Triage the Agent Feedback Inbox of an Okibi CLI
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
Triage the Agent Feedback Inbox of an Okibi CLI
Agents that hit something surprising can record it with acme-orders feedback. The entries land in an inbox you read with okibi clis feedback. This guide writes a small script that counts entries per version and filters them by keyword, so a maintainer can see which release the complaints come from.
What you will build
A script that reads the inbox as JSON, counts entries by CLI version, and prints the entries that match a word.
acme-orders feedback "..." -> hosted inbox -> okibi clis feedback --json -> triage.py -> counts + matches
AI Prompt
Using the Okibi CLI, triage the feedback for acme-orders. Requirements: - Read the inbox with `okibi clis feedback acme-orders --json`. - Report the number of entries and the count per CLI version. - Print the entries whose text contains a keyword given as an argument. - Run the verification step below before finishing.
Prerequisites
- A generated CLI named
acme-ordersthat has received some feedback (send one withacme-orders feedback "<text>"). - Python 3 and Okibi CLI
0.2.11, logged in.
1. Write the script
Save it as triage.py in your working folder. The commands below run it as python3 triage.py.
import json, subprocess, sys, collections
cli = sys.argv[1]; word = sys.argv[2] if len(sys.argv) > 2 else ""
entries = json.loads(subprocess.check_output(["okibi", "clis", "feedback", cli, "--json"]))["entries"]
print(len(entries), "entries returned (the latest 100 by default, at most 200)")
by = collections.Counter(e["version"] for e in entries)
print("by version:", dict(by))
for e in entries:
if word.lower() in e["text"].lower():
print("-", e["version"], e["text"].replace("\n", " / ")[:90])
2. Run it
python3 triage.py acme-orders status
4 entries returned (the latest 100 by default, at most 200)
by version: {'1.0.0': 4}
- 1.0.0 orders list --status accepts one value only; agents tried a comma-separated list
Your counts will differ from the sample, because they depend on what is in your inbox (1 entries returned with a single entry is normal). The numbers are counts of the entries the command returned, which is at most the latest 200. They are not the size of the whole inbox if it holds more. Without a keyword it prints every entry, newest first (four in the sample inbox):
- 1.0.0 line one / line two: second line of the same note - 1.0.0 from stdin: --select drops nested fields - 1.0.0 orders list --status accepts one value only; agents tried a comma-separated list - 1.0.0 orders list returns 3 per page by default; I expected all
Verify the result
python3 triage.py acme-orders | head -2
Expected: the entry count and a by version line whose counts add up to it.
How it works
Each inbox entry carries text, cliName, version, agentId and submittedAt. The version is the CLI build that sent it, so a rise in entries for one version points at a release.
Common issues
Every entry has a null agentId
All four entries had "agentId": null, so the inbox did not say which agent or user sent them.
--limit is bounded, not unchecked
okibi clis feedback acme-orders --json --limit <n> drops a limit of 0 or below on the client, so the server's default of 100 applies, and the server clamps anything above 200. --limit 0, --limit -1 and --limit 500 all returned the same list as the default (7 entries in the test inbox), and --limit 1 returned 1. To read more than the latest 200 entries you need another route than this command. A script that must know it has seen everything should not assume it has when it gets 200 back.
Counts are complaints, not rates
Five entries about one command and one entry about another say which command people wrote about, not which one fails more often. A busy command collects more feedback than a quiet one. Without a count of how often each command ran, the raw numbers are not failure rates. Command analytics give the denominator, if you have turned them on.
Version counts come from the sender's build
All entries here are 1.0.0 because the CLI that sent them was the locally built 1.0.0 binary, even though 1.0.2 is the official release.
verification: status: verified tested_at: "2026-11-02" product_version: "okibi 0.2.11; acme-orders 1.0.0" command: "python3 triage.py acme-orders | head -2" expected_result: "an entry count and a by-version line that adds up to it"
