naveen@failloud:~/playground$ failloud check agent_output.json

FailLoud

FailLoud catches agents claiming “done” when the work is placeholder, unbacked, stale, miscounted, or missing pieces. Paste an agent’s JSON output below and watch it get checked — deterministically, line by line. No model grades the homework.

Same checks, in your browser. This playground runs the five real FailLoud checks client-side — nothing you paste is uploaded anywhere. The CLI + MCP server is the production tool.
FL001
placeholder_shipped
Done, but the payload is empty or still holds lorem ipsum, {{template}} braces, “coming soon”.
FL002
unbacked_claim
“Sent / posted / published” with no message id, URL, receipt, or artifact to back it.
FL003
stale_state
Work based on state older than allowed (24h by default), untimestamped, or version-skewed.
FL004
count_mismatch
The report says 12, the itemized record lists 9. The list is the only part that was ever true.
FL005
missing_components
Fields present but silently empty, zero-length segments, dead stretches in the deliverable.
failing_output.json— the exact examples shipped in the repo

Or paste / drop any JSON an agent produced. The envelope that verifies best: status, generated_at, claims[] with evidence, counts + items, state with context_updated_at, output, segments — every field optional, each one present is one more thing checked instead of trusted.

failure report— errors fail the run, warnings don’t
VERDICT
—
Running the failing example…

Liked what it caught? Get the code.

The full FailLoud — the command-line tool and the MCP server an agent can call inside the loop, before acting on a sub-agent’s output — lives on GitHub, MIT-licensed, zero-dependency core, 68 tests. Star it, clone it, wire it into CI: exit code 0 is a pass, 1 is a failed output, 2 is a tool error.

$ pip install failloud# or pip install . from a clone · add [mcp] for the server
Get the code on GitHub ↗