Jetformat

Add the GPT Skill

The Skill gives GPT or Codex a clear rule. Call Jetformat for Office and PDF work instead of parsing files directly.

By Jetformat

The Skill is a set of instructions, not a remote conversion service. It tells the host model when the local CLI is required, which commands are valid, and how to report outputs.

Install it

Add the Jetformat plugin from the command line:

codex plugin marketplace add jetformat/jetformat-agent
codex plugin add jetformat@jetformat

What the model should do

  1. Check that jetformat is installed.
  2. Run one command with the user’s real input path.
  3. Never unzip Office files or invent PDF bytes.
  4. Report the output path Jetformat wrote.

Try a request

Ask: “Run ppt to-pdf on board-deck.pptx and show the first page.” The model should run ppt to-pdf, then pdf render for the requested preview.

The Skill never sells

When the account has no credits left, the CLI prints a link to how credits work. The Skill does not quote prices or open checkout.

Read the full model guide.

Verify the local execution environment

  1. Run jetformat --version from the same environment the agent uses for shell commands.
  2. Confirm a free read can open the real input file before attempting conversion.
  3. Sign in to the CLI when the requested operation is a metered write.
  4. Convert to a new output path, then inspect that output before reporting success.

The Skill supplies instructions; it does not install a renderer inside the model or make files on another machine available automatically. A working terminal installation is useful evidence, but an agent running in a different environment may need its own PATH and file access.

Diagnose the failed stage

Symptom Check first
Binary cannot be found Executable availability and PATH
Source cannot be opened Real input path and execution environment
Write is refused by billing CLI authorization and available credits
File exists but layout is wrong Fonts, feature limits, and reviewed output pages

Do not inspect an older output after a failed conversion. Choose a new path when testing, check the exit outcome, and retain the original source for comparison.

Report the artifact and the review scope

For an illustrative 10-page PDF output, conversion uses 10 credits. Rendering 2 previews adds 2 credits; reading the page count adds 0 credits. These units describe the example, not a measured result from the user’s document.

The final report should identify the file actually written, its observed page count, and which pages or values were checked. If the agent cannot inspect images, it can still generate previews, but should leave visual assessment unresolved. See local PDF workflows and file-processing boundaries before passing extracted document text to a separate host.

Last updated