runledger guard
Add real-time safety guards, risk scoring, and auditable session receipts to AI coding agents.
Free
Works with the AI Agents you already use
runledger guard
Example session with this skill installed
Install the runledger guard for Claude Code in this project and generate a markdown receipt for the last session to check for risks.
- Read your context and instructions
- Compiled the runledger guard
I have installed the RunLedger PreToolUse guard for Claude Code and generated a risk assessment of your most recent agent session.
RunLedger Security Audit
| Metric | Value | Status |
|---|---|---|
| Session ID | rl_9a2b4c81 | Active |
| Risk Score | 18/100 | LOW |
| Quality Grade | A- | Passed |
| Total Cost | $0.14 | Within Budget |
| Guard Status | Installed | Intercepting |
Session Summary
The session involved 4 tool calls primarily focused on FileRead and Grep. One Bash command was intercepted and allowed as it was a non-destructive npm test. No high-risk patterns (e.g., recursive deletes, credential exfiltration) were detected.
Hook Configuration
{
"agent": "claude-code",
"hook_path": ".claude/settings.json",
"status": "enabled",
"commands": [
"runledger guard --session rl_9a2b4c81"
]
}
Next steps
- Run
runledger receipt --opento view the full HTML diffs in your browser. - Add
runledger receipt --fail-on 60to your GitHub Actions workflow to prevent high-risk PRs. - Use
runledger doctorif you notice hooks aren't firing after an agent update.
Connects securely to your tools. The creator never sees your data.
About this skill
Manual oversight of AI coding agents is a bottleneck for high-velocity teams. This skill provides a programmable safety layer and audit trail for agents like Claude Code, Codex, and Factory Droid, ensuring they don't execute destructive commands or exceed risk thresholds.
What it does
- Real-time guarding intercepting
PreToolUseevents to block or flag risky agent actions before they execute. - Risk scoring assigning a 0 to 100 score to agent sessions based on tool usage and file diffs.
- Session receipts generating HTML, Markdown, or JSON audit trails with cost estimates and quality grades.
- CI/CD gating failing builds automatically if an agent session exceeds a predefined risk threshold.
- Multi-agent support configuring hooks for Claude Code, Codex CLI, and Factory Droid simultaneously.
How it works
- Verify the environment by running
runledger doctorto check for misconfigured hooks. - Install the guard using
runledger guard installfor specific agents or the entire project. - Test policy logic with the
guard testcommand using a JSON payload of a hypothetical tool call. - Audit results by generating a receipt after the agent run to review diffs and risk grades.
Frameworks & tools
This skill utilizes the runledger-ai Python package (3.9+). It integrates directly with the configuration files of Claude Code, Codex CLI, and Factory Droid.
Why this beats prompting it yourself
System prompts for safety are easily bypassed by jailbreaks or complex reasoning loops. This skill moves safety logic to the execution layer, providing a hard programmatic gate that agents cannot ignore or "convince" to disappear.
Use cases
- Prevent agents from executing
rm -rfor other destructive filesystem commands. - Review token costs and quality grades for every agent-generated PR.
- Block agent runs in CI if the generated code risk score exceeds your team's limit.
- Maintain a compliance-ready history of all autonomous tool calls in a repository.
Known limitations
Requires Python 3.9 or higher and manual trust of hooks in Codex CLI or Factory Droid interfaces. Guarding logic depends on the agent correctly triggering the PreToolUse hook.
Requires Python 3.9+. User must manually trust hooks within Codex or Droid interfaces. Protection is limited to agents that support lifecycle hooks.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 3 days ago
- Free with an Agensi account
Requires