model
+
+The Superpowers adapter runs Claude Code with candidate skills that control agent behavior. A malicious skill can:
+
+- Execute arbitrary shell commands
+- Read/write files in the project directory
+- Access environment variables (including API keys)
+- Make network requests
+
+## Current Mitigations
+
+1. **Scrubbed environment**: Only essential vars passed (HOME, PATH, TERM, LANG, ANTHROPIC_API_KEY)
+2. **Isolated HOME**: Each scenario gets its own HOME directory
+3. **No `--dangerously-skip-permissions` by default**: Permission prompts required unless explicitly bypassed
+4. **Project directory isolation**: Each scenario gets its own project directory
+
+## Known Limitations
+
+- **API key exposure**: ANTHROPIC_API_KEY is passed to the subprocess
+- **No OS-level isolation**: Without Docker/bubblewrap, candidate code runs with user privileges
+- **SKILLOPT_UNSAFE bypass**: When enabled, full filesystem access
+
+## Recommendations
+
+### For Local Testing (trusted candidates)
+```bash
+SKILLOPT_UNSAFE=1 python -m skillopt_sleep.adapters.superpowers --candidate my_skill.md
+```
+
+### For Untrusted Candidates (future work)
+
+Docker isolation (not yet implemented):
+```bash
+# Build sandbox image
+docker build -t skillopt-sandbox .
+
+# Run with isolation
+python -m skillopt_sleep.adapters.superpowers --candidate untrusted.md --sandbox docker
+```
+
+## Follow-up Work
+
+- [ ] Docker sandbox implementation
+- [ ] Bubblewrap (bwrap) support for Linux
+- [ ] Network isolation option
+- [ ] API key injection via Docker secrets

scripts/smoke_superpowers.sh

@@ -0,0 +1,71 @@
+#!/bin/bash
+# Smoke test for Superpowers adapter integration.
+# Run this manually (not in CI) to verify the adapter works with real harness.
+#
+# Prerequisites:
+# - Harness installed and authenticated
+# - Same model/settings for baseline and candidate runs
+#
+# Usage:
+#   SKILLOPT_UNSAFE=1 ./scripts/smoke_superpowers.sh [candidate_skill_path]
+#
+# Output: writes results + raw output to smoke_results/ for PR evidence.
+# Fails on any runner error (no silent swallowing).
+
+set -euo pipefail
+
+SKILL="${1:-}"
+OUTDIR="smoke_results/$(date +%Y%m%d_%H%M%S)"
+mkdir -p "$OUTDIR"
+
+echo "Smoke test: Superpowers adapter"
+echo "Output: $OUTDIR"
+echo "SKILLOPT_UNSAFE=${SKILLOPT_UNSAFE:-0}"
+echo ""
+
+run_scenario() {
+    local name="$1"
+    local candidate="${2:-}"
+    local outfile="$OUTDIR/${name}.json"
+
+    echo "=== $name ==="
+
+    local args=(
+        --skill verification-before-completion
+        --scenario test-passes-verify
+        --json
+    )
+    if [[ -n "$candidate" ]]; then
+        args+=(--candidate "$candidate")
+    fi
+
+    # No || true - fail if runner errors
+    python -m skillopt_sleep.adapters.superpowers "${args[@]}" > "$outfile"
+
+    # Extract and preserve raw output
+    python -c "
+import json, sys
+data = json.load(open('$outfile'))
+for s in data.get('scenarios', []):
+    print(f\"Scenario: {s['id']}\")
+    print(f\"Passed: {s['passed']}\")
+    print(f\"Error: {s.get('error', 'none')}\")
+    # Raw output preserved in JSON, print preview
+    out = s.get('output', '')
+    if out:
+        print(f\"Output preview ({len(out)} chars):\")
+        print(out[:500])
+    print()
+"
+}
+
+# Baseline run (stock skill)
+run_scenario "baseline"
+
+# Candidate run if provided
+if [[ -n "$SKILL" ]]; then
+    run_scenario "candidate" "$SKILL"
+fi
+
+echo "Results saved to $OUTDIR"
+echo "Include these files in your PR as evidence of smoke test."

skillopt_sleep/adapters/__init__.py

@@ -0,0 +1 @@
+"""SkillOpt adapters for external skill frameworks."""