One repeatable job
Each skill is scoped around a concrete workflow rather than a vague promise that an agent can handle everything.
// maker and method
Focused tools. Inspectable evidence. Explicit limits.
I’m H.J. Westerfield, the person behind JustHandled Labs. My background is in communications, editing, project coordination, and practical AI systems. I build small agent skills and evidence workflows that do one named job without hiding the judgment calls.
How the work is built
The common pattern is deliberately narrow: define the task and its boundary, make the inputs and outputs inspectable, rehearse the workflow on safe fixtures, and report what passed, failed, remains ambiguous, or cannot be assessed.
Safe defaults
Audits and scanners are read-only by default. Workflows that can change files preview the proposed change before writing.
Evidence over confidence
Fixtures, validation checks, checksums, and receipts make results easier to inspect and reproduce.
Limits stay visible
A passing preflight is evidence about the supplied inputs, not a certification, legal opinion, or guarantee about an external system.
What counts as a result
Publication, indexing, views, testing, and deployment are useful operating facts, but they are not automatically demand or revenue. JustHandled Labs keeps those categories separate from buyer questions, qualified use, paid transactions, and demonstrated workflow value.
That distinction also shapes the products: conflicting or incomplete evidence stays marked for review instead of being promoted into certainty.
Where to start
Use the catalog for a focused self-serve task, the guides for the underlying method, or the Launch Evidence Review when a bounded AI-built app needs a human-reviewed release ledger.