Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/raphaelchristi/harness-evolver/dev-dry-runnpx skills add raphaelchristi/harness-evolver --skill dev-dry-rungit clone --depth 1 https://github.com/raphaelchristi/harness-evolverWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00049 | $0.01098 |
| Opus 5 | $0.00024 | $0.00549 |
| Sonnet 5 | $0.00010 | $0.00220 |
| Haiku 4.5 | $0.00005 | $0.00110 |
Grade C, and why
dev:dry-run scanned grade C with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
rm -rf "$TMPDIR" How it starts
The opening of the file, as written. The whole thing — 144 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/dev:dry-run
Smoke-test the evolve pipeline. Two modes depending on whether LANGSMITH_API_KEY is available.
Resolve Paths
TOOLS="${EVOLVER_TOOLS:-$([ -d "tools" ] && echo "tools" || echo "$HOME/.evolver/tools")}"
EVOLVER_PY="${EVOLVER_PY:-$([ -f "$HOME/.evolver/venv/bin/python" ] && echo "$HOME/.evolver/venv/bin/python" || echo "python3")}"
Check: Online or Offline?
if [ -n "$LANGSMITH_API_KEY" ]; then
echo "MODE: Online (LANGSMITH_API_KEY found)"
MODE="online"
else
echo "MODE: Offline (no LANGSMITH_API_KEY)"
MODE="offline"
fi
Offline Mode (no API key)
Validate tool syntax and argparse consistency:
echo "=== Tool Syntax Check ==="
for f in $TOOLS/*.py; do
python3 -c "import ast; ast.parse(open('$f').read())" 2>&1
if [ $? -eq 0 ]; then echo "OK: $(basename $f)"; else echo "FAIL: $(basename $f)"; fi
done
echo ""
echo "=== Argparse Flags Check ==="
for f in $TOOLS/*.py; do
$EVOLVER_PY "$f" --help > /dev/null 2>&1
if [ $? -eq 0 ]; then echo "OK: $(basename $f) --help"; else echo "FAIL: $(basename $f) --help"; fi
done
echo ""
echo "=== Skill Cross-Reference Check ==="
# Check every tool referenced in evolve skill exists
for TOOL in $(grep -oh '\$TOOLS/[a-z_]*.py' skills/evolve/SKILL.md | sed 's/\$TOOLS\///' | sort -u); do
if [ -f "$TOOLS/$TOOL" ]; then
echo "OK: $TOOL referenced and exists"
else
echo "FAIL: $TOOL referenced in evolve skill but not found"
fi
done
Online Mode (with API key)
Run the full pipeline with a mock agent:
1. Create temp directory with mock agent
TMPDIR=$(mktemp -d)
cat > "$TMPDIR/agent.py" << 'PYEOF'
import json, sys
input_path = sys.argv[1] if len(sys.argv) > 1 else None
if input_path:
with open(input_path) as f:
data = json.load(f)
question = data.get("input", data.get("question", ""))
print(json.dumps({"output": f"Mock answer to: {question}"}))
else:
print(json.dumps({"output": "No input provided"}))
PYEOF
cat > "$TMPDIR/test_inputs.json" << 'JSONEOF'
[
{"input": "What is 2+2?"},
{"input": "Name a color"},
{"input": "What is Python?"}
]
JSONEOF
echo "Mock agent created at $TMPDIR"
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 144 lines · 49 tokens per session scan C b0e519fd1078
dev:dry-run is a skill published in the GitHub repository raphaelchristi/harness-evolver (49 stars, last pushed 4mo ago), licensed MIT. It adds 49 tokens to every session and 1,098 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
playwright-cli
Automates browser interactions for testing and validating your own web applications using playwright-cli. Use when you need terminal-first browser control for navigation, form filling, screenshots, tracing, bound browser sessions, debugging, or generating Playwright test code. Only use against applications you own or…
cross-browser-qa
Run cross-browser QA for websites and apps. Use before production or after CSS, media, forms, animation, canvas, layout, or interaction changes to test Chromium, WebKit/Safari-like, and Firefox behavior, browser-specific CSS/media issues, focus/keyboard differences, and responsive rendering.
motion-performance-qa
Run motion performance QA for websites and apps. Use after adding Motion, GSAP, Lenis, scroll scenes, page transitions, hover states, carousels, WebGL sync, observers, timers, RAF loops, or any animation-heavy UI to catch jank, layout shift, leaks, cleanup bugs, reduced-motion gaps, and mobile issues.
webapp-testing
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
integration
Run integration and e2e tests after unit tests pass. Use after /supergraph:fix when unit tests are green.