How do I install the Pptagent skill?

Run npx skillmds@latest add icip-cas/pptagent in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Pptagent safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Pptagent?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Pptagent free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Pptagent?

icip-cas (@icip-cas) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Pptagent by icip-cas — Agent Skill

Name: Pptagent
Author: icip-cas

Pptagent

Create and revise editable PowerPoint presentations from a brief using HTML slide sources, rendering, and visual review. Use when the requested deliverable is a PPTX deck. The host authors the slides; text-only hosts can use an external visual reviewer.

icip-cas Updated Sep 22, 2026

File contents

PPTAgent

Use the host agent to research and author slides. The CLI handles rendering, editable conversion, and verification that the delivered PPTX matches the reviewed sources. Do not start another authoring agent.

Setup

Resolve SKILL_ROOT to this directory. Use its .venv/bin/python when present, otherwise an activated Python 3.11+ environment with the dependencies installed. Set SKILL_PYTHON to that interpreter and run:

"$SKILL_PYTHON" "$SKILL_ROOT/scripts/pptagent.py" doctor --workspace "$PWD"

Read references/task-contract.md for file layout and HTML conversion constraints. The default mode: multimodal uses the host's image viewer. Only for mode: text, read references/text.md. When review_slides and review_deck MCP tools are available in text mode, use them for the corresponding review steps. Configuration is optional; defaults live in config.example.yaml.

Create or revise

Resolve the user's audience, message, page count, language, and aspect ratio. For a new workspace, run init --workspace "$PWD" --slides N, with optional --aspect-ratio 16:9|4:3|A1 and --language. Existing tasks use task.json.
Use available research tools as needed and preserve sources for checkable claims. A manuscript or asset manifest is useful for complex decks, not a prerequisite for every edit.
Use scaffold if starter HTML is useful; it preserves existing slides. Author exactly the requested page count and replace the starter text.
Run review-slides, inspect every returned image, then record each result:

"$SKILL_PYTHON" "$SKILL_ROOT/scripts/pptagent.py" record-slide-review   --workspace "$PWD" --slide slides/slide01.html --verdict pass   --summary "Readable; no clipping, overlap, or broken images."

Run build, then review-deck. Inspect the exported deck's contact sheets; open full-size pages for dense content. Record record-deck-review with --verdict and --summary. Text mode records visual results automatically.
Fix issues that affect the user's requirements or visible quality. Source changes require fresh renders, reviews, and a build. Shared resource changes invalidate all slide reviews. Rendering alone does not establish a pass.
Run finalize. Deliver answer.pptx only when complete=true. If the user explicitly accepts an incomplete draft, set delivery.mode: best-effort; delivery_ready=true then permits delivery with the failed checks disclosed.

Commands take --workspace with an absolute path and emit JSON. Use --help for arguments. A successful delivery includes a current final-report.json. Keep API keys in environment variables or the skill's .env, outside artifacts.

---name: pptagentdescription: Create and revise editable PowerPoint presentations from a brief using HTML slide sources, rendering, and visual review. Use when the requested deliverable is a PPTX deck. The host authors the slides; text-only hosts can use an external visual reviewer.---# PPTAgentUse the host agent to research and author slides. The CLI handles rendering,editable conversion, and verification that the delivered PPTX matches thereviewed sources. Do not start another authoring agent.## SetupResolve `SKILL_ROOT` to this directory. Use its `.venv/bin/python` when present,otherwise an activated Python 3.11+ environment with the dependencies installed.Set `SKILL_PYTHON` to that interpreter and run:```bash"$SKILL_PYTHON" "$SKILL_ROOT/scripts/pptagent.py" doctor --workspace "$PWD"```Read [references/task-contract.md](references/task-contract.md) for file layoutand HTML conversion constraints. The default `mode: multimodal` uses the host'simage viewer. Only for `mode: text`, read [references/text.md](references/text.md).When `review_slides` and `review_deck` MCP tools are available in text mode, usethem for the corresponding review steps. Configuration is optional; defaultslive in `config.example.yaml`.## Create or revise- Resolve the user's audience, message, page count, language, and aspect ratio. For a new workspace, run `init --workspace "$PWD" --slides N`, with optional `--aspect-ratio 16:9|4:3|A1` and `--language`. Existing tasks use `task.json`.- Use available research tools as needed and preserve sources for checkable claims. A manuscript or asset manifest is useful for complex decks, not a prerequisite for every edit.- Use `scaffold` if starter HTML is useful; it preserves existing slides. Author exactly the requested page count and replace the starter text.- Run `review-slides`, inspect every returned image, then record each result:```bash"$SKILL_PYTHON" "$SKILL_ROOT/scripts/pptagent.py" record-slide-review --workspace "$PWD" --slide slides/slide01.html --verdict pass --summary "Readable; no clipping, overlap, or broken images."```- Run `build`, then `review-deck`. Inspect the exported deck's contact sheets; open full-size pages for dense content. Record `record-deck-review` with `--verdict` and `--summary`. Text mode records visual results automatically.- Fix issues that affect the user's requirements or visible quality. Source changes require fresh renders, reviews, and a build. Shared resource changes invalidate all slide reviews. Rendering alone does not establish a pass.- Run `finalize`. Deliver `answer.pptx` only when `complete=true`. If the user explicitly accepts an incomplete draft, set `delivery.mode: best-effort`; `delivery_ready=true` then permits delivery with the failed checks disclosed.Commands take `--workspace` with an absolute path and emit JSON. Use `--help`for arguments. A successful delivery includes a current `final-report.json`.Keep API keys in environment variables or the skill's `.env`, outside artifacts.

Frequently asked questions