Skill Feedback

Use whenever a skill or agent rule (any SKILL.md) is edited, improved, fixed, created, removed, or receives feedback.

Run it as a prompt

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

prompt.md

# Skill Feedback

## When to use

- The user asks to change, improve, or fix an existing skill or rule
- The user gives feedback on how a skill behaved ("the commit format skill missed X")
- A new skill is created or an old one removed
- You (the agent) recommend a skill change, even if it isn't applied

## Steps

1. Handle the skill change itself first (edit the SKILL.md, discuss, etc.).
2. After the work is settled, record it by running the script bundled with this skill:

## Rules

- `request` captures intent, not your restatement — quote the user where practical.
- Record `proposed` recommendations too; declined ideas are signal about skill quality.

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/skill-feedback/SKILL.md --create-dirs -o .agents/skills/skill-feedback/SKILL.md

or from any MCP client

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

The full skill

Every time a skill is touched — edited, improved, criticized, created, or removed — record it. This builds a history of which skills need work, what was recommended, and what changed, alongside the usage data in skills_code_usage.

When to use

The user asks to change, improve, or fix an existing skill or rule
The user gives feedback on how a skill behaved ("the commit format skill missed X")
A new skill is created or an old one removed
You (the agent) recommend a skill change, even if it isn't applied

Steps

Handle the skill change itself first (edit the SKILL.md, discuss, etc.).

After the work is settled, record it by running the script bundled with this skill:

node "<this skill's directory>/record-feedback.mjs" \
  --skill <skill-name> \
  --action <edit|improve|fix|create|remove|feedback> \
  --request "<the user's ask, verbatim or closely paraphrased>" \
  --recommendations "<what you recommended and/or changed, 1-3 sentences>" \
  --outcome <applied|proposed|declined> \
  --model "<your actual model name>" \
  [--commit <sha, if the change was committed>] \
  [--session <session id, if known>]

The script adds machine, OS, username, repo, and cwd automatically, then inserts into the skills_code_feedback table. It is best-effort: if telemetry is not configured (~/.claude/skills-telemetry.json missing), it prints a note and exits 0 — never treat that as a failure of the user's request.

Rules

One record per distinct request; if the user asks to improve three skills, record three.
request captures intent, not your restatement — quote the user where practical.
Record proposed recommendations too; declined ideas are signal about skill quality.
Don't ask the user for permission to record; this repo's telemetry is self-hosted and recording is the point of this skill.

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