A documented workflow nothing could execute
A guard blocked the one step a documented runbook required. The fix was not a weaker guard, it was building the sanctioned command it was asking for.
guardrails
30 posts
Running an AI agent in CI without it spending your budget
An agent in CI cannot be asked a question. Budget caps that actually throw, local reproduction before pushing, and the gate that degraded instead of failing.
What should an AI agent never be allowed to do?
A permissions model derived from incidents rather than imagination: three tiers, human-only bypasses, and capabilities removed from reach entirely.
The most useful thing my agent does is refuse
Refusals are only useful if they name the remedy. Four that fired during one session, what each redirected me to, and the one that fired on nothing at all.
The agent that writes the check that blocks it
When the same agent writes the fix and the gate, the gate agrees by construction. RED before GREEN is what makes a machine-written check mean anything.
Where to put the human
Gate placement for autonomous agents. Reversibility is the criterion, not confidence, and the consent surface matters as much as the question.
A gate that fixes what it finds
An auto-healing loop must fix the source, never the symptom, and must never edit the thing that found the problem. Confidence floors and bounded rounds.
Six guardrails that stop my AI agent mid-command
Six PreToolUse hooks that refuse an agent's command before it runs. What each one caught, and why a guard the agent can talk its way past is not a guard.
A verification step that can pass while the bug is present is not a verification step
Static checks read what was written; the defect lives in what was drawn. Four artifacts that passed every gate this session and failed the moment someone looked.
Teaching an agent when to stop
Enumerated halt classes make stopping checkable; a mood does not. The forbidden reasons that stall a drive, and the undeclared halt I made anyway.
The agent said 'fixed'. The list said otherwise
A delete returned deleted:true while the item was still there. Four failures in one day shared a shape: the action reported on itself, and nothing read the state back.
A test that cannot fail is not a test. Mine passed twice.
A RED test fixture that passed because a comment explaining the omission contained the omitted string. The failure mode is generic: any test whose input mentions what it removes.