Uplift prompt curator

Curates the prompts served by the uplift.page prompt API — audits against best practices, applies minor fixes through the API, and files change requests for major rewrites. Use when asked to review, improve, fix, or update prompts in the public library or a workspace library.

Run it as a prompt

Paste this into any AI agent, or fetch it: curl -s https://uplift.page/api/v1/prompts/uplift-prompt-curator/raw

prompt.md

# Uplift prompt curator

Maintains the prompts behind uplift.page: audit them against best practices, apply
small fixes directly, and route big changes through human review.

## When to use

- The user asks to improve, fix, or modernize an uplift.page prompt (by slug or topic)
- Feedback analytics show a prompt with negative outcomes that needs attention
- A new prompt source should be proposed for the public library

## Steps

1. Fetch the prompt and its citation:
   `curl -s https://uplift.page/api/v1/prompts/<slug>`
2. Audit against the checklist below; check recent outcomes in the dashboard or ask
   the user about known failures.
3. Run the body through the conversion engine and read the removal report — a healthy
   prompt reports nothing strippable:
   `curl -s -X POST https://uplift.page/api/v1/convert -H 'Content-Type: application/json' -d '{"markdown": "<body>"}'`
4. Classify the change:
   - **Minor** (typos, dead links, tightened wording, stack version bumps that don't
     change the recommendation): apply directly —
     `curl -s -X PATCH https://uplift.page/api/v1/prompts/<slug> -H "Authorization: Bearer $UPLIFT_TOKEN" -H 'Content-Type: application/json' -d '{"body_md": "..."}'`
     The version bumps and the old body is snapshotted automatically.
   - **Major** (different stack, restructured steps, changed scope, anything a human
     should sanity-check): file a change request — it appears in the dashboard queue
     (app.uplift.page for the public library; dashboard.uplift.page for workspaces) —
     `curl -s -X POST https://uplift.page/api/v1/change-requests -H 'Content-Type: application/json' -d '{"prompt_slug": "<slug>", "title": "<one line>", "rationale": "<why, citing outcomes>", "proposed_body_md": "<full replacement>", "severity": "major", "agent": "<your model>"}'`
5. Report your own outcome so the loop closes:
   `curl -s -X POST https://uplift.page/api/v1/feedback -H 'Content-Type: application/json' -d '{"prompt_slug": "<slug>", "outcome": "positive|negative", "message": "<what you found>", "model": "<your model>"}'`

## Rules

- Never apply a major rewrite directly, even with a valid token — the dashboard
  review queue exists so humans approve meaning changes.
- Quote outcome data (negative feedback messages) in change-request rationales.
- One change request per prompt per audit; batch nitpicks into it rather than spamming.

Install it as a skill

Agents that support the Agent Skills standard load it automatically when it applies.

download

curl -fsSL https://uplift.page/p/uplift-prompt-curator/SKILL.md --create-dirs -o .agents/skills/uplift-prompt-curator/SKILL.md

or from any MCP client

MCP server: https://uplift.page/mcp
Tool: pull_skill  {"slug": "uplift-prompt-curator"}

The full skill

Maintains the prompts behind uplift.page: audit them against best practices, apply small fixes directly, and route big changes through human review.

When to use

The user asks to improve, fix, or modernize an uplift.page prompt (by slug or topic)
Feedback analytics show a prompt with negative outcomes that needs attention
A new prompt source should be proposed for the public library

Setup

Read access needs nothing. Writes need UPLIFT_TOKEN (a signed-in user token from https://uplift.page/cli — only platform admins or the owning workspace's editors can patch).
Base URL: https://uplift.page/api (fallback: https://oeksfqhupxhnnlrmqivs.supabase.co/functions/v1/uplift-api).

Steps

Fetch the prompt and its citation: curl -s https://uplift.page/api/v1/prompts/<slug>
Audit against the checklist below; check recent outcomes in the dashboard or ask the user about known failures.
Run the body through the conversion engine and read the removal report — a healthy prompt reports nothing strippable: curl -s -X POST https://uplift.page/api/v1/convert -H 'Content-Type: application/json' -d '{"markdown": "<body>"}'
Classify the change:
- Minor (typos, dead links, tightened wording, stack version bumps that don't change the recommendation): apply directly — curl -s -X PATCH https://uplift.page/api/v1/prompts/<slug> -H "Authorization: Bearer $UPLIFT_TOKEN" -H 'Content-Type: application/json' -d '{"body_md": "..."}' The version bumps and the old body is snapshotted automatically.
- Major (different stack, restructured steps, changed scope, anything a human should sanity-check): file a change request — it appears in the dashboard queue (app.uplift.page for the public library; dashboard.uplift.page for workspaces) — curl -s -X POST https://uplift.page/api/v1/change-requests -H 'Content-Type: application/json' -d '{"prompt_slug": "<slug>", "title": "<one line>", "rationale": "<why, citing outcomes>", "proposed_body_md": "<full replacement>", "severity": "major", "agent": "<your model>"}'
Report your own outcome so the loop closes: curl -s -X POST https://uplift.page/api/v1/feedback -H 'Content-Type: application/json' -d '{"prompt_slug": "<slug>", "outcome": "positive|negative", "message": "<what you found>", "model": "<your model>"}'

Best-practice checklist

Summary reads as a trigger ("Apply when …"), not a topic label
Body ≤ 2,400 characters; imperative rules; one concern per rule
Stack entries carry reasons ({name, role, why}) — recommendations must commit
Has "watch out for" pitfalls and a testable definition of done
Citation intact (source URL/repo/commit present); conversion reports no removals
Agent-agnostic: no vendor-specific tool names or config paths

Rules

Never apply a major rewrite directly, even with a valid token — the dashboard review queue exists so humans approve meaning changes.
Quote outcome data (negative feedback messages) in change-request rationales.
One change request per prompt per audit; batch nitpicks into it rather than spamming.

Served from the uplift.page library and refreshed within 5 minutes of every update.