mirror of
https://github.com/github/awesome-copilot.git
synced 2026-02-21 02:45:12 +00:00
3436da9619
Weekly scheduled workflow that scans agents, prompts, instructions, and skills for potential duplicates based on name, description, and content similarity. Reports findings as a GitHub issue with task list checkboxes for review. Checks previous duplicate-review issues to exclude known accepted pairs. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
128 lines
5.1 KiB
Markdown
128 lines
5.1 KiB
Markdown
---
|
|
description: Weekly scan of agents, prompts, instructions, and skills to identify potential duplicate resources and report them for review
|
|
on:
|
|
schedule: weekly
|
|
permissions:
|
|
contents: read
|
|
issues: read
|
|
tools:
|
|
github:
|
|
toolsets: [repos, issues]
|
|
safe-outputs:
|
|
create-issue:
|
|
max: 1
|
|
close-older-issues: true
|
|
labels:
|
|
- duplicate-review
|
|
noop:
|
|
---
|
|
|
|
# Duplicate Resource Detector
|
|
|
|
You are an AI agent that audits the resources in this repository to find potential duplicates — resources that appear to serve the same or very similar purpose.
|
|
|
|
## Your Task
|
|
|
|
Scan all resources in the following directories and identify groups of resources that may be duplicates or near-duplicates based on their **name**, **description**, and **content**:
|
|
|
|
- `agents/` (`.agent.md` files)
|
|
- `prompts/` (`.prompt.md` files)
|
|
- `instructions/` (`.instructions.md` files)
|
|
- `skills/` (folders — check `SKILL.md` inside each)
|
|
|
|
### Step 1: Gather Resource Metadata
|
|
|
|
For each resource, extract:
|
|
|
|
1. **File name** (the path)
|
|
2. **Front matter `description`** field
|
|
3. **Front matter `name`** field (if present)
|
|
4. **First ~20 lines of body content** (the markdown after the front matter)
|
|
|
|
Use bash to read files efficiently. For skills, read `skills/<name>/SKILL.md`.
|
|
|
|
### Step 2: Identify Potential Duplicates
|
|
|
|
Compare resources and flag groups that look like potential duplicates. Consider resources as potential duplicates when they share **two or more** of the following signals:
|
|
|
|
- **Similar names** — file names or `name` fields that share key terms (e.g., `react-testing.prompt.md` and `react-unit-testing.prompt.md`)
|
|
- **Similar descriptions** — descriptions that describe the same task, technology, or domain with only minor wording differences
|
|
- **Overlapping scope** — resources that target the same language/framework/tool and the same activity (e.g., two separate "Python best practices" instructions)
|
|
- **Cross-type overlap** — an agent and a prompt (or instruction and agent) that cover the same topic so thoroughly that one may make the other redundant
|
|
|
|
Be pragmatic. Resources that cover related but distinct topics are NOT duplicates. For example:
|
|
- `react.instructions.md` (general React coding standards) and `react-testing.prompt.md` (React testing prompts) are **not** duplicates — they serve different purposes.
|
|
- `python-fastapi.instructions.md` and `python-flask.instructions.md` are **not** duplicates — they target different frameworks.
|
|
- `code-review.agent.md` and `code-review.prompt.md` that both do the same style of code review **are** potential duplicates worth flagging.
|
|
|
|
### Step 3: Check for Known Accepted Duplicates
|
|
|
|
Before finalizing the report, search for **previous issues** labeled `duplicate-review` in this repository:
|
|
|
|
```
|
|
Search for issues with label "duplicate-review" that are closed
|
|
```
|
|
|
|
Read the comments and body of those past issues to find any pairs or groups that reviewers have explicitly marked as **"accepted"** or **"not duplicates"**. Look for phrases like:
|
|
- "accepted as-is"
|
|
- "not duplicates"
|
|
- "intentionally separate"
|
|
- "keep both"
|
|
- checked task list items (i.e., `- [x]`)
|
|
|
|
Exclude those known-accepted pairs from the current report. If you include a group that was previously reviewed, add a note: `(previously reviewed — see #<issue-number>)`.
|
|
|
|
### Step 4: Produce the Report
|
|
|
|
Create an issue titled: `🔍 Duplicate Resource Review`
|
|
|
|
Format the body as follows:
|
|
|
|
```markdown
|
|
### Summary
|
|
|
|
- **Potential duplicate groups found:** N
|
|
- **Resources involved:** M
|
|
- **Known accepted (excluded):** K pairs from previous reviews
|
|
|
|
### How to Use This Report
|
|
|
|
Review each group below. If the resources are intentionally separate, check the box to mark them as accepted. These will be excluded from future reports.
|
|
|
|
### Potential Duplicates
|
|
|
|
#### Group 1: <Short description of what they share>
|
|
|
|
- [ ] Reviewed — these are intentionally separate
|
|
|
|
| Resource | Type | Description |
|
|
|----------|------|-------------|
|
|
| `agents/foo.agent.md` | Agent | Does X for Y |
|
|
| `prompts/foo.prompt.md` | Prompt | Also does X for Y |
|
|
|
|
**Why flagged:** <Brief explanation of the similarity>
|
|
|
|
---
|
|
|
|
#### Group 2: ...
|
|
|
|
<repeat for each group>
|
|
```
|
|
|
|
Use `<details>` blocks to collapse groups if there are more than 10.
|
|
|
|
### Safe Output Guidance
|
|
|
|
- If you find potential duplicates: use `create-issue` to file the report.
|
|
- If **no** potential duplicates are found (after excluding known accepted ones): call `noop` with the message: "No potential duplicate resources detected. All resources appear to serve distinct purposes."
|
|
|
|
## Guidelines
|
|
|
|
- Be conservative — only flag resources where there is a genuine risk of redundancy.
|
|
- Group related duplicates together (don't list the same pair twice in separate groups).
|
|
- Sort groups by confidence (strongest duplicate signals first).
|
|
- Include cross-type duplicates (e.g., an agent and a prompt doing the same thing).
|
|
- Limit the report to the top 20 most likely duplicate groups to keep it actionable.
|
|
- For skills, use the folder name and description from `SKILL.md`.
|
|
- Process resources in batches to stay within time limits — prioritize name and description comparison, then spot-check content for top candidates.
|