Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Work Order Writer: Guardrailed Coding-Agent Handoffs

    2

    Writes a complete work order for a coding agent, with guardrails, a measure-first step, what must and must not change and checkable acceptance criteria, then checks the agent's report against it.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    2 usesSecurity scanned
    Agent Work Order Writer: Guardrailed Coding-Agent Handoffs

    Agent Work Order Writer: Guardrailed Coding-Agent Handoffs

    Example session with this skill installed

    Here is the order and the agent's report. Did it actually do what I asked?

    • Read your context and instructions
    • Compiled the agent work order

    I ran the order check on the bundled demo: an invented work order (WO-142) for a small meal-planning app that should export dates in UTC.

    Step 1: check the order as written

    sectionstatusline
    Goalpresent9
    Contextpresent13
    Step 0 (measure before changing)present18
    Guardrailspresent22
    Out of scopepresent27
    End statepresent31
    Acceptance criteriapresent36
    Stop conditionsempty43
    Reportpresent47

    Acceptance criteria (4)

    idlinenotes
    AC138ok
    AC239vague wording (works)
    AC340nothing measurable; no denominator (all without a number)
    AC441ok

    Result: 5 flag(s)

    • Stop conditions, line 43: section empty
    • End state, line 31: end state does not say what must not change
    • AC2, line 39: vague wording (works)
    • AC3, line 40: nothing measurable
    • AC3, line 40: no denominator (all without a number)

    Step 2: rewrite the flagged parts

    
### Must not change
The CSV header line and column order (compared with the line recorded in Step 0), the PDF export and the database
schema (`git diff --stat main -- app/export/pdf.py migrations/` prints nothing).

## Acceptance criteria

AC1: `pytest tests/export` passes, 38 of 38.
AC2: An export made with the fixture account set to UTC+10 shows 2026-03-01T00:00:00Z for the meal planned on 1 March (checked in the downloaded file, not in the code).
AC3: `grep -c` for dates without a trailing Z in that export gives 0 of 126 date cells.
AC4: `cmp` of the export's first line with the header line recorded before the change exits 0.

## Stop conditions

Stop and report if the Step 0 test count is not 38 of 38, if the header line differs before any change, or if the change needs a schema migration.

    Check on the rewrite

    Result: 0 flag(s)

    No flags. This checks structure and wording only; it cannot tell whether the order is right.

    Step 3: check the agent's report against the order

    criterionresultwhyreport line(s)
    AC1metstatus met with evidence24
    AC2not documentedsays met but shows no evidence (no code, link or count such as 3 of 3)25
    AC3not metthe report says it is not met26
    AC4not documentednot in the report's acceptance criteria section-

    1 of 4 criteria met

    A criterion that is not documented counts as not verified. Ask for the missing checks.

    What this means

    The order check caught an empty Stop conditions section, a missing "Must not change" part, a vague criterion ("works as expected") and one with no number to count against.

    The report check caught a criterion claimed as met with no evidence (AC2), one the agent itself marks as not met (AC3) and one it never mentions (AC4).

    • Next: send the agent back for AC2 to AC4 with the exact checks the order names, and do not merge on this report.

    Want me to write a work order for your own task? I ask a few setup questions first, then check the order before you hand it over.

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

    About this skill

    Use when you hand a task to a coding agent and want it done the way you meant, or when you need to check whether an agent's report really shows the work is done. The skill writes the order in nine fixed sections: goal, context, a measure-first step, guardrails, out of scope, an end state with "must change" and "must not change", acceptance criteria with numbers and commands, stop conditions, and the report format. A standard-library Python script checks the order for missing or empty sections, vague or unmeasurable criteria, criteria without a denominator and strings that look like credentials (shown as type and length, never the value). A second script reads the agent's report and marks each criterion met, not met or not documented; a claim of "met" without evidence, or a "met" line that also uses a listed failure word (such as failed, error or not run), counts as not documented. The checks read structure and wording; they cannot tell whether the goal is right or the evidence is true. Do not use when you want the code written here and now, a project plan or roadmap, or a code review that is not tied to a work order.

    Example prompts

    1. "Write a work order for the agent: move our CSV export to UTC dates, and don't let it touch the database."
    2. "Turn this ticket into an agent task with acceptance criteria I can check."
    3. "Here is the order and the agent's report. Did it actually do what I asked?"

    Requires

    Python 3.9+ (standard library only) and bash or zsh for the report command. On claude.ai the demo runs with code execution enabled. The orders it writes work with any coding agent that reads Markdown.

    Compatibility

    Written for SKILL.md-compatible agents, for example Claude Code, Claude.ai skills, Codex CLI, Cursor or Gemini CLI. Each documents the SKILL.md format; not every agent was tested. The seller suggests Claude Sonnet 4.5 or later.

    Limits

    Section headings must stay in English as in the template; the vague-word and status-word lists are English only. "Measurable" is a text rule (a digit or a code span), and the word lists are the skill's own heuristics. The report check reads status words, failure words and evidence markers; it does not compare a count such as "2 of 3" with the criterion. Credential detection covers common formats only.

    Independence

    Not affiliated with or endorsed by Anthropic, OpenAI, Anysphere (Cursor) or Google. Product names are used only to describe compatibility.

    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

    2 uses

    Times used to date

    Money-back guarantee

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

    Trust & safety

    Security scan

    Security scan passed 5 days ago

    Requires

    Python 3.9+ (standard library only)
    bash or zsh (for the report command)
    Python
    Git

    Listed5 days ago
    Updated1 day ago

    What's inside

    Frequently Asked Questions