harness scorecard

    1

    Audit, grade, and secure your coding-agent harness with an evidence-based 10-dimension scorecard.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    harness scorecard

    Example session with this skill installed

    Run a full audit on the current repository's agent harness and output the scorecard to the ./audits directory.

    • Read your context and instructions
    • Compiled the harness scorecard
    • Generated the document

    Audit complete. I have processed the 10-dimension scanner groups and verified findings against the repository evidence. The run has been sealed and the integrity of the artifacts verified.

    Harness Audit Result: Grade B

    DimensionMaturity LevelStatus
    D1: Instruction ClarityLevel 4 (Managed)Passed
    D2: Tool DefinitionLevel 3 (Defined)Passed
    D3: State ManagementLevel 2 (Repeatable)Warning
    D4: Subagent DelegationLevel 4 (Managed)Passed
    D5: Security HooksLevel 1 (Initial)Critical
    D6: Error HandlingLevel 3 (Defined)Passed
    D7: Context EngineeringLevel 2 (Repeatable)Warning
    D8: CI/CD IntegrationLevel 0 (None)Failed
    D9: ObservabilityLevel 2 (Repeatable)Passed
    D10: Safety GuardrailsLevel 1 (Initial)Critical

    Top 3 Fixes Needed:
    1.

    Implement Write Guards: Current security hooks allow broad filesystem access; implement hooks/readonly_guard.py or equivalent.
    2.

    Formalize CI Integration: No evidence of agent evaluation or linting in CI workflows; add .harness-scorecard checks to your pipeline.
    3.

    Strengthen State Isolation: Agent state is currently leaked across subagent calls; use structured state handoffs.

    The full report is available at ./audits/.harness-scorecard/run_8829/report.html.

    Next steps

    • Review the report.en.md file for specific code locations of the D5 and D10 failures.
    • Run HSC verify-seal ./audits/.harness-scorecard/run_8829/ to confirm artifact authenticity.
    • Apply the suggested readonly_guard.py to move from Grade B to A.

    report.html

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Manual review of agent instructions and hooks is slow and prone to bias. This skill provides a structured, evidence-graded audit for your repository's AI agent harness, evaluating everything from tool definitions and subagent delegation to security hooks and CI workflows. It replaces subjective assessments with a verifiable 10-dimension scorecard.

    What it does

    • Automated Discovery identifies agent instructions, tool definitions, and orchestration logic within any repository.
    • Multi-Dimension Scoring evaluates your harness across 10 specific categories including state management, security hooks, and subagent delegation.
    • Evidence Verification ensures every finding is backed by repo data, assigning maturity levels from E0 to E5.
    • Safe Auditing operates in a read-only mode to prevent accidental modification of the codebase during the review.
    • Artifact Generation produces standardized reports in Markdown, HTML, and SARIF formats, alongside SVG badges for documentation.

    How it works

    1. Initialize the audit by running HSC start on the target repository to set up a dedicated run directory.
    2. Scan the codebase using dimension-specific scanners to gather candidates for instructions, tools, and hooks.
    3. Verify findings through an inference phase that cross-references candidates against the repository to confirm evidence levels.
    4. Finalize the run to calculate the letter grade, generate the sealed reports, and verify the integrity of the output artifacts.

    Frameworks & tools

    Built for environments using Claude Code, Cursor, GitHub Copilot, or custom Python-based agent harnesses. The underlying engine requires Python 3.11+.

    Why this beats prompting it yourself

    Generic prompts often hallucinate security capabilities or miss subtle configuration flaws in complex harnesses. This skill uses a rigid, script-driven flow that forces evidence collection and forbids the model from "guessing" the final grade, ensuring an objective audit.

    Use cases

    • Auditing a new agent harness for production readiness and security gaps.
    • Benchmarking the maturity of internal AI engineering configurations.
    • Generating standardized SARIF reports for CI/CD compliance checks.
    • Reviewing third-party agent repositories before integration.

    Known limitations

    Requires Python 3.11 or higher. Probing runtime hooks requires a temporary clone of the repository. Does not support third-party network scanners without explicit opt-in flags.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.

    Requires Python 3.11+. Runtime probing requires a temporary clone. No network access or external sensor usage unless explicitly enabled via flags.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    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 to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 7 days ago

    • Free with an Agensi account

    Accounts
    GitHub
    Software
    Python

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Environment Variables

    Allowed Hosts

    github
    api.github.com
    Listed7 days ago

    What's inside

    Frequently Asked Questions

    The scorecard evaluates your harness across 10 critical dimensions, including security hooks, tool definitions, state management, and subagent delegation, assigning a maturity level from E0 to E5 based on verifiable repository data.

    Yes, it is designed to audit harnesses for popular environments like Claude Code, Cursor, and GitHub Copilot, as well as custom-built Python agent frameworks.

    The skill operates in a read-only mode for the audit itself, but a temporary clone of your repository is required to safely probe runtime hooks without affecting your production code.

    Beyond the core audit engine, you receive standardized reporting templates in Markdown, HTML, and SARIF formats, plus dynamic SVG badges to display your harness maturity level in your README.

    This skill requires Python 3.11 or higher and works best on repositories where agent instructions and orchestration logic are stored as code or configuration files.

    Standard LLM prompting often 'hallucinates' security features; this skill uses a script-driven workflow that mandates evidence collection, preventing the AI from guessing or overlooking subtle configuration flaws.

    New in Prompt & Skill Engineering

    Free