More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Tool-Call Contract Auditor Pro

    1

    Schema lint, replay fixtures, CI gate, and repo-wide MCP tool-surface scan. Free teaches the checklist; Pro automates the merge gate.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Tool-Call Contract Auditor Pro

    Tool-Call Contract Auditor Pro

    Example session with this skill installed

    Run a contract audit on my create_invoice tool using the schemas/create_invoice.schema.json and the fixtures/bad-args.json trace. Output a verdict report.

    • Read your context and instructions
    • Compiled the tool-call contract auditor

    I have performed a contract audit on the create_invoice tool. The linting process identified a contract break where the agent attempted to pass a string to a field requiring an integer.

    Tool-Call Contract Audit Results

    CheckResultDetails
    Schema LintFAILFound 1 validation error in fixtures/bad-args.json
    Contract MatchPASSCall matches defined templates/tool-contract.yml rules
    Side-Effect ClassCRITICALNon-idempotent: Requires Idempotency-Key which was missing
    VerdictBAD_ARGSType mismatch on property amount_cents

    Error Log

    {
  "status": "error",
  "error_type": "arg_mismatch",
  "field": "amount_cents",
  "expected": "integer",
  "actual": "string",
  "message": "Value '1200.50' is not of type 'integer'"
}

    Post-Call Verdict

    The agent attempted a double-fire on a non-idempotent tool after a timeout. Because the Idempotency-Key was not present in the original schema, this would have resulted in a duplicate invoice in production.

    Next steps

    • Update schemas/create_invoice.schema.json to strictly enforce the integer type for amount_cents.
    • Implement an Idempotency-Key requirement in the tool contract to prevent double-fire errors.
    • Run python3 scripts/replay_tool_trace.py against the updated contract to verify the fix.

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

    About this skill

    Schema lint, replay fixtures, CI gate, and repo-wide MCP tool-surface scan. Free teaches the checklist; Pro automates the merge gate.

    Free vs Pro

    • Free audits tool use in conversation with a checklist and failure taxonomy; Pro lints JSON Schema and OpenAPI tool args via scripts/lint_tool_schemas.py
    • Free stops at chat guidance; Pro replays recorded traces against contracts via scripts/replay_tool_trace.py
    • Free has no CI story; Pro ships templates/github-action-tool-contract.yml that fails the merge on contract breaks
    • Free leaves reporting to chat; Pro exports Markdown/JSON verdict reports for pull requests
    • Free covers one session; Pro adds per-tool contract YAML — allow/deny, max retries, side-effect class

    Upgrade Path

    Upgrade when you maintain more than three tools, ship MCP surfaces to others, or need a merge gate. Free stays useful for a single-session review; Pro is for teams that change tools every sprint.

    Free version: https://www.agensi.io/skills/tool-call-contract-auditor

    The Problem

    A checklist catches a bad call in conversation, then the same contract break ships in the next pull request because nothing in CI reads the tool schema. Schema drift stays invisible until a caller passes a string where the schema says integer, and by then the invoice exists twice.

    The free skill has no artifact a reviewer can open and no gate that fails a branch. Reviewers get a chat transcript, not a verdict.

    What You Get

    • scripts/lint_tool_schemas.py — validate tool arguments against JSON Schema or OpenAPI params, with JSON output for CI
    • scripts/replay_tool_trace.py — replay recorded fixtures against a contract file and return PASS or FAIL per fixture
    • templates/tool-contract.yml — per-tool allow/deny, max retries, side-effect class, required idempotency
    • templates/github-action-tool-contract.yml — CI wiring that fails the merge on a contract break
    • templates/verdict-report.md — a pull-request-ready export a reviewer can read in one page
    • fixtures/ — recorded traces for good-call, bad-args, double-fire, partial-as-success, skip-retry
    • schemas/ — sample tool JSON Schemas for create_invoice and transfer_funds
    • references/failure-taxonomy.md and references/idempotency-patterns.md — the same shared vocabulary as the free skill

    Why This Beats Prompting It Yourself

    Asking an agent to check the tool schema returns prose. The Pro scripts return an exit status. A demo fixture pair makes the difference concrete: before, the agent retries create_invoice four times and creates four invoices; with Pro, the replay fixture INV-create-001 FAILs on a missing Idempotency-Key and CI blocks the merge until the contract YAML is updated.

    Example Session

    Run the lint against a schema and a recorded bad call, then replay the fixture set against the contract.

    Lint reports one validation error: amount_cents received the string "1200.50" where the schema requires an integer. Replay reports the double-fire fixture as FAIL — a non-idempotent tool re-called after a timeout with no idempotency key in the original schema. Verdict is BAD_ARGS on the first, CRITICAL side-effect on the second, and the merge is blocked until both are fixed.

    Use Cases

    • Block pull requests that break a tool JSON Schema or OpenAPI parameter contract
    • Replay historical agent traces against a contract to find which failure modes actually occurred
    • Standardize retry and idempotency rules across a repo of MCP tools
    • Produce a verdict report a reviewer can attach to a pull request

    Known Limitations

    • Uses the Python standard library subset for JSON Schema validation unless jsonschema is installed
    • Does not detect agent loops or prompt-injection attempts
    • Limited to tools exposed to the agent harness
    • Does not replace provider-side idempotency

    Tags: agents, mcp, tools, reliability, ci, json-schema Version: 1.0

    Standard library subset used for JSON Schema validation unless jsonschema is installed. Does not detect agent loops or prompt-injection attempts. Limited to tools exposed to the agent harness.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 day ago

    Requires

    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    The skill provides a specialized framework to enforce strict JSON Schema validation and idempotency checks, preventing common issues like silent failures, corrupted state from accidental double-fires, and mismanaged partial successes that generic prompting often misses.

    It is built for engineers using Claude Code, Cursor, and Codex, and is fully compatible with any agent supporting the Model Context Protocol (MCP) tool-calling standard and SKILL.md.

    You receive a complete reliability kit including YAML contract templates, Python validation scripts, GitHub Action templates for CI/CD integration, and a standardized failure taxonomy for debugging.

    Setup involves three main steps: defining your tool contracts using the provided templates, running the Python linting scripts against your schemas, and wiring the GitHub Action template into your existing PR workflow.

    The skill uses Python's standard library for basic validation but can be enhanced by installing the `jsonschema` library for more complex validation needs; however, it does not currently handle loop detection or security-focused prompt injection reviews.

    Yes, the included trace replaying feature allows you to run recorded execution fixtures against your defined contracts to verify how your agent handles complex scenarios like partial successes or repeated calls.