The `feedback` command — and its canonical `--source agent` example —
appeared in zero skill files. Skills load by task relevance, so agents
had no path to discover it and the `--source agent` reporting channel
stayed dormant. Adding the note to bulk-operations (the always-loaded
foundation skill) means every agent sees it regardless of task, without
duplicating the note across all 15 skills.
Framed as an open "Submit Feedback" section — bugs, things you wish
you'd known, user frustration, unmet requests, and wished-for
capabilities — rather than bug-reporting only, so the channel captures
the full range of agent experience, not just defect reports.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The removed block prescribed a UI round-trip to discover action shapes
and asserted there is no endpoint to enumerate valid action types. That
guidance is about to be superseded: once HubSpot exposes endpoints that
list valid actions, this section will be rewritten to point at them.
Removing it now keeps the skill from shipping soon-to-be-stale workarounds
and a claim ("no endpoint exists") that won't hold. SKILL.md's parallel
sentence + its now-dangling cross-reference to that note are removed too.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The skill told agents to author flows by `get`-ing an existing workflow
and copying its real action objects — but that advice collapses on a
brand-new portal with no workflows to copy from. Document the honest
fallback: build the step(s) once in the HubSpot UI, then `get` that flow
to harvest the exact actionTypeId/actionTypeVersion/fields.
Also note why there's no shortcut: no CLI command or public endpoint
lists the built-in action types. The automation platform keeps an
action-type catalog internally (with field schemas), but the `hubspot`
CLI only proxies flow CRUD and doesn't surface it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`create --dry-run` is a no-op echo: it makes no API call and returns
ok:true even for a body that is not a valid create (e.g. {"nope":true},
or one missing type/flowType/objectTypeId/actions). A green dry-run
therefore only proves the input is well-formed JSON, not that the create
would succeed. Without this note an agent would trust the dry-run as
validation and ship an invalid body — and the trap is easy to miss
because `update --dry-run` does reject bodies missing required fields.
Gives agents a grounded skill for authoring and managing HubSpot
workflows (v4 Flows). The JSON reference and worked examples are grounded
against the real AutomationFlowApi shape rather than guessed: branches CAN
converge (via a shared connection.nextActionId), and the enrollmentCriteria
DATASET discriminator is the on-the-wire value (not DATASET_BASED).
Also disambiguates the `hubspot` agent CLI from the `hs` CLI in the README,
and clarifies when to reach for bulk-operations vs workflow-automation.
Matt Rollender