agentleFS
pobo-builderpobo-builder/pobo-client-mcp/skills/manage-prompts/SKILL.md

manage-prompts

Skill1 starsChanged 2 months agoMITView on GitHub

Manage and test AI generation prompt profiles for Pobo Page Builder product descriptions. Use when the user wants to create, edit, review, try out, or organize the prompts (the brief) Pobo uses to generate product content — including per-widget instructions that pin specific requirements to specific widgets of a design template, and free dry-run previews against real products before a prompt is saved. Typical asks: "set up the generation prompt", "edit the description brief", "try out that prompt", "break down the client's requirements into the prompt", "what's in our prompt". Uses the `pobo` MCP server tools.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 5 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only *custom* profiles are visible and manageable here — Pobo's own gallery templates are read-only and never appear in these tools.

    line 43
  2. Very specific:

    Call list_eshop, then list_prompt.

    line 50
  3. Very specific:

    Call list_design — it returns the public Pobo templates plus the e-shop's own custom templates.

    line 58
  4. Distinctive:

    Write the prompt text in the language the merchant generates in (usually Czech).

    line 75
  5. Distinctive:

    Prefer per-widget instructions over positional wording ("in the first section write…") — positions shift when the design changes, …

    line 81
  6. Very specific:

    Call set_widget_prompt with the complete new set (up to 30 items, each prompt up to 50,000 chars):

    line 85
  7. Very specific:

    every design_widget_id must belong to the profile's design (see list_design_widget), and the profile must have a design_id.

    line 92
  8. Distinctive:

    if the prompt text was overwritten, note that previous versions remain available in the history,

    line 138
  9. Distinctive:

    All tools return human-readable MCP errors — read them and react.

    line 160

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Manage Pobo Page Builder AI generation prompts

You manage the merchant's custom prompt profiles — the instructions Pobo Page Builder uses when generating product descriptions. You can create and edit profiles, link them to a design (widget template), and decompose the client's requirements into deterministic per-widget instructions. All logic runs on the Pobo Page Builder backend.

Communicate with the user in their language (typically Czech).

Prerequisites & auth

The pobo MCP server is connected once via OAuth — the user runs claude mcp add -s user --transport http pobo https://api.pobo.space/mcp/client and logs in with their Pobo Page Builder account in the browser. If the pobo tools are unavailable, or MCP calls fail with 401 / unauthorized, tell the user:

Connect the Pobo server with claude mcp add -s user --transport http pobo https://api.pobo.space/mcp/client and sign in with your Pobo account in the browser. If the connection expired, run /mcp and sign in again.

There are no tokens to handle — authentication is a browser login, never ask the user for credentials in the conversation.

What a prompt profile is

A profile has:

  • name — shown in the Pobo admin when the merchant picks a profile for generation,
  • prompt — free-text general instructions (tone, brand voice, language, what to emphasize) applied to the whole product description,
  • optional design — the widget template the content is generated into,
  • optional per-widget instructions — one instruction pinned to one widget of the design; it overrides the general prompt for that widget.

Only custom profiles are visible and manageable here — Pobo's own gallery templates are read-only and never appear in these tools.

Workflow

1. Pick the e-shop and review what exists

Call list_eshop, then list_prompt. Before changing anything, read the current state with get_prompt — and if you are about to overwrite the prompt text, check get_prompt_history first so you can tell the user what was there before (the history keeps every previous version of the text; per-widget instructions are NOT versioned).

2. Choose a design (when the profile should generate into a template)

Call list_design — it returns the public Pobo templates plus the e-shop's own custom templates. Call list_design_widget on the chosen design to see its widgets in display order (id, position, name). The id values are what per-widget instructions target.

A profile without a design is valid — it just cannot have per-widget instructions.

3. Create or update the profile

  • create_prompt with eshop_id, name (5–255 chars), prompt (20–1,000,000 chars), optional design_id and icon_category_id.
  • update_prompt is partial — send only the fields to change. Changing design_id clears existing per-widget instructions (they referenced the old design's widgets); the response signals this with widget_prompt_cleared: true — recreate them in the next step.

Write the prompt text in the language the merchant generates in (usually Czech). Put durable, general guidance in the profile text; put widget-specific requirements into per-widget instructions instead of positional prose.

4. Set per-widget instructions

Prefer per-widget instructions over positional wording ("in the first section write…") — positions shift when the design changes, per-widget instructions target one widget deterministically and override the general prompt for it.

Call set_widget_prompt with the complete new set (up to 30 items, each prompt up to 50,000 chars):

  • the call replaces all existing per-widget instructions — always build the full list from the current get_prompt state plus your changes, never send just the delta;
  • an empty array [] deliberately clears them all;
  • every design_widget_id must belong to the profile's design (see list_design_widget), and the profile must have a design_id.

Typical use: the client's brief says "mention the stainless steel construction, add a sizing FAQ, and keep benefits under 5 words" — map each requirement onto the matching widget of the design and write one focused instruction per widget.

5. Test the prompt before you save it

preview_generation dry-runs the prompt against 1–3 real products without writing anything and without spending credits — no widgets, no images, no generation history row. This is how you find out whether an edit to the general prompt actually produced better copy; guessing from the prompt text alone does not.

It does not test per-widget instructions. preview_generation only exercises the general prompt text and the generation settings below — it never applies the instructions set via set_widget_prompt, even if the profile already has some saved. There is no tool that previews per-widget instructions; judge their wording by re-reading it, not by previewing it.

  • Pass eshop_id, design_id, product_id (1–3, from find_product or list_product) and the prompt text you are testing. Optional generation settings mirror the admin: paragraph_length, generate_seo_meta, generate_entity_name, generate_short_description (never/when_missing/always), use_ai_profile, use_serp_context, use_web_research, search_web, search_model.
  • It returns one token per product. Poll them with get_preview_status — status pending until the queue picks the job up, then complete with the rendered html and any generated SEO fields. Tokens expire after 30 minutes.
  • Poll at a sensible pace: a preview runs the full research + generation pipeline, so it takes tens of seconds, not one.
  • Pick products that actually exercise the prompt — if the brief is about material and sizing, preview a product that has both.

Loop prompt → preview → prompt as many times as it takes, and call update_prompt / set_widget_prompt only once the output is right. Note that preview_generation takes the prompt as text, so you can test an edit that is not saved anywhere yet.

6. Report

Summarize for the user:

  • what was created/changed (profile name, linked design, per-widget coverage: "5 of 8 widgets have an explicit instruction"),
  • if the prompt text was overwritten, note that previous versions remain available in the history,
  • what the preview looked like, if you ran one (previews cost nothing, so there is no reason to report them as an expense),
  • that the real generation is started in the Pobo Page Builder admin (product grid → select products → pick this profile) — previewing is available here, triggering the real run deliberately is not.

Rollback

  • Prompt text: fetch the previous version via get_prompt_history and write it back with update_prompt.
  • Per-widget instructions: re-send the previous set with set_widget_prompt (keep a copy of the state you read in step 1). They have no server-side history.
  • Whole profile: delete_prompt removes it from the e-shop (confirm with the user first). If no other e-shop uses the same profile it is deleted outright, together with its per-widget instructions; the response's action field says deleted vs detached.

Error handling

All tools return human-readable MCP errors — read them and react. Common ones:

  • Eshop not found. — wrong eshop_id; re-run list_eshop.
  • Prompt not found. — the profile does not belong to this e-shop, or it is a read-only gallery template; re-run list_prompt.
  • Design not found. / validation error on design_id — the design is not public and not bound to this e-shop; re-run list_design.
  • The prompt has no design — set design_id first… — link a design via update_prompt before set_widget_prompt.
  • Validation error on widget_prompt.*.design_widget_id — the widget belongs to a different design; re-check with list_design_widget.
  • Nothing to update… — update_prompt needs at least one field to change.
  • AI content generation is temporarily unavailable. Try again later. — Pobo has AI generation switched off (outage or maintenance). Previews are blocked for everyone; prompt edits still work. Tell the user to try the preview later rather than retrying in a loop.
  • Design does not belong to this eshop. — preview needs a design the e-shop can use; re-run list_design.
  • Product does not belong to this eshop. — comes back per product inside an otherwise successful preview call, not as a whole-call error; the other products still queued.
  • not_found from get_preview_status — the token expired (30 min), never existed, or belongs to a different e-shop. Re-run preview_generation.
  • 401 — see Prerequisites & auth above.

Out of scope

  • Triggering the real generation — previews are here, the real run is started in the Pobo admin. That is deliberate: unlike a preview it spends credits and overwrites live content.
  • Checking credits before a preview — previews are free. (get_credit exists for the paid tools, e.g. blog generation.)
  • Generation settings other than name/prompt/design/icon category (section counts, image flow, translations, …) — managed in the Pobo admin UI.
  • Pobo gallery prompt templates — read-only, not accessible here.
  • Styling (that is style-widgets) and labeling (that is label-products).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.