- Home
- Skills
- Agents & Orchestration
- Did the Agent Actually Do It? — Post-Run Verification & Side-Effect Audit for Delegated Agent Tasks
Did the Agent Actually Do It? — Post-Run Verification & Side-Effect Audit for Delegated Agent Tasks
Every agent-testing bundle on this marketplace tests code: it diffs two runs, turns a failed trace into a regression fixture, builds a JSONL eval dataset, and gates a dataset's provenance. Every deleg
$1
Works with the AI Agents you already use
Did the Agent Actually Do It? — Post-Run Verification & Side-Effect Audit for Delegated Agent Tasks
Example session with this skill installed
My browser agent says it ended the duplicate marketplace listing for my belt lot. Did it actually do it, and did it touch anything else?
- Read your context and instructions
- Compiled the did the agent
Verdict: VERIFIED LANDED (or LANDED ELSEWHERE / UNKNOWN OUTCOME) with the contract, claim-by-claim handback check, independent second-surface verification, collateral sweep, and a ledger entry.
Connects securely to your tools. The creator never sees your data.
About this skill
Every agent-testing bundle on this marketplace tests code: it diffs two runs, turns a failed trace into a regression fixture, builds a JSONL eval dataset, and gates a dataset's provenance. Every delegation-brief writer stops at the brief — by its own stated limits, it does not dispatch, monitor, or verify the agent. Nobody verifies the other kind of delegation: the real-world task you handed an agent in your own logged-in browser, account, marketplace, storefront, or filing system, where the agent came back and reported "done."
That report is a claim, not a result. The agent may have clicked on a locked or asleep screen and reported the click as effective. It may have edited the wrong variant, the wrong listing, or a draft instead of the live record. It may have done the one authorized change — and three unauthorized ones nobody asked it to check for. Or it may have failed in a way whose outcome is genuinely unknown, where repeating the action to "make sure" would double-charge, double-submit, or double-sell.
This skill is the post-run audit for that delegation. Before dispatch it locks a one-change task contract: the canonical identifier copied never retyped, the single authorized change, the explicitly forbidden changes, and the independent verification route named in advance. After the run it verifies by that independent route — an export diff, a fresh live read-back, a second surface — never by the agent's own screenshots or summary — sweeps for collateral side effects, classifies any failure, and issues one honest verdict. REPORTED is never upgraded to VERIFIED without that independent read.
Built from a live delegated-operations practice: account and listing changes executed by a separate agent in the operator's own logged-in browser under anchored, guarded, one-change prompts and verified by independent export diff; remote-screen work where a locked screen (LockApp present) invalidated every reported click; marketplace work where a duplicate listing created a double-sell risk that only a collateral sweep caught; and a standing rule that an unknown outcome is never retried just to obtain a cleaner result.
Audits one delegated task at a time using records, exports, and live surfaces the user can access. It does not monitor an agent live, does not execute the delegated task itself, and cannot verify a task for which no independent read route exists — in that case it caps the verdict at REPORTED and says why.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 days ago
Permissions
No special permissions declared or detected