skill: post-deploy workflow completion — trigger required runs after apify push

[DaveHanns]

Problem

Under agent workflows that require both a deploy AND subsequent remote runs (e.g. seeding cross-run KV state, scheduled execution, or a task with a two-step "push then invoke" shape), agents commonly terminate after apify push succeeds. The Actor is deployed but never invoked — downstream steps that require run artifacts fail because no run exists.

Observed in multiple runs (sc05/all + sc07/both — 2026-06-23 sweep): high-quality source code is pushed, then the agent declares completion. Downstream checks find "0 remote runs" against the newly-deployed Actor.

Suggested addition to the skill

Add a "Post-deploy workflow completion" section to the skill:

### After `apify push` — trigger the runs the task actually needs

`apify push` deploys the Actor but does not run it. If your task requires cross-run
KV state, a scheduled invocation, or any observable output from the deployed Actor,
you must explicitly trigger the run(s) after pushing.

Preferred invocation surfaces, in order:
1. `apify call <actor-name>` — CLI, easiest, runs to completion, returns dataset.
2. MCP `call-actor` tool — for MCP-based agents.
3. `curl -X POST https://api.apify.com/v2/acts/<user~name>/runs?token=$APIFY_TOKEN`
   — REST fallback for api-only stacks.
4. `curl <standby-url>` — for Standby-mode Actors (do NOT use `apify call` — it
   opens a long-lived process that stays alive until `defaultTimeoutSecs` expires,
   silently burning cost; see F50).

Rule of thumb: "the task is complete when the deployed Actor has produced the
artifacts (dataset items, KV records, log lines) the task described" — not when
`apify push` returned.

Related

• Companion runner-side change in the eval framework: when the "remote-run count" checkpoint fails, dump the agent's final 5 tool calls into the report so we can distinguish "agent thought push completed the task" from "agent intended to invoke but tool call errored".

Impact

High (agent-gap). Task completion depends on the agent understanding push ≠ invoke.

---

Surfaced during an evaluation of Apify surfaces for agent-driven Actor development.