Index of every page on this site: llms.txt
All skills
modelflare-ai avatar

/ppt-runner

@da623fc
by Modelflaremodelflare-ai/anymodelbot4.8k stars
109
Run it yourself

Read this first for every presentation request in this bot, and before using the ppt-master skill. It covers how to run PPT Master here - the runtime, command paths, confirmation cards, images, app/state.json and delivery - and overrides ppt-master wherever they differ.

  • 2 files
  • 28.9 KB
  • Updated 3 weeks ago
  • GitHub
Use this Skill: https://skilld.dev/gh/modelflare-ai/anymodelbot/ppt-runner

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈71 tokens for metadata: the name and description. ≈2.2k when used: this file.

Description uses 3.5% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈283 characters, or 3.5%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Running PPT Master in this app

ppt-master owns the workflow: routing, stages, design rules and exports. This file only says how to run it here. Where the two differ, this file wins. These are the run instructions for this host.

Commands

PM below means this script. The app sets HOME to the bot's own folder:

python3 "$HOME/skills/ppt-runner/scripts/pm.py" <command>        # macOS, Linux
python "$env:HOME\skills\ppt-runner\scripts\pm.py" <command>     # Windows PowerShell

If python3 doesn't start on Windows, use python. If neither exists, set runtime in app/state.json to {"ready": false, "problem": "no_python"}, ask the user to install Python 3.10–3.14 from python.org, and stop.

Every upstream command goes through PM run, because the packages are only installed there. Upstream's

python3 ${SKILL_DIR}/scripts/svg_to_pptx.py <project_path> --quick-generate --no-notes

becomes

PM run svg_to_pptx.py <project_path> --quick-generate --no-notes

SKILL_DIR is the skillDir printed by PM doctor. Use it to read upstream's files. Long commands move to the background on their own; you are told when they finish.

Before any work

  1. Run PM doctor. It also runs upstream's integrity guard.
  2. guardOk is false: stop, report the failure as it is. Never inspect, repair or bypass the guard.
  3. Packages missing: tell the user in one line that the first run installs components, then run PM setup. If pip fails on what looks like a network problem, retry once with --mirror https://pypi.tuna.tsinghua.edu.cn/simple.
  4. Read ppt-master's SKILL.md and workflows/routing.md, then select the route as they say.

On-demand components, installed the first time they're needed:

Needed for Command
PDF sources PM setup --extra pdf. Tell the user this installs PyMuPDF (AGPL-3.0) into this bot's folder only
Narration (Edge voices need no key) PM setup --extra narration
Jupyter notebook sources PM setup --extra notebook

Where this host differs

  • Files. Every project lives under app/projects/, because the view can only read app/: PM run project_manager.py init <name> --dir app/projects, pptx_to_svg.py "<source.pptx>" -o "app/projects/<slug>_<YYYYMMDD>" --roundtrip …, and template workspaces at app/projects/<template_id>. User files arrive in app/inbox/. Never write outside the workspace.
  • Confirmation. Treat this as the user's explicit instruction to confirm here rather than on a page: always use upstream's chat branch and never launch confirm_ui. Each ⛔ BLOCKING gate becomes a card: write a JSON file, run PM pending <file>, say in one line that it's ready for review, and end your turn. See "Cards" below.
  • Live preview is off by run instruction. Don't start svg_editor, and in Step 6 report that preview is shown in the app. The view shows pages from svg_final/, or from the working SVGs until those exist.
  • Image to PPTX is unavailable. Upstream supports it only on Codex and defines no alternative. Say so in one sentence and offer Beautify or a normal Generate instead. Never pretend one is the other.
  • Images. Never ask for an API key, and don't configure IMAGE_BACKEND or a .env. AI images are upstream's Path B (host-native): use generate_image from the media skill, then copy the file from media/ to <project_path>/images/<filename>. Stock images: PM run image_search.py … needs no key. If the user has no use for images, don't generate any.
  • Icons. Only tabler-filled is bundled. icons.library is tabler-filled or none, and where a template or reference suggests another library, use tabler-filled. Brand logos must come from the user.
  • Sound effects and auto-update aren't included. Skip upstream's optional sound pass. update_repo.py doesn't exist here.
  • Video. Native MP4 export needs Windows PowerPoint 2016+. Elsewhere say that it isn't available, and deliver the PPTX.
  • Sponsors. Never read SPONSORS.md or SPONSORS_CN.md. If asked which model to use, point to this app's model settings.
  • Long decks. One turn in this app stops after 30 minutes. For more than about 12 pages:
    • Default: confirm generation_mode: split at Stage 2. When upstream's split handoff says to open a fresh chat, call task_start instead, asking it to continue the project by its absolute path through upstream's resume-execute stage, following this file. Run PM sync … --phase authoring --background.
    • Quick: there is no plan to resume from, so run the whole thing in task_start.
    • A background task can't ask questions, so all confirmation happens before it starts. It ends with task_report, listing the PPTX under files.

Keeping the view up to date

The view only shows app/state.json. Use PM for the job, previews and exports; don't write those by hand. They are read from disk, and done is refused without a valid deck.

PM sync <project_path> --kind generate --name "<title the user would recognize>" --format ppt169 --route generate --profile quick --phase authoring --total 5

Run it when the phase changes and after every page you author. Later calls for the same project need only --phase. The first sync also removes the sample data.

--phase When
importing Converting sources, research, project setup
spec_pending Waiting at a card (PM pending sets this)
authoring Writing pages. --total is the page count
finalizing finalize_svg.py and quality gates
exporting svg_to_pptx.py
done The deck exists. Add --warn "…" for anything the user should know
error Add --message "…": where it stopped and what the user can do. Keep the files
cancelled The user stopped the job. Keep the files

--kind: generate, template, edit_native. --route: generate, create_template, edit_native. --profile: default, quick, beautify.

Finishing

  1. Export with the route's own command. With --native-charts-and-tables the result is a second file; both are listed.
  2. PM verify <file.pptx>. If it fails, repair at the owning layer and export again.
  3. PM sync <project_path> --phase done.
  4. Deliver the file with present, and say in one line what you made. Details are in the view.

Don't overwrite a previous export without asking. Exports are timestamped, so the normal path doesn't.

Cards

PM pending <file> shows a card in the view. The file:

{
  "stage": "stage1",
  "title": "Confirm the audience and the template",
  "summary": "One or two sentences on what you recommend and why.",
  "fields": [{ "label": "Audience", "value": "Investors" }, { "label": "Canvas", "value": "16:9" }],
  "choices": [{ "id": "free_design", "label": "Free design", "description": "Designed for this deck", "recommended": true }]
}

stage is stage1 or stage2 in Default and Beautify, and plan for Edit Native's plan. Show what upstream requires at that gate in fields and choices, in the user's language; summary is short and the rest stays in the project's files. Nothing counts as confirmed until the user answers: silence confirms nothing, and you never answer for them.

Messages from the view

Buttons arrive as [ui-action] {…}, usually with a line of text from the user:

Action Do
{"type":"start","route":"generate","profile":"quick","files":[…]} Start that route and profile with those files. profile is default, quick or beautify
{"type":"start","route":"create_template","files":[…]}, {"type":"start","route":"edit_native","files":[…]} Start that route. Ask only what the route's prerequisites require, such as new material
{"type":"confirm_spec","stage":"…","projectId":"…","choice":"…"} The user confirmed this gate. Record it the way upstream's chat branch does, run PM pending --clear, continue
{"type":"revise_spec","stage":"…","projectId":"…"} The text says what to change. Revise and show the card again
{"type":"resume","projectId":"…"} The job stopped mid-way (stopped by the user or cut off). Check the files, then continue from there, or restart a Quick run whose context is gone
{"type":"recheck"} Run PM doctor, and PM setup if it says so

If the user stopped a job and sends something unrelated, first run PM sync <project_path> --phase cancelled.

In chat apps

Messages from WeChat, Feishu, Telegram and the like have no view. State still updates as above, so the view is current back on the computer. Write each card as a short message and take the answer as text, deliver decks with present, and don't mention buttons or the view.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at da623fc. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 3 weeks ago

README badge

README badge for modelflare-ai/anymodelbot/ppt-runner