← Back to Skills Marketplace
agisearch

Converter

by AGIsearch · GitHub ↗ · v2.0.0 · MIT-0
cross-platform ✓ Security Clean
461
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install converter
Description
A local-first conversion router and format strategist. Identifies the safest local path for document, image, audio, video, archive, and data transformations....
README (SKILL.md)

Converter — The Local Conversion Router

Format is not a detail. Format is a decision about the future of information.

This skill is a local-first conversion router.
It does not "magically" convert every file. It identifies the safest local path, selects the right toolchain, explains what will be preserved or lost, and executes the conversion only when the required local tools are available.

If the local tools are missing, it does not bluff.
It returns a conversion plan.


Access & Tool Model

This skill is an instruction-only orchestrator that depends on host-available local binaries.

  • Local-First: no third-party upload by default
  • Toolchain-Dependent: execution depends on installed local tools such as FFmpeg, Pandoc, ImageMagick, LibreOffice (soffice), and 7-Zip
  • Planner Mode: if a required binary is missing, the skill returns a manual conversion route or fallback plan instead of pretending success
  • Privacy Boundary: no DRM bypass, no password cracking, no unauthorized decryption of protected files
  • Explicit External Consent: if a conversion would require an external service, the skill must ask before any file leaves the local environment

What This Skill Does NOT Do

  • It does not bypass DRM or password protection.
  • It does not crack encrypted archives or locked documents.
  • It does not guarantee perfect fidelity for structurally complex conversions.
  • It does not pretend lossy conversions are reversible.
  • It does not upload files to third-party services unless the user explicitly authorizes that route.

Supported Local Toolchains

1. Pandoc

Used for:

  • Markdown ↔ DOCX
  • Markdown ↔ HTML
  • DOCX → Markdown
  • HTML ↔ EPUB
  • text-centric document restructuring

Best for:

  • structural document conversion
  • preserving headings, lists, links, and basic tables
  • format migration where content structure matters more than exact page layout

2. FFmpeg

Used for:

  • audio transcoding
  • video transcoding
  • codec/container migration
  • bitrate and stream handling

Best for:

  • WAV ↔ MP3
  • MOV/AVI/MKV ↔ MP4
  • extracting audio tracks
  • making files compatible with devices and platforms

3. ImageMagick (magick)

Used for:

  • raster image conversion
  • resizing
  • color space adjustments
  • batch image transformations

Best for:

  • PNG ↔ JPEG
  • JPEG ↔ WebP
  • TIFF ↔ PNG
  • simple local image pipeline work

4. LibreOffice (soffice)

Used for:

  • headless export of office files
  • spreadsheet and presentation conversions
  • office documents to PDF

Best for:

  • DOCX → PDF
  • XLSX → PDF
  • PPTX → PDF
  • ODT/ODS/ODP exports

5. 7-Zip (7z)

Used for:

  • archive extraction
  • archive repackaging
  • common compression path conversions

Best for:

  • ZIP / 7z / TAR operations
  • multi-file bundling
  • local archive handling

Standard Output Format

Every request should begin with a route diagnosis.

CONVERSION ROUTE ANALYSIS

  • Path: [Source] → [Target]
  • Type: [Lossless / Lossy / Structural Reconstruction / Legacy-to-Modern / Unsupported]
  • Primary Tool: [FFmpeg / Pandoc / ImageMagick / soffice / 7z / Fallback]

FIDELITY NOTES

  • Preserved: [what should survive]
  • Lost / At Risk: [what may degrade or disappear]
  • Compatibility: [device / platform / software caveats]

EXECUTION STATUS

  • Status: [Ready to Execute / Missing Binary / Unsupported Path / Partial-Fidelity Only]
  • Next Step: [command / local route / fallback recommendation]

Fidelity Philosophy

1. The One-Way Door

Lossy-to-lossless is not restoration.
MP3 → FLAC does not recover discarded audio detail.
JPEG → PNG does not reconstruct compression damage.

The skill should explicitly flag fake quality boosts.

2. Structural Reconstruction Is Not Exact Reproduction

Some conversions are not simple transcoding. They are reconstruction.

Examples:

  • PDF → editable text formats
  • slide deck → text document
  • complex layout → reflowable ebook

The skill should state clearly when exact layout fidelity is unlikely.

3. Intent Matters

The same requested conversion may require different settings depending on why the user wants it.

Examples:

  • “for email”
  • “for printing”
  • “for editing”
  • “for web upload”
  • “for archival storage”

The route should optimize for actual intent, not just target extension.


Partial-Fidelity / Unsupported Paths

These paths require explicit caution:

  • scanned PDF → editable DOCX / Markdown
  • raster image → clean vector logo
  • DRM-protected ebooks
  • encrypted archives
  • proprietary design files with incomplete local tool support
  • codec/container combinations unsupported by installed binaries

When a path is only partially reliable, the skill should say so plainly and offer the safest fallback.


Common Conversion Domains

Documents

Common paths:

  • Markdown ↔ DOCX
  • DOCX → PDF
  • HTML ↔ EPUB
  • ODT → DOCX
  • TXT / Markdown → structured document output

Images

Common paths:

  • PNG ↔ JPEG
  • JPEG ↔ WebP
  • TIFF ↔ PNG
  • HEIC → JPEG (if supported by local tools)
  • batch image format migration

Audio

Common paths:

  • WAV → MP3
  • FLAC → AAC / MP3
  • M4A ↔ MP3
  • extract audio from video

Video

Common paths:

  • MOV / MKV / AVI → MP4
  • H.264 / H.265 transcodes
  • compatibility versions for web/device playback
  • bitrate / quality / file-size balancing

Data / Office

Common paths:

  • DOCX / XLSX / PPTX → PDF
  • CSV ↔ TSV
  • structured text export when office tools are available

Archives

Common paths:

  • ZIP extraction
  • 7z extraction
  • repackage to ZIP / 7z
  • bundle directories for delivery or storage

Interaction Patterns

Scenario A: Office to PDF

Input:
“Convert this DOCX to PDF.”

Diagnose:
Document export -> local office conversion -> likely soffice

Output:
Route analysis + command or execution path + fidelity note


Scenario B: High-quality audio conversion

Input:
“Turn this WAV recording into a high-quality MP3.”

Diagnose:
Audio transcoding -> lossy target -> ffmpeg

Output:
Bitrate-aware FFmpeg path + note on what is lost


Scenario C: Batch web images

Input:
“Convert all JPEGs in this folder to WebP for my website.”

Diagnose:
Batch raster conversion -> web optimization -> magick

Output:
Batch command + size/fidelity tradeoff note


Scenario D: Missing tool fallback

Input:
“Convert this HEIC photo to JPEG.”

Diagnose:
Image conversion -> tool availability check -> possible magick or OS-native fallback

Output:
If tool exists: execute locally
If tool is missing: provide a fallback plan such as sips on macOS or local install guidance


Scenario E: PDF to Markdown

Input:
“Turn this PDF into Markdown.”

Diagnose:
Structural reconstruction problem, not a guaranteed direct-format conversion

Output:
Warn about layout loss, scanned/OCR risk, and whether preprocessing is required before Markdown export


Privacy & Safety

Converter is local-first.
It should either:

  1. convert locally using trusted binaries, or
  2. provide a local fallback route.

It should not quietly upload sensitive files to websites, APIs, or third-party services.

The user should always know:

  • which tool is being used
  • what quality risks exist
  • whether the conversion is exact, lossy, reconstructed, or only partially reliable

Engineering Identity

  • Type: Instruction-only Local Conversion Router
  • Execution Model: toolchain-dependent local orchestration
  • Fallback Philosophy: if direct execution is unavailable, return the safest honest plan
  • Bias: privacy, fidelity, and explicit tradeoff disclosure over magical promises

The point of this skill is not to promise impossible conversion.

It is to make format change honest, local, and predictable.

Usage Guidance
This skill appears coherent and local-first, but remember it will run host binaries and operate on your local files when invoked. Only enable it if you trust the agent and the host environment to run ffmpeg/pandoc/magick/soffice/7z safely. Verify those binaries are installed from trusted sources, and be cautious about converting DRM-protected or encrypted files — the skill says it won't bypass protections but will need explicit permission before any upload. If you want stricter control, require manual approval before execution or limit the skill's ability to access sensitive directories.
Capability Analysis
Type: OpenClaw Skill Name: converter Version: 2.0.0 The skill is a legitimate local-first file conversion orchestrator designed to use standard system binaries like ffmpeg, pandoc, and ImageMagick. The instructions in skill.md and examples.md are well-structured, emphasize user privacy, and explicitly prohibit data exfiltration or unauthorized activities such as DRM cracking. No evidence of malicious intent, obfuscation, or intentional vulnerabilities was found.
Capability Assessment
Purpose & Capability
Name/description match required binaries and capabilities. Declared dependencies (ffmpeg, pandoc, magick, soffice, 7z) are exactly the kinds of tools needed for the stated document/image/audio/video/archive conversions.
Instruction Scope
SKILL.md is an instruction-only orchestrator that advises and runs local toolchains and returns conversion plans when tools are missing. It does not instruct the agent to read unrelated files, request unrelated credentials, or quietly upload data without consent. It explicitly requires explicit consent before any external upload.
Install Mechanism
No install spec and no code files — lowest-risk delivery. The skill relies on host-provided binaries rather than downloading or extracting code at install time.
Credentials
No environment variables, credentials, or configuration paths are requested. The declared footprint is minimal and proportional to a local conversion tool orchestrator.
Persistence & Privilege
always is false and the skill is user-invocable. It does not request persistent or elevated privileges, nor does it modify other skills or system-wide settings.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install converter
  3. After installation, invoke the skill by name or use /converter
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.0.0
Converter 2.0.0 — Major local-first overhaul: - Now acts as a local-first "conversion router," identifying the safest local path and toolchain for every conversion. - Uses host-available binaries (e.g., FFmpeg, Pandoc, ImageMagick, LibreOffice, 7-Zip); if tools are missing, provides a manual fallback plan instead of bluffing. - Strong privacy focus: no cloud uploads or DRM bypass; no file leaves local environment without explicit user consent. - Adds detailed "Conversion Route Analysis" for every request, including fidelity notes and command recommendations. - Fully documents supported toolchains, conversion domains, and best-practice guidelines for each file type.
v1.0.0
Initial release
Metadata
Slug converter
Version 2.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 2
Frequently Asked Questions

What is Converter?

A local-first conversion router and format strategist. Identifies the safest local path for document, image, audio, video, archive, and data transformations.... It is an AI Agent Skill for Claude Code / OpenClaw, with 461 downloads so far.

How do I install Converter?

Run "/install converter" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Converter free?

Yes, Converter is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Converter support?

Converter is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Converter?

It is built and maintained by AGIsearch (@agisearch); the current version is v2.0.0.

💬 Comments