Inspecting

status

pikopod agent status

Reads /healthz from the running agent and prints it as JSON: requests proxied, upstream errors, recordings written and dropped, observer panics, events emitted, alert delivery health, the number of tracked fingerprints, and every baseline family with its sample count and whether it is warmed up. With a spec watcher armed, a spec_watch block reports each source’s last check and last error. Every upstream with recordings gets a truthfulness block: how much of what the provider sent the linked sandbox reproduces, over the most recent 200 recordings. See Truthfulness. Needs pikopod up. See Control plane for the fields.

report

pikopod agent report

The endpoint inventory from the baselines on disk: what your app actually calls, by method and template, with the status class, sample count, field count, and whether the baseline is frozen or still warming up. Worth reading after about 48 hours. Works offline.

examplepay — endpoint inventory (what your app ACTUALLY calls)
  GET    /charges                                       2xx    12 samples  4 fields  [baseline frozen]
  POST   /charges                                       2xx     6 samples  5 fields  [baseline frozen]

1 provider failures recorded · 1 reproduced locally · 1 committed as scenarios · 0 recurred after being committed

The last line is the outcome the whole loop exists for, computed from the event log and the scenarios directories. It is printed whenever the event log has anything in it, even before any baseline is frozen.

Number	Definition
provider failures recorded	Distinct fingerprints in `<data_dir>/events.ndjson` whose kind is an incident (`upstream_error`, `upstream_unreachable`, `rate_limited`, `client_error`, `webhook_duplicate`, `webhook_out_of_order`) or a `behaviour_divergence`. Drift events are not failures and are not counted.
reproduced locally	Of those, the ones with a pack `incident-<fingerprint>.yaml` under `<data_dir>/scenarios`, which `pikopod reproduce` writes, or committed as below.
committed as scenarios	Of those, the ones with that pack under `./scenarios` in the working tree, the directory `scenario check` reads first and the one you commit.
recurred after being committed	Of the committed ones, the ones whose event was last seen after the committed pack’s file time. A failure that recurs after its regression test was committed is the number you want at zero.

inspect

pikopod agent inspect
pikopod agent inspect --last 20 --format curl
pikopod agent inspect --format har > recordings.har

Prints stored records with tokenized fields visible, so you can confirm what was kept before you trust it. --last sets how many records per upstream (default 5). Formats:

Format	Output
`json`	One pretty-printed record per line, then a footnote describing exactly what was redacted and the one residual: a numeric secret under a field name pikopod does not recognise passes through.
`curl`	One `curl` command per record, with the sanitized headers and body, safe to share.
`har`	One HAR 1.2 document across the selected records.

Everything agent inspect prints is post-sanitizer: tokens and placeholders, never raw traffic. See Redaction.

The event log

pikopod agent incidents --format json --limit 200

The reader for <data_dir>/events.ndjson. See Incidents and pikopod agent incidents.

Getting started

Gate: catch declared changes in CI

Sandbox: rehearse failures

Scenarios

Observe: the agent

Reproduce and fix

Operations

Project

status

report

inspect

The event log

​status

​report

​inspect

​The event log

status

report

inspect

The event log