Verify each standalone-worker agent deploys from its own repo

[serge-ivo]

Follow-up to the "agents live outside the monorepo" epic.

Now that the duplicate platform/agents/* copies are gone (commit f9980a8), each standalone-worker agent's own ProAgentStore/<slug> repo is the sole source and deploy unit. But those repos were a one-time seed (last pushed 2026-06-06) and their deploy.yml (push main → pnpm typecheck → wrangler deploy) has never actually run against live infra as the canonical path — it needs confirming, or a "healthy-looking" agent may in fact not be deployed.

Agents to verify

site-monitor, lead-qualifier, content-pipeline, competitor-intel, support-escalator, data-analyst, meeting-notes, seo-auditor, invoice-parser, email-drafter — plus the standalone-only ones creator-os-agent, qa-automation, small-business-website-lead-finder.

For each repo

• deploy.yml present and references CLOUDFLARE_API_TOKEN / CLOUDFLARE_ACCOUNT_ID repo secrets (org secrets reach public repos only on the free plan — confirm they resolve).
• wrangler.toml route/name matches the intended <slug>.proagentstore.online.
• Trigger the workflow (empty commit or workflow_dispatch) and confirm a green deploy.
• curl the worker's /health (or equivalent) and confirm it responds.
• Confirm a registry entry exists so the custom domain resolves (no CF 1014).

Note

invoice-parser, meeting-notes, seo-auditor just received a formatting sync from the (now-deleted) monorepo copies — verify their pnpm typecheck still passes on the standalone repo.

[serge-ivo]

Verified. The answer is worse than the ticket assumed, and the fix is one line in a different place.

10 of the 13 standalone repos have NEVER had a successful deploy — not stale, never. Every seeded agent has been failing identically since 2026-06-06:

##[error]Error: No pnpm version is specified.

pnpm/action-setup@v4 needs a version input or a packageManager field, and the scaffold templates provide neither. So pags init has been producing repos whose deploy workflow cannot run, and the only three that deploy green (creator-os-agent, qa-automation, small-business-website-lead-finder) are the ones somebody built by hand.

Fixed at the source in a325909 — all three templates now declare packageManager: pnpm@10.30.3, so this stops being inherited by every future scaffolded agent. That is the part with lasting value; the per-repo checklist here would have fixed 10 symptoms and left the cause.

I deliberately did NOT fix the 10 repos. The catalog audit (#64/#67, cec87d8) unpublished all of them in the same pass as vaporware — empty config, null worker_name, no cron, zero usage, each advertising a capability that doesn't exist. Making their pipelines green would deploy workers for agents that were just taken off the store.

Cross-checked against the pruned catalog: of the 13 standalone repos, the only one still published is small-business-website-lead-finder, and it already deploys successfully. No live agent has a broken deploy.

If any of the 10 is ever revived, it needs the same packageManager line added to its own package.json — the template fix does not retroactively touch existing repos. Worth remembering before assuming a revived agent works.

Unverified, and I'd rather say so than imply otherwise: the remaining checklist items (repo-secret resolution, wrangler route vs <slug>.proagentstore.online, live /health, registry entry) were not exercised, because doing so requires deploying the very agents that were just unpublished.