Skills face-swap

📦

face-swap

Name: face-swap
Author: agentspace-so

Content revision r2 High Risk ⚙️ External commands🌐 Network access📁 Filesystem access

Create Consented Face Swaps with RunComfy

Choosing a face-swap model for each image or video workflow can be difficult. This skill routes consented requests to suitable RunComfy endpoints and CLI commands.

Supports: Claude Codex Code(CC)

⚠️ 38 Poor

Stop for confirmation before installing.

Review the plan and obtain explicit user consent before changing files.

Install with my Agent

Copy this request to your Agent. It includes the canonical Skill page and manifest.

Agent request

Review the Skillstore skill "face-swap" from https://skillstore.io/skills/agentspace-so-face-swap.md and its manifest at https://skillstore.io/api/skills/agentspace-so-face-swap/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

Your Agent should still show its plan and request any confirmation required by the security policy.

Agent-readable resources

Use these links when an AI agent, crawler, or script needs clean context instead of reading the full page.

Markdown detail GET /skills/agentspace-so-face-swap.md Signed manifest GET /api/skills/agentspace-so-face-swap/manifest Signed lockfile GET /api/skills/agentspace-so-face-swap/lockfile

Test it

Using "face-swap". Replace the actor in this approved campaign still with the consenting spokesperson from my reference portrait.

Expected outcome:

Selected GPT Image 2 Edit for precise multi-reference composition. The scene remains primary, and the portrait supplies only the approved facial identity.

Using "face-swap". Make this approved character perform the movement from my reference video.

Expected outcome:

Selected Kling Motion Control Pro because motion preservation is required. The performance video supplies movement, and the character image supplies appearance.

Using "face-swap". Use one approved identity across twelve product scenes with consistent dimensions.

Expected outcome:

Selected Nano Banana Edit for batch identity consistency. The identity reference comes first, followed by scenes with locked aspect ratio and resolution.

Security Audit

High Risk

v5 • 7/23/2026 Open versioned report

Most static findings are false positives caused by Markdown formatting, documentation links, placeholder URLs, and model-selection prose. Confirmed risks include third-party command execution, external processing of biometric media, unsafe shell quoting, and advisory-only consent controls.

Files scanned

321

Lines analyzed

Review items

False positives ignored

Confirmed security concerns (3)

High

SKILL.md:40-42 SKILL.md:305

Unsafe Shell Quoting for User-Controlled JSON

Examples place JSON inside single-quoted Bash arguments, while the skill claims prompts and URLs have no shell-injection surface. A single quote in user data can terminate the shell string unless arguments are safely constructed.

The Bash examples visibly wrap JSON in single quotes and the skill expects prompts and asset URLs from users. No escaping or shell-free argument construction is specified.

High

SKILL.md:55-61 SKILL.md:302

Unenforced Consent and Impersonation Controls

The skill enables identity substitution and states that it does not gate inputs. Refusal guidance exists, but no mandatory consent check prevents nonconsensual impersonation or harmful synthetic media.

The file explicitly says the skill does not gate anything and the API runs supplied inputs. The safeguards are advisory instructions rather than enforced checks.

Medium

SKILL.md:107-117 SKILL.md:298-309

Biometric Media Sent to External Services

The workflow posts portrait, video, and audio references to RunComfy model endpoints. The file provides endpoint limits but no provider retention, deletion, or per-run privacy confirmation.

The invocation example supplies identity and audio URLs, and the implementation description says the CLI posts to the Model API. Facial media is inherently privacy-sensitive.

Capability review items (10)

These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.

Medium

SKILL.md:32-43

Ruby/shell backtick execution

```bash

The block tells the agent to install or execute @runcomfy/cli, authenticate, and run a remote model. Package execution and credentialed commands create supply-chain and data-transfer risk.

Medium

SKILL.md:49-51

Ruby/shell backtick execution

```bash

The block executes npx skills add against a community repository with global installation. This downloads and runs third-party package tooling, so the external-command finding is real.

Medium

SKILL.md:111-118

Ruby/shell backtick execution

```bash

The command invokes runcomfy and submits portrait and audio URLs to a hosted model endpoint. This intended behavior exposes sensitive media to an external process and service.

Medium

SKILL.md:137-144

Ruby/shell backtick execution

```bash

The command invokes runcomfy with a performance video and character image. It executes an external CLI and transfers potentially sensitive identity media to a remote service.

Medium

SKILL.md:170-181

Ruby/shell backtick execution

```bash

The command invokes runcomfy with scene and identity images. This is genuine external command execution that sends facial reference media to a hosted model.

Medium

SKILL.md:200-214

Ruby/shell backtick execution

```bash

The command invokes runcomfy with an identity reference and multiple scene images. The external process performs network-backed batch editing on user media.

Medium

SKILL.md:233-240

Ruby/shell backtick execution

```bash

The command invokes runcomfy with a user image and generated prompt. This is intended external command execution with remote processing of user media.

Medium

SKILL.md:298

Ruby/shell backtick execution

The skill classifies user intent — video vs still, motion-preserving vs identity-preserving, single

The text explicitly states that the skill invokes runcomfy, posts to the Model API, and downloads results. This confirms external command and network-backed processing behavior.

Medium

SKILL.md:303

Ruby/shell backtick execution

- **Install via verified package manager only.** Use `npm i -g @runcomfy/cli` or `npx -y @runcomfy/c

The line directs installation or execution of @runcomfy/cli through npm or npx. Running third-party package tooling creates a real supply-chain execution risk.

Medium

SKILL.md:311

Ruby/shell backtick execution

- **Scope of bash usage**: declared `allowed-tools: Bash(runcomfy *)`. The skill never instructs the

The line confirms Bash permission for any runcomfy subcommand. The scope is constrained, but the skill has genuine external command capability.

Risk Factors

⚙️ External commands (50)

🌐 Network access (35)

📁 Filesystem access (2)

SKILL.md:304 SKILL.md:304

Audited by: codex View Audit History →

Share & cite this report

Share the versioned assessment report, neutral badge, embed card, and citations. Skillstore reports evidence without deciding whether this Skill is safe.

Open versioned report

Copy report link

https://skillstore.io/skills/agentspace-so-face-swap/audits/5?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown badge

[![Skillstore security assessment](https://skillstore.io/badges/skills/agentspace-so-face-swap/security.svg)](https://skillstore.io/skills/agentspace-so-face-swap?utm_source=security_passport_badge)

HTML badge

<a href="https://skillstore.io/skills/agentspace-so-face-swap?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/agentspace-so-face-swap/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Embed card

<iframe src="https://skillstore.io/embed/skills/agentspace-so-face-swap.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Academic citations (APA · BibTeX · CFF)

APA citation

agentspace-so. (2026). face-swap security audit report (audit version 5) [Author version unspecified]. Skillstore. https://skillstore.io/skills/agentspace-so-face-swap/audits/5

BibTeX citation

@techreport{agentspace-so-agentspace-so-face-swap-2026,
  author = {agentspace-so},
  title = {face-swap security audit report (audit version 5)},
  institution = {Skillstore},
  year = {2026},
  number = {5},
  url = {https://skillstore.io/skills/agentspace-so-face-swap/audits/5},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "face-swap security audit report (audit version 5)"
version: "unspecified"
type: report
authors:
  - name: "agentspace-so"
date-released: "2026-07-23"
url: "https://skillstore.io/skills/agentspace-so-face-swap/audits/5"
identifiers:
  - type: other
    value: "skillstore:agentspace-so-face-swap:audit:5"
    description: "Skillstore immutable audit report identifier"

Compare variants

3 installable variants

Each author remains a separate installable skill. The recommended variant is ranked by Skillstore evidence.

Why this variant is first

Higher Skillstore usage

Author / source Skillstore Score Evidence Confidence Skillstore usage Updated

doany-ai Recommended

doany-ai-face-swap

Skillstore Score 38

Evidence Confidence High

Skillstore usage 47

Updated

2026-09-09

agentspace-so Current

agentspace-so-face-swap

Skillstore Score 38

Evidence Confidence High

Skillstore usage 40

Updated

2026-09-09

runcomfy-com

runcomfy-com-face-swap

Skillstore Score 38

Evidence Confidence Medium

Skillstore usage 46

Updated

2026-09-09

Skillstore Score

Why this score Evidence Confidence: High

Architecture

Maintainability

Content

Community

Spec Compliance

What You Can Build

Prepare campaign portraits

Brand designers can place a consenting spokesperson into approved scenes while preserving pose, clothing, lighting, and background.

Transfer character motion

Video editors can apply an approved performance to a target character while preserving the source movement.

Create consistent image sets

Creative developers can apply one approved identity across multiple product images or narrative frames.

Try These Prompts

Choose a route

I have [still image or video] and consent from every depicted person. Choose the best RunComfy route and explain the required inputs.

Edit one portrait

Use the approved face in [identity image] to replace the face in [scene image]. Preserve pose, clothing, hair, lighting, and background.

Process a consistent batch

Apply the approved identity in [reference image] across [scene images]. Keep aspect ratio and resolution consistent, and change no other scene elements.

Plan motion-preserving video

Use [performance video] and [approved character image] for motion transfer. Select the model, validate inputs, run the job, and summarize disclosure needs.

Best Practices

Confirm identity rights, source rights, and platform disclosure requirements before processing media.
Use clear, well-lit references and state exactly which visual details must remain unchanged.
Review every result for identity accuracy, unintended edits, privacy exposure, and required synthetic-media labels.

Avoid

Do not create impersonation, defamatory, sexual, fraudulent, or nonconsensual media involving real people.
Do not place secrets, local paths, or private media URLs into prompts or public asset links.
Do not assume a successful model response proves consent, factuality, or platform compliance.

Frequently Asked Questions

What do I need before using this skill?

You need the RunComfy CLI, authentication, network access, media rights, identity consent, and any required synthetic-media disclosure plan.

Which model handles a precise still-image face swap?

Use GPT Image 2 Edit for multiple references. Use Flux Kontext when one scene and a written face description are sufficient.

Can this skill process video?

Yes. Wan 2-2 Animate supports audio-driven character video, while Kling Motion Control preserves movement from a reference performance.

Can it apply one identity to several images?

Yes. Nano Banana Edit supports batch workflows with one identity reference and multiple scene images.

Does the skill enforce consent?

No. It provides refusal guidance, but users and agents must verify rights, consent, disclosure, and platform rules before every run.

Where is my media processed?

The RunComfy CLI sends request data to hosted model endpoints. Review provider privacy, retention, and deletion terms before submitting sensitive media.

Developer Details

Author

agentspace-so

License

MIT

Skillstore revision

Version notice

The author did not declare a version.

Repository

https://github.com/agentspace-so/runcomfy-agent-skills/tree/main/face-swap/

Ref

181fdefcafd96b041926e61c4b2e306ca7e7820e

Maintenance freshness

7/24/2026

Usage

3 downloads · 40 views

File structure

📄 SKILL.md