Deckprobe logo

Deckprobe

Organization
deckflow
deckprobe

Inspect PDF, Microsoft Office (docx/xlsx/pptx and legacy doc/xls/ppt), and Apple iWork (key/numbers/pages) files without opening or rendering them. Use for page and slide and sheet counts, title/author/created metadata, encryption and macro and digital-signature and JavaScript risk signals, sheet and table structure, embedded assets and fonts, corruption checks, and verifying a file really is the format its extension claims. Also use to inventory many documents at once. Returns bounded, deterministic JSON on stdout. Never renders, never runs macros, never follows external references, never sends the file anywhere.

Overview

Publisherdeckflow
Repositorydeckprobe
Skill namedeckprobe
Stars
114
Forks
10
Bundled files
5
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 5 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by deckflow on GitHub. Read the source before you install it.

Installation

Install the Deckprobe AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/deckflow/deckprobe/tree/main/skills/deckprobe
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/deckflow/deckprobe.git /tmp/deckprobe
mkdir -p .claude/skills
cp -r /tmp/deckprobe/skills/deckprobe .claude/skills/deckprobe
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Deckprobe in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Deckprobe on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Deckprobe is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

DeckProbe

ffprobe for documents. One shell command answers a specific question about a PDF, Office, or iWork file; the answer comes back as JSON on stdout and the exit status carries the verdict.

Reach for it when a task needs facts about a document — how many slides, is it encrypted, does it have macros, is it corrupt, is this really a .pptx — rather than the text inside it. It is safe on untrusted input: nothing is rendered, no macro runs, no external reference is followed.

1. Make sure it runs

npx skills add installs this document only, not the binary. Check first:

bash
deckprobe --version
  • Prints a version → use deckprobe as written throughout this document.
  • Command not found, but node exists → the command is npx -y @deckflow/deckprobe. Prefix every example below with it. First run downloads the package; later runs are cached.
  • Neither → ask the user to install one of:
    • curl --proto '=https' --tlsv1.2 -LsSf https://github.com/deckflow/deckprobe/releases/latest/download/deckprobe-installer.sh | sh
    • npm install -g @deckflow/deckprobe
    • cargo install --git https://github.com/deckflow/deckprobe --locked deckprobe

If you cannot get it running, say so and stop. Do not fall back to unzipping the package, grepping XML, or parsing PDF bytes by hand — avoiding exactly that is the point of this tool.

2. The shape of every call

bash
deckprobe [-t TARGETS] [-l LEVEL] [-c CONFIDENCE] [--view values] [--pretty] FILE

Start here when you do not yet know what you need:

bash
deckprobe -t @summary --view values --pretty report.pdf

--view values returns a compact target -> value map. Drop it when you also need each value's confidence, the path that produced it, and the measured cost.

Check the exit status before parsing stdout. A non-zero status still writes valid JSON, but it is an error envelope, not a report.

3. Choosing targets

-t accepts short names (slide_count), canonical names (powerpoint.slide_count), and selector presets. It is repeatable and comma-separated: -t @header,title,page_count.

SelectorExpands to
@headerContainer identity only — format, size, extension match, encryption flag
@defaultThe driver's defaults for the active level. Used when -t is omitted
@summaryIdentity, common metadata, and primary structure
@securityEncryption, macros, signatures, external references, active content
@structureFormat-owned counts, names, and dimensions
@assetsImages, media, previews, fonts, embedded objects
@qualityIntegrity, repair, extension match, conformance
@formatEvery format-specific target at the active level
@allEverything available at the active level

@summary deliberately omits a statistic the current driver can only get from a full-file read. PDF page count is the notable case — ask for page_count or @structure explicitly.

Never guess a target name. The tool documents itself:

bash
deckprobe targets --format pptx --pretty   # ids, aliases, min level, value type, cost class
deckprobe formats --pretty                 # drivers, profiles, support boundaries
deckprobe schema --pretty                  # the authoritative report JSON Schema

references/targets.md has the full catalogue per format if you would rather read than run.

4. Probe level and confidence

-l / --level picks the budget and which paths are eligible. Accepts h/m/d too.

LevelUse it for
headerIdentity and container properties. ~500 ms budget
metadataDefault. Bounded metadata and common structural counts. ~500 ms
deepHigher-cost paths. ~5 s — only when a target's min_level says deep

-c / --confidence filters eligible paths: low < medium < high (default) < exact. Lower it when a target comes back unresolved and an approximate answer is acceptable; raise it with -C target=exact for a single target that must be authoritative.

5. Reading the report

jsonc
{
  "schema_version": 2,
  "status": "ok",              // "ok" | "partial" | "error"
  "input":  { "display_name": "deck.pptx", "source_kind": "local_file" },
  "driver": { "id": "powerpoint", "profile": "pptx" },
  "results": {
    "powerpoint.slide_count": {
      "status": "resolved",     // see the eight values below
      "value": 31,
      "confidence": "high",     // none | low | medium | high | exact
      "path": "powerpoint.app_statistics",
      "source": "docProps/app.xml saved statistic"
    }
  },
  "execution": { "unresolved_targets": [], "actual_cost": { } },
  "diagnostics": []
}

A result's status is one of resolved, estimated, planned, unknown, unsupported, invalid, budget_exceeded, failed. Only resolved and estimated carry a value. unknown is the common one: the path ran and the document simply does not record that fact.

  • The report's own status is ok or partial. partial is not a failure — at least one requested target could not be resolved at the requested confidence, and the rest are still valid. Check execution.unresolved_targets.
  • Add --strict when an unresolved target must fail the command. The full report is still written, and the exit status becomes 5.
  • results is keyed by the canonical target name even when you asked with a short alias.

6. Exit status — and what to do about it

StatusMeaningYour next move
0SuccessParse stdout. Unresolved targets are allowed unless --strict
1Invalid request or unsupported targetYou named a target this format has no path for. Run deckprobe targets --format <fmt>
2CLI syntax error, or the file is missing/unreadableFix the flags or the path
3Unsupported or unrecognized formatCheck the extension against deckprobe formats
4Malformed input or budget exceededGenuinely damaged file, or raise -b/-x/-e/-T. See references/limits.md
5--strict and a target was unresolvedThe report is on stdout; decide whether the missing target matters
6Internal failureA bug. Report it

7. Many files at once

bash
printf '%s\n' '{"path":"a.pdf"}' '{"path":"b.pptx"}' | deckprobe --jsonl -t @summary

One compact JSON per non-empty input line. A per-record error does not stop the run; the process exits with the highest per-record status. Each record is a JSON string path, {"path":"..."}, or {"name":"report.pdf","data_base64":"..."}. Add an id to an object record when the caller needs stable attribution; JSONL error envelopes echo that id and the original path under input.

For bytes on stdin, -n supplies the logical filename that selects the format:

bash
curl -sL "$url" | deckprobe -n download.pdf -t @security -

8. What DeckProbe will not do

It has no renderer, no OCR, no text extraction, and no macro interpreter. It never opens a network connection and never resolves an external reference. Legacy XML iWork packages are rejected by design — only modern IWA is supported. .xlsb is identity-only. PDF XMP is not merged into the metadata targets.

Routing is by filename extension, then verified against the container. Renaming a .pptx to .docx returns MALFORMED_INPUT, and -f/--input-format is an assertion, not an override — it cannot force an unrelated parser onto a file.

Never parse stderr. Every result, including every error, is one JSON value on stdout.

References

Read these only when you need them:

  • references/targets.md — the full target catalogue per format, with aliases and minimum levels. Read before naming a target you have not verified.
  • references/recipes.md — task → command lookup for the common questions.
  • references/output.md — the complete report shape, error codes, and cost accounting, for when you are consuming the JSON programmatically.
  • references/limits.md — budgets, the -b/-x/-e/-T overrides, -O format options, and how to diagnose an exit 4.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Deckprobe AI skill do?

Inspect PDF, Microsoft Office (docx/xlsx/pptx and legacy doc/xls/ppt), and Apple iWork (key/numbers/pages) files without opening or rendering them. Use for page and slide and sheet counts, title/author/created metadata, encryption and macro and digital-signature and JavaScript risk signals, sheet and table structure, embedded assets and fonts, corruption checks, and verifying a file really is the format its extension claims. Also use to inventory many documents at once. Returns bounded, deterministic JSON on stdout. Never renders, never runs macros, never follows external references, never...

Why use Deckprobe on TypingMind?

Because you install it once and use it with any model. Deckprobe is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Deckprobe in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/deckflow/deckprobe/tree/main/skills/deckprobe. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Deckprobe?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Deckprobe?

As many as you like. As long as a model supports skills, you can use Deckprobe with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Deckprobe AI skill free?

Yes. It is published on GitHub by deckflow under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇