claude-runner
Execute an external Claude CLI prompt in headless print mode. Use only for an explicit Claude CLI request, a selected Claude route unavailable as native host delegation, or an approved external fallback.
How do I install this agent skill?
npx skills add https://github.com/robsonrung/rar-skills --skill claude-runnerIs this agent skill safe to install?
- Gen Agent Trust Hubpass
A secure wrapper for the Claude CLI that enables workspace interaction with configurable tool permissions, session persistence, and automated result validation.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
Claude Runner
Use shared/references/model-preview.md to choose the branch first. Deterministic checks need no model selection. Actual dispatch uses one concrete approved snapshot; nested calls consume it without re-resolution or extra workers.
Execute the caller's scoped prompt through the local CLI. Shared roles, envelope keys, and result handling live in shared/references/runner-common.md. Preserve seat fidelity: name the provider that actually answered, and report unavailable or unverified seats honestly.
Native routing
Read shared/references/host-model-execution.md before choosing an external route. If the current host can dispatch the exact selected Claude model with the required effort, isolation, and receipt policy in a native subagent or task thread, use that route. agents/openai.yaml is host UI metadata; it does not dispatch a native model.
Invocation
Set SKILL_DIR to this loaded skill directory in the same shell call:
SKILL_DIR="<absolute path of this skill directory>";
python3 "$SKILL_DIR/scripts/run_claude.py" "<scoped prompt>" \
--model <approved-model-id> --effort <approved-effort> --disable-fallback --json
Copy the exact model and effort from the selected snapshot. Add its role, tool mode, receipt policy, and limits. Omit an effort flag only for a selected runtime controlled route. Nested calls use these saved options without resolving defaults again. The example does not grant source sharing or tool authority.
Runtime boundaries
- Use the local authenticated
claudeCLI in print mode. Permission checks remain enabled. - Analysis roles default to
repo_read_only: only Read, Glob, and Grep, with customizations disabled and an empty MCP configuration.--restrict-toolsforces this profile;--tool-profile no_toolsremoves all tools.--allow-writeor--tool-profile writekeeps normal write permissions within the caller's authority. - Prompt and file context may be sent to the configured provider. Bare mode changes authentication behavior; read the runtime reference before using it.
- The wrapper forwards model and effort. Primary assistant events can prove the serving model. Requested labels, initialization labels, synthetic events, and usage labels cannot. Missing evidence stays unverified; conflicting primary IDs or an exact model mismatch fail the result.
- Direct calls can fall back once when the CLI is missing and the fallback can preserve the constraints. Tool profiles and native budget limits block fallback. Approved routes disable fallback; native session continuation must preserve the selected runner.
Use --max-turns, --max-budget-usd, and --timeout for native turn, reported cost, and elapsed time limits. All must be positive and finite. A requested response length is advisory.
Load by need
- references/command-reference.md: flags, examples, native mappings, and return codes when composing a non-basic invocation.
- references/runtime-reference.md: authentication, fallback conditions, configuration, and extended envelope fields when configuring a route or interpreting its result.
- references/continuation.md: only when resuming a session or managing a background job.
Result
Read agent_message and the wrapper envelope. Check success, serving-model receipt, effective runner/provider, and any fallback or malformed-output status. A completed process alone does not prove the requested task or model was used. Report touched files when the caller authorized edits.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/robsonrung/rar-skills/claude-runner">View claude-runner on skillZs</a>