Uplift Evaluate

Evaluates a skill distributed on uplift.page using live platform data — conversion cleanliness, size, citation, freshness, workspace usage, and outcome feedback. Use when asked whether a skill is healthy, worth keeping, or needs work, when auditing a workspace library, or before promoting a skill to the public library.

Run it as a prompt

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

prompt.md

# Uplift Evaluate

A skill's quality is measurable: does it convert clean, fit the budget, carry a citation,
get used, and produce positive outcomes? The platform holds all of that data; this skill
turns it into a verdict and an action, not just a report.

## Steps

1. Call `evaluate_skill {slug}` on the uplift MCP server (workspace key required — see
uplift-connect). It returns a score, a pass/fail checklist, the conversion report,
feedback tallies with recent negative messages, and 30-day workspace usage.
2. Turn each failed check into its action:
- **trigger-style summary** → rewrite the description as "Use when …" so agents load it
- **over budget** → split the skill or cut the weakest section
- **conversion residue** → tracking or vendor-specific text survives in the served
body; clean the source and re-save
- **cited** → add `source: {repo, path, commit, url}` on the next save
- **fresh** → re-read against current practice, then save (a save is the freshness
signal, so don't save without actually reviewing)
- **used this month** → check the trigger description against what the team actually
asks for; unused + stale together makes an archive candidate
- **positive outcomes** → read `recent_negative` messages and fix the instruction
they blame
3. Route the fix by ownership: workspace skill → `save_skill`; public-library skill →
change request quoting the evidence (`POST /v1/change-requests`).
4. Close the loop: after acting on a skill, report your own outcome with `send_feedback`.

## Rules

1. Quote the numbers and the feedback messages in every recommendation — never "seems fine".
2. Evaluate before any public promotion and after any negative-feedback report.
3. Batch an audit: `list_skills` first, evaluate each, then present one ranked table
(healthy / fix / archive) instead of a stream of per-skill verdicts.
4. Without MCP support, degrade gracefully: fetch the prompt over REST, run its body
through `POST /v1/convert`, and judge the static checks locally — say that usage and
feedback data were unavailable.

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-evaluate/SKILL.md --create-dirs -o .agents/skills/uplift-evaluate/SKILL.md

or from any MCP client

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

The full skill

A skill's quality is measurable: does it convert clean, fit the budget, carry a citation, get used, and produce positive outcomes? The platform holds all of that data; this skill turns it into a verdict and an action, not just a report.

Steps

Call evaluate_skill {slug} on the uplift MCP server (workspace key required — see uplift-connect). It returns a score, a pass/fail checklist, the conversion report, feedback tallies with recent negative messages, and 30-day workspace usage.
Turn each failed check into its action:
- trigger-style summary → rewrite the description as "Use when …" so agents load it
- over budget → split the skill or cut the weakest section
- conversion residue → tracking or vendor-specific text survives in the served body; clean the source and re-save
- cited → add source: {repo, path, commit, url} on the next save
- fresh → re-read against current practice, then save (a save is the freshness signal, so don't save without actually reviewing)
- used this month → check the trigger description against what the team actually asks for; unused + stale together makes an archive candidate
- positive outcomes → read recent_negative messages and fix the instruction they blame
Route the fix by ownership: workspace skill → save_skill; public-library skill → change request quoting the evidence (POST /v1/change-requests).
Close the loop: after acting on a skill, report your own outcome with send_feedback.

Judgement calls

Zero usage on a skill younger than ~30 days is noise, not failure — flag it, keep it.
Real outcomes outrank static checks: negative feedback on a checklist-perfect skill still means it fails in practice.
Archive candidates need all three: unused ≥90 days, stale, and no positive feedback.

Rules

Quote the numbers and the feedback messages in every recommendation — never "seems fine".
Evaluate before any public promotion and after any negative-feedback report.
Batch an audit: list_skills first, evaluate each, then present one ranked table (healthy / fix / archive) instead of a stream of per-skill verdicts.
Without MCP support, degrade gracefully: fetch the prompt over REST, run its body through POST /v1/convert, and judge the static checks locally — say that usage and feedback data were unavailable.

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