SuperAPI AI Directory

artifact-radar

Office & productivity

View repository

Updated 10-15 22:14

About this skill

Artifact Radar

Purpose

Artifact Radar turns the current conversation's generated files into a clean, clickable delivery board.

It answers:

  • What files did this session create or modify?
  • When were they generated or last changed?
  • Where are they on disk?
  • Which ones can the user click or preview immediately?

When to Use

Use when the user asks:

  • "What files have been generated in the current session"
  • "What md, Excel, and images are there this time"
  • "Give me a clickable list of artifacts"
  • "List the images and documents just generated"
  • "where are the files you created"
  • "show generated artifacts with timestamps"
  • "make an artifact index for this session"

Naming

User-facing name: Artifact Radar

Chinese name: Artifact Radar

Positioning line:

把一次 AI 会话里散落的文档、图片、表格和交付物,一键变成可点击的产物雷达。

Core Principle

Do not claim the runtime has a perfect session artifact ledger.

Build a best-effort index from evidence, ordered by confidence:

  1. Files explicitly created, edited, copied, exported, rendered, or generated in the current conversation.
  2. Git staged, unstaged, or untracked files in the current repository.
  3. Recently modified deliverable files under relevant workspace roots.
  4. Known generated asset roots mentioned in the conversation, such as image generation directories.
  5. User-provided roots, file names, extensions, or time ranges.

If evidence is incomplete, say so and report the scan roots and time window.

Artifact Types

Include user-facing deliverables by default:

  • Markdown/text: .md, .txt, .rst
  • Documents: .docx, .doc, .pdf
  • Spreadsheets: .xlsx, .xls, .csv, .tsv
  • Presentations: .pptx, .ppt
  • Images: .png, .jpg, .jpeg, .webp, .gif, .svg
  • Web artifacts: .html, .htm
  • Data/handoff files: .json, .yaml, .yml
  • Script/tooling deliverables: .ps1, .sh, .bat, .cmd
  • Archives: .zip, .7z, .tar, .gz

Exclude source code, dependencies, build caches, logs, and internal binaries by default unless the user explicitly asks for them.

Required First Check

  1. Resolve the working root:
    • current working directory
    • active repository root if available
    • user-provided path if any
  2. Resolve the session window:
    • explicit user time range if provided
    • remembered files from the current conversation
    • recent modified files, defaulting to the last 8 hours
  3. Add extra roots when known:
    • image generation folders
    • export folders
    • temp/output folders used by the agent
  4. Prefer git status --short for high-confidence repo artifacts.
  5. Use the bundled scanner for recent deliverables.

Scanner

Run from the skill directory or pass the script path explicitly:

.\scripts\find-artifacts.ps1 -Roots . -Hours 8

Include extra roots when useful:

.\scripts\find-artifacts.ps1 -Roots . -ExtraRoots "$env:USERPROFILE\.codex\generated_images" -Hours 8

Use -GitChangedOnly when a recently cloned repository creates too much noise:

.\scripts\find-artifacts.ps1 -Roots . -GitChangedOnly

Output Format

Default to a task artifact strip, not a table.

The first visible block should look like a compact product UI artifact bar:

任务产生制品(3个):

[📄 刘宏库-详细报告.md](D:/path/刘宏库-详细报告.md)  [📄 高级Java后端-简历筛选报告.md](D:/path/高级Java后端-简历筛选报告.md)  [🖼️ 项目概览图.png](D:/path/project-overview.png)  [查看全部制品 →](#全部制品)

Rules:

  • Keep artifact labels short: file name only, no long path.
  • Use one inline Markdown link per artifact so each item is directly clickable.
  • Use a small type hint before the name when useful: 📄, 📊, 🖼️, 📽️, 🌐, 🗂️.
  • Show up to 5 artifacts in the strip. If there are more, add 查看全部制品 →.
  • Put the full path, timestamp, evidence, and confidence in the detailed list below.
  • If the client cannot render emoji cleanly, omit icons and keep the clickable file names.

Then provide a complete details section:

### 全部制品

| File | Type | Time | Evidence | Confidence | Notes |
|---|---|---|---|---|---|
| [report.md](D:/path/report.md) | Markdown | 2026-06-04 18:12 | git-status | high | Design brief |

For small results, the strip plus table is enough. For large results, group details by type or task.

Legacy compact table format is acceptable only when the user asks for a spreadsheet-like listing:

| File | Type | Time | Evidence | Confidence | Notes |
|---|---|---|---|---|---|
| [report.md](D:/path/report.md) | Markdown | 2026-06-04 18:12 | git-status | high | Design brief |

For images, include a preview when helpful:

![Preview](D:/path/image.png)
[Open image](D:/path/image.png)

Use absolute filesystem paths in Markdown links. Do not use file:// URIs.

For local file links, always wrap the link target in angle brackets:

[SKILL.md(源)](<C:/Users/admin/Documents/New project 2/artifact-radar/SKILL.md>)

This is required even when the path appears simple, because Windows paths often contain spaces, parentheses, or non-ASCII text. Without angle brackets, Markdown renderers may show a broken duplicate-looking row where the label and raw path appear separately.

Confidence Labels

  • high: explicitly created/edited in the conversation or appears in git status.
  • medium: matches the time window and artifact extensions under a relevant root.
  • low: only filename similarity or broad recent scan links it to the session.

Boundaries

  • Do not scan entire drives by default.
  • Do not open GUI applications unless the user explicitly asks.
  • Do not delete, move, rename, or modify artifacts while indexing them.
  • Do not expose file contents or secrets unless the user asks to inspect a specific file.
  • Do not list every recent source file unless the user asks for code changes too.
  • Do not call the list "complete" when roots or session start time are uncertain; call it a best-effort Artifact Radar.

Final Checklist

Before answering:

  1. Include timestamps.
  2. Use absolute clickable paths.
  3. Start with the task artifact strip when there are user-facing artifacts.
  4. Render image previews when useful.
  5. State scan roots and time window.
  6. Mention excluded categories and uncertainty briefly.