- Home
- Skills
- Prompt & Skill Engineering
- Skill Validate And Fix
Works with the AI Agents you already use
Skill Validate And Fix
Audits a SKILL.md against the skill template and hands back a findings table, the complete corrected file, and a change list, so a skill that never triggers starts to
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Skill Validate And Fix
Example session with this skill installed
Audit my SKILL.md and fix it - it never triggers.
name: Meeting Notes
description: helps with meetings
meeting notes skill
This skill is very powerful and makes great notes.
Steps
- read the transcript and summarise it and find action items and owners and dates and send them to slack
- make it good
Output
A summary. Rate the summary 1-10 at the end.
- Read your context and instructions
- Compiled the validate and fix
Resolved source: inline text
The reason it never triggers is the frontmatter. name isn't a valid skill identifier, and description: helps with meetings gives the agent nothing to match a request against. The description is the only part of a skill the agent reads before it decides to load it, so a request like "pull the action items out of this call" has nothing to hit. The body has problems too: one step does five things, one step is a side effect with nothing guarding it, and the output asks the skill to grade its own work.
Findings
| Severity | Location | Finding |
|---|---|---|
| BLOCKER | frontmatter name | Meeting Notes has uppercase letters and a space. A skill name must be lowercase kebab-case (meeting-notes), and it should match the skill's folder name. |
| BLOCKER | frontmatter description | helps with meetings doesn't say what the skill produces, what input it needs, or which requests should trigger it. With no trigger phrases, the skill never gets selected. This is the main cause of the bug you reported. |
| MAJOR | body structure | Three of the four required level-2 sections are missing: Purpose, When to Use and Validation. ## Steps and ## Output are non-canonical stand-ins. |
| MAJOR | Steps, bullet 1 | One bullet bundles five separate actions: read, summarise, extract action items, attach owners and dates, send to Slack. A bundled step tends to be done partly, and you can't tell which part failed. |
| MAJOR | Steps, bullet 1 ("send them to slack") | Posting to Slack is an outward side effect. No channel is named, no tool is assumed and no confirmation is asked for. When no Slack tool is available, the likely failure is the agent saying it sent the message when it didn't. |
| MAJOR | Steps, bullet 1 ("owners and dates") | It's a hallucination risk. Nothing says what to do when the transcript names no owner or no date, so the model fills the gaps with plausible names and deadlines. |
| MAJOR | Steps, bullet 2 ("make it good") | It's a vague instruction that nobody can check. It doesn't change behaviour and nobody can verify it. |
| MAJOR | Output ("A summary") | No output shape is defined. The section never says where action items, owners or dates go, so the format changes from run to run. |
| MAJOR | Output ("Rate the summary 1-10") | This is a self-scoring rubric inside the output. The model nearly always gives itself a high number, and the score tells the reader nothing. Quality checks belong in a Validation section as checkable items. |
| MAJOR | Steps (whole section) | There's no handling for a missing transcript. With no transcript supplied, the skill has nothing to tell it to stop and ask, so it may write notes from the surrounding conversation. |
| MINOR | # meeting notes skill | The level-1 title isn't canonical. It should be the skill's name in Title Case, without the word "skill". |
| MINOR | body, line 1 | "This skill is very powerful and makes great notes." is promotional filler with no instruction in it. |
Fixed SKILL.md
---
name: meeting-notes
description: Turns a meeting transcript or raw call notes into a short summary, a decisions list, and an action-item table whose owners and due dates come only from what was said, plus a Slack-ready recap to paste or post on confirmation. Activate when a user says "summarise this meeting", "write up the notes from this call", "pull the action items out of this transcript", "who owns what from this meeting", or pastes or attaches a meeting transcript and asks for notes or next steps.
---
# Meeting Notes
## Purpose
Meeting transcripts are long, and the parts people need - what was decided, who agreed to do what, and by when - get buried in the discussion. This skill reads a supplied transcript or set of raw notes and produces a short summary, a list of decisions, and an action-item table in which every owner and due date can be traced back to the transcript. It never invents an owner or a date, and it never posts anything to Slack without the user naming the channel and confirming.
## When to Use
1. You have a meeting transcript - pasted, attached, or at a file path - and want written notes from it.
2. You need the action items from a call, each with its owner and due date.
3. You want the decisions made in a meeting listed separately from the discussion around them.
4. You want a Slack-ready recap of the meeting to share with your team.
### SCOPE — This skill does NOT handle:
- Recording, transcribing, or joining a live meeting.
- Scheduling follow-up meetings or sending calendar invites.
- Creating tickets or tasks in a project tracker.
- Posting to Slack or any other channel without the user naming the destination and confirming.
## Process
1. **Resolve Input** — Locate the transcript: pasted text, an attached file, or a path the user gave. If no transcript is supplied or the file cannot be read, stop and ask for it instead of writing notes from the surrounding conversation.
2. **Record Meeting Details** — Take the meeting title, date, and participants from the transcript; write "not stated" for any detail the transcript does not give.
3. **Write Summary** — Write one short paragraph covering the meeting's purpose and the main topics, in the order they were discussed.
4. **Extract Decisions** — List each decision the participants agreed on, one per bullet, each closely paraphrasing the line it came from; if there are none, write "No decisions recorded."
5. **Extract Action Items** — For each commitment, record the task, the owner, the due date, and the speaker or timestamp it came from. Take owners and dates only from the transcript, writing "Unassigned" or "No date stated" where it is silent, and convert a relative date such as "next Friday" to a calendar date only when the meeting date is known.
6. **Assemble Notes** — Output a header line with title, date, and participants, then the Summary, then Decisions, then an Action items table with the columns Task, Owner, Due, and Source.
7. **Draft Slack Recap** — Below the notes, output the summary and action items formatted for Slack in a code block the user can paste.
8. **Post Only on Confirmation** — Post the recap only when a Slack tool is available and the user has named the channel and confirmed; otherwise state that the recap is ready to paste and do not claim it was sent.
## Validation
- [ ] The notes contain a header line, Summary, Decisions, and Action items, in that order.
- [ ] Every owner and due date in the action-item table appears in, or is directly derived from, the transcript, and missing values read "Unassigned" or "No date stated".
- [ ] Every action item has a Source entry naming the speaker or timestamp it came from.
- [ ] No message was posted to Slack unless the user named a channel and confirmed.
- [ ] The output contains no self-assigned score or rating.
- [ ] When no transcript was supplied, the reply asks for one and contains no generated notes.
Change summary
[Frontmatter] Renamed Meeting Notes to meeting-notes. Rename the skill's folder to meeting-notes as well so the two match.
[Frontmatter] Rewrote description so it says what goes in (a transcript or raw notes) and what comes out (summary, decisions, action items with traceable owners and dates, Slack recap), and lists the trigger phrases that should fire it. This is the fix for "never triggers".
- [Title] Changed
# meeting notes skillto the canonical# Meeting Notes.
[Purpose] Added the section. It replaces the promotional opening line with the problem the skill solves and the two guarantees it makes: no invented owners or dates, and no posting without confirmation.
[When to Use] Added the section with four numbered use cases and a SCOPE list that rules out transcription, scheduling, ticket creation and unconfirmed posting.
[Process] Replaced the two ## Steps bullets with eight single-action steps. The bundled bullet is split into separate steps for reading, summarising, extracting decisions, extracting action items and drafting the recap.
- [Process] Added not-found handling as step 1: with no transcript, the skill stops and asks for one.
[Process] Owners and dates now come only from the transcript, with fixed fallback wording and a rule for relative dates. This closes the hallucination risk.
[Process] Changed "send them to slack" so the skill drafts the recap and posts it only when a Slack tool exists and you've named and confirmed the channel. It never claims a post it didn't make.
- [Process] Removed "make it good". The concrete output shape in step 6 replaces it.
[Process] Replaced ## Output with a defined layout: header, Summary, Decisions, and an Action items table with Task, Owner, Due and Source columns.
[Validation] Added the section. It replaces the "Rate the summary 1-10" self-score with six checkboxes you can verify against the output.
- No minor findings left unresolved.
Verdict: needs human review. The structure and triggers are fixed, but step 8 changes the behaviour from "always send to Slack" to "draft, then post only on confirmation". Check that this is what you want before deploying, then test it by pasting a transcript and saying "pull the action items out of this transcript".
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
Checks a SKILL.md the way the agent will read it, then repairs it. Most skills that "never trigger" fail in the frontmatter: a name that isn't a valid identifier, or a description too vague to match anything a user says. This finds that and the body problems that make a skill behave badly once it does load.
What it does
It audits structure (valid frontmatter, name and description rules, the canonical title and sections in order) and content: bundled steps, side effects with no guard, steps that ask the model to grade itself, vague triggers, unjustified hard limits, hallucination risks. Each finding is labelled BLOCKER, MAJOR or MINOR. You get a findings table, the complete fixed SKILL.md in one block, and a change summary tagged by section, ending with a verdict and a note on any behaviour change you should confirm.
Try it first
Paste a short SKILL.md, or give its path, and say "audit my SKILL.md and fix it, it never triggers". The first finding should point at the description.
What it needs
The skill as pasted text, a file path, or a skill name it can find in your skills folders. Read access for paths and names. No network access and no API keys.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
It audits against one template, the standard name/description frontmatter plus Purpose, When to Use, Process and Validation sections, so a skill written for a different house format gets findings it may not care about. It can't test whether the fixed skill triggers in your agent; it tells you the phrase to try. When a fix changes what the skill does, not just how it's written, it says so and asks you to confirm. Bundled scripts and reference files are read only to check the references to them.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago