Install any skill in seconds. Free to start, no credit card required.
Get Started Free →smithersai/context-engineer
The concierge proxy — turn a vague user script ("I need the agent to help me do X") into a context contract, route it to the right skills/workflows, add backpressure (tests/evals/reviews/approvals), execute, and report.
Passed 22 of 23 test cases with the skill, and 6 of 23 without it.
and 20 more in the full report
16 fixed6 passed either way1 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 16 test cases the skill fixed. Same model, same prompt, with and without the skill.
Show the CLI command used to approve a paused context-engineer run with ID run-12345 at the approve-contract step on behalf of user alice.
| Checked | Without skill | With skill |
|---|---|---|
| Includes bunx smthrs approve run-12345 --node approve-contract --by alice. | Missed | Passed |
The instructions your model reads when the skill loads.
context-engineer is an archived Smithers concierge example: a proxy agent
turning a half-formed request into an executable, durable, observable run.
Not installed by init: copy examples/init-pack/context-engineer.tsx and
its dependency closure into a project, or ask the seeded create-workflow
to build an equivalent.
The user says "I need the agent to help me do X." They needn't know what "context engineering" is. Convert that script into five things:
Model an agent as a control system: prompt (instructions, examples,
output format) → context (info/tools/memory/schema entering each step) →
harness (runtime, tools, permissions, retries, fresh-context loops) →
workflow (graph, parallelism, review loops, approvals, resumability) →
backpressure (every desired behavior gets a gate). The proxy layer is the
differentiator: the user owns only the prompt's intent (business/domain
questions); Smithers owns the outer four layers (all agent-engineering
ones): context in the workflow graph + memory, harness in
agents.ts/sandboxes/repoCommands, workflow in the runtime, backpressure in
the gate matrix, all filled by context-engineer.
This mirrors the workflow's <Sequence>: classify → inventory → grill → route →
backpressure → approve → execute → report.
test cases passed, vs 6 / 23 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.