Skip to content

Two workflows that chain primitives, composed from the skills' own commands - #19

Draft
pablomogamon wants to merge 1 commit into
ZenRows:mainfrom
pablomogamon:act/multi-primitive-workflows
Draft

Two workflows that chain primitives, composed from the skills' own commands#19
pablomogamon wants to merge 1 commit into
ZenRows:mainfrom
pablomogamon:act/multi-primitive-workflows

Conversation

@pablomogamon

Copy link
Copy Markdown

Overview's "workflows" were single-tool example prompts typed by hand in the app repo — not workflows, and not faithful to anything the toolkit ships. This adds the real thing on the toolkit side, where it belongs:

  • price-monitoring (Fetch → Extract → Batch): prove the page (--wait-for), prove the structure once (--css … --validate), batch estimate before spending, batch create --wait + results, retry-failed on a schedule.
  • gated-page-extraction (Fetch → Browser → Extract): the escalation-only rule end to end — the browser enters only when the cheap step failed with evidence, browser run script.json preferred because scripted sessions close themselves.

Fidelity rule: every command line is lifted verbatim from the skills (protected-fetch, extract, batch-jobs, interact-browser), so a workflow cannot promise a flag the CLI does not have. Both registered in registry/workflows.json alongside competitor-intelligence, same schema.

Why now: the app's Overview stops hand-typing this content — a sync script there (scripts/sync-agent-catalog.mjs, companion app PR) vendors this registry + the WORKFLOW.md steps and renders them verbatim, so what the product shows and what zenrows workflow explain says come from one source.

Toolkit suite: 184/184 green.

🤖 Generated with Claude Code

…mmands

price-monitoring (fetch -> extract -> batch): prove the page, prove the
structure, estimate before spending, run managed, retry-failed on a
schedule. gated-page-extraction (fetch -> browser -> extract): the
escalation-only rule end to end - the browser enters only when the cheap
step failed with evidence, scripted sessions preferred because they
close themselves.

Every command line is lifted from the skills (protected-fetch, extract,
batch-jobs, interact-browser), so the workflows cannot promise a flag
the CLI does not have. Registered in registry/workflows.json; the app's
Overview vendors this registry (scripts/sync-agent-catalog.mjs there),
so these cards render in the product from this same source.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@pablomogamon
pablomogamon requested a review from AnderRV September 5, 2026 18:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant