Skip to content

When should platform skills (vercel, stripe, neon, etc.) be formally consulted within the CE workflow loop? #810

@digitalcostas

Description

@digitalcostas

When should platform skills (vercel, stripe, neon, etc.) be formally consulted within the CE workflow loop?

Background

The CE spine (/ce-brainstorm/ce-plan/ce-work/ce-code-review/ce-compound) doesn't currently specify when platform-specific skills should be formally invoked, or whether using ce-best-practices-researcher to hit official docs directly is a sufficient substitute.

I ran into this concretely while brainstorming a Vercel Cron feature. During the brainstorm, ce-best-practices-researcher fetched Vercel's official cron docs and returned accurate factual coverage — frequency limits, no-retry behavior, maxDuration, etc. But vercel:vercel-functions (the curated platform skill) was never explicitly invoked. The information came in through research; the skill's opinionated patterns and conventions did not.

The question

Is there an intended trigger rule for when platform skills should be formally consulted in the CE loop? A few sub-questions:

  1. Which phase owns it? Is formal platform skill invocation a /ce-plan responsibility (before implementation decisions are locked), a /ce-work responsibility (during coding), or both?

  2. Does ce-best-practices-researcher satisfy the same intent? If the researcher fetches the same official docs the skill would consult, is that equivalent — or does the skill add curated guidance, project conventions, or sequenced setup steps that a raw fetch misses?

  3. Should brainstorms touch platform skills at all? For feasibility grounding (e.g. "can Vercel Crons run every minute on Pro?"), invoking a platform skill during brainstorming seems useful. But it also feels like it might be pulling implementation-layer concerns in too early.

  4. Should there be a named trigger? Something like: "when a plan touches a platform domain (Vercel, Stripe, Neon), explicitly invoke the relevant skill before drafting implementation steps" — added to workflow documentation or AGENTS.md.

Why it matters

Without a clear rule, platform skill coverage is ad hoc — dependent on whether the model happens to recognise the domain in time. For infrastructure features where the wrong architectural choice gets baked into a plan before any code runs, missing the skill at planning time is more costly than missing it during coding.

What I'm not asking

Not asking for platform skills to be auto-invoked on every task — that would add noise. Looking for a lightweight trigger rule that makes coverage deliberate rather than accidental.

Curious whether others have run into this and what patterns you've settled on.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions