hyperlight-codeact Skill
scriptsCodeAct via Hyperlight. Use when looping over many files (8+), cross-referencing results from multiple sources, aggregating data across directories, or chaining 5+ dependent tool calls. Collapses N round-trips into one sandboxed Python run via scripts/codeact.py. NOT beneficial for <5 files or simple grep-then-view — direct tool calls have less overhead at small scale. ESPECIALLY valuable when MCP servers are loaded — fewer turns means the MCP tool catalog context is replayed fewer times. Run `scripts/codeact.py --discover` to see available sandbox tools. Trigger: "codeact", "chain tools", "sandbox", "batch", "for each", "hyperlight sandbox", "collapse tool calls", "run in sandbox", "sandbox execution".
- GitHub Stars
- 44
- Dateien
- 3
- Letzter Commit
- 1. Mai 2026
- Lizenz
- —
- Erlaubte Tools
- —
Was macht der hyperlight-codeact Skill?
hyperlight-codeact ist ein Agent Skill. CodeAct via Hyperlight. Use when looping over many files (8+), cross-referencing results from multiple sources, aggregating data across directories, or chaining 5+ dependent tool calls. Collapses N round-trips into one sandboxed Python run via scripts/codeact.py. NOT beneficial for <5 files or simple grep-then-view — direct tool calls have less overhead at small scale. ESPECIALLY valuable when MCP servers are loaded — fewer turns means the MCP tool catalog context is replayed fewer times. Run `scripts/codeact.py --discover` to see available sandbox tools. Trigger: "codeact", "chain tools", "sandbox", "batch", "for each", "hyperlight sandbox", "collapse tool calls", "run in sandbox", "sandbox execution".
So installierst du den hyperlight-codeact Skill
Jeder Befehl kopiert den Skill-Ordner in das Skills-Verzeichnis des Clients. Starte den Client neu oder beginne eine neue Session, um ihn zu laden.Den hyperlight-codeact Skill in Claude Code installieren
~/.claude/skills/hyperlight-codeact/SKILL.md
$ npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact ~/.claude/skills/hyperlight-codeactDen hyperlight-codeact Skill in Claude Code (project) installieren
.claude/skills/hyperlight-codeact/SKILL.md
$ npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact .claude/skills/hyperlight-codeactDen hyperlight-codeact Skill in Codex installieren
~/.agents/skills/hyperlight-codeact/SKILL.md
$ npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact ~/.agents/skills/hyperlight-codeactDen hyperlight-codeact Skill in Gemini CLI installieren
~/.gemini/skills/hyperlight-codeact/SKILL.md
$ npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact ~/.gemini/skills/hyperlight-codeactDen hyperlight-codeact Skill in Claude.ai installieren
npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact hyperlight-codeact && zip -r hyperlight-codeact.zip hyperlight-codeact # claude.ai → Customize → Skills → + → Create skill → Upload a skill → hyperlight-codeact.zip
SKILL.md
Auf GitHub ansehenHyperlight CodeAct
Collapse multi-step tool chains into a single sandboxed Python execution inside an isolated Hyperlight micro-VM (WebAssembly).
Syntax
Use call_tool(name, **kwargs) — built-in global, no import needed:
content = call_tool('view', path='src/main.py')
files = call_tool('glob', pattern='**/*.py')
hits = call_tool('grep', pattern='TODO', paths='src')
result = call_tool('bash', command='git log --oneline -5')
call_tool('edit', path='config.json', old_str='"debug": false', new_str='"debug": true')
Return types (critical — wrong assumptions cause retries)
call_tool('glob', ...)→ list of strings like["src/app.py", ...]call_tool('view', ...)→ string (file content)call_tool('bash', ...)→ dict withstdout,stderr,returncodecall_tool('mcp_call', ...)→ string
Pattern
for f in call_tool('glob', pattern='src/**/*.py'):
try:
content = call_tool('view', path=f)
except Exception:
continue
# analyze content...
print(f"{f}: {result}")
Rules
- One bash call, one program. Do not scout with separate tool calls first.
- Wrap file reads in try/except.
glob()returns workspace-relative paths — pass directly toview().- Brace expansion works:
call_tool('glob', pattern='src/{db,services}/**/*.py')
Discover tools
python3 scripts/codeact.py --discover # JSON manifest
python3 scripts/codeact.py --instructions # LLM-ready reference
Tools are auto-detected based on what's installed on the host.
Execute
uv run --python 3.13 --with 'hyperlight-sandbox[wasm,python_guest]>=0.3.0' \
python3 scripts/codeact.py --auto --workspace . --code '...'
Output: {"stdout": "...", "stderr": "...", "exit_code": 0, "success": true}
Trust model
Sandboxed code can only reach the outside world through call_tool() bridges.
Tools run on the host with full process access. Use --workspace to restrict
file tools to a directory tree.
Prerequisites
- Python 3.10–3.13 (Wasm backend has no wheels for 3.14+)
uv(recommended) orpip install 'hyperlight-sandbox[wasm,python_guest]>=0.3.0'
References
Dateien in diesem Skill
3hyperlight-codeact Skill-Fragen
Wie installiere ich den hyperlight-codeact Skill in Claude Code?
Führe npx degit jsturtevant/copilot-codeact-plugin/plugins/codeact/skills/hyperlight-codeact ~/.claude/skills/hyperlight-codeact aus. Claude Code lädt Skills aus ~/.claude/skills (alle Projekte) oder .claude/skills (ein Projekt) und nutzt einen Skill, wenn seine Beschreibung zur Aufgabe passt.
Welche Clients können den hyperlight-codeact Skill nutzen?
Jeder Client, der das Agent Skills-Format liest: Claude Code, Codex, Gemini CLI und claude.ai (als ZIP hochladen). Skripte im Skill laufen nur dort, wo der Client Codeausführung erlaubt.