Agent Visio Use

Bootstrap skill for DrawForge. Use this skill to onboard an agent into the DrawForge GitHub repository, understand the project structure, run the canonical c...

Security Scan

The skill is an instruction-only bootstrap for the DrawForge repo; its declared requirements (git, python, VISIO_BRIDGE_TOKEN) match the workflow it describes, but the runtime instructions tell an agent to clone and execute Python scripts from a remote repository so you should audit the repo and be careful with any bridge token before running tests.

Purpose & Capability

Name/description claim to onboard an agent into the DrawForge repo and run a smoke test. Required binaries (git, python) and a VISIO_BRIDGE_TOKEN for the Visio bridge are consistent with cloning and running the repository's Python smoke-test workflow.

Instruction Scope

SKILL.md instructs the agent to clone the GitHub repo and run three Python scripts from Setup/ (prepare_smoke_test.py, run_draw_job.py, execute_drawdsl.py). That is coherent with the stated purpose, but it does involve executing code pulled from a remote repository—users should audit the repository and the referenced scripts before running them, and be cautious about any network activity those scripts may perform or any sensitive inputs they request.

Install Mechanism

There is no install spec; the skill is instruction-only. This is the lowest-risk model for an onboarding helper and is consistent with the described purpose.

Credentials

Only one environment variable is declared (VISIO_BRIDGE_TOKEN). The QUICKSTART clarifies this is a local token for a user's Visio bridge and is only needed for the bridge-backed smoke test; that is proportionate to the skill's purpose. No unrelated credentials or config paths are requested.

Persistence & Privilege

always is false and there is no install-time persistence requested. The skill does not request elevated platform presence or modify other skills' configuration.

Assessment

This skill is coherent for onboarding into DrawForge, but it directs the agent to clone a GitHub repo and run Python scripts from that repo. Before running the smoke test, verify the upstream GitHub repository and inspect the referenced Setup/*.py scripts for any network calls, credential use, or unexpected behavior. Only provide VISIO_BRIDGE_TOKEN if you trust and understand the token's scope; prefer an ephemeral or minimally-scoped token. Note the SKILL.md example uses an SSH clone ([e-mail]...), which will use your SSH keys — if you prefer, use the HTTPS clone URL. Run the smoke test in an isolated or sandboxed environment if you are unsure about the code, and avoid supplying unrelated secrets. Finally, remember this is instruction-only: the agent will execute commands you allow, so audit the repository before granting runtime access.

DrawForge Bootstrap

This is a lightweight onboarding skill for the DrawForge repository.

It is not the full DrawForge system. Its job is to guide an agent to the correct GitHub repository, documents, smoke test, and execution flow.

DrawForge itself is an agent-driven closed loop built on top of Microsoft Visio. Its goal is to turn reference figures into directly editable diagram assets by helping agents operate Visio more like a capable human user rather than as a blind API caller.

What this skill can do

This skill can help an agent:

• understand what DrawForge is trying to achieve
• find the correct GitHub repository and entry documents
• avoid random first-run behavior and jump into the intended workflow
• run the canonical cold-start smoke test
• start reproducing reference figures through the DrawForge Visio loop
• move toward a result that a human can continue editing directly in Visio

Typical outcomes

After using this skill, an agent should be able to:

• explain the DrawForge workflow clearly
• bootstrap itself into the repository with the correct read order
• validate that the Visio bridge and execution path are working
• begin work on figure reproduction with better layer awareness
• help produce editable .vsdx outputs instead of dead image copies

What this skill is for

Use this skill when an agent needs to:

• find the DrawForge source repository
• understand the top-level architecture quickly
• avoid free-form blind retries
• run the canonical cold-start smoke test
• begin work in the correct layer

When to use it

Use this skill when:

• an agent is entering DrawForge for the first time
• a new environment needs to be validated before real drawing work
• a user wants an agent to help reproduce a figure through Visio
• the goal is not only to look similar to the reference, but to obtain a directly editable diagram asset

What this skill is not

This skill does not bundle the whole repository. It does not include Visio bridge code, benchmark PNGs, or runtime artifacts.

The full project lives in the GitHub repository:

https://github.com/qweadzchn/DrawForge

Recommended workflow

1. Clone the GitHub repository locally.
2. Read the cold-start entry documents.
3. Run the canonical smoke test before doing open-ended drawing work.
4. Only then move on to real jobs or system improvements.

Clone the repository

git clone [e-mail]:qweadzchn/DrawForge.git
cd DrawForge

If SSH is not available, use HTTPS instead.

Read order

Read these files first:

1. AGENT_START_HERE.md
2. AGENT_GUIDE.md
3. GET_STARTED.md
4. docs/human/setup/AGENT_COLD_START_SMOKE_TEST.md
5. MODE_POLICY.md

Canonical smoke test

From the repo root:

python Setup\prepare_smoke_test.py --config Setup\examples\smoke-test-inputpng-1.json
python Setup\run_draw_job.py --config Setup\examples\smoke-test-inputpng-1.json
python Setup\execute_drawdsl.py --config Setup\examples\smoke-test-inputpng-1.json --round 1 --save-final

Expected outputs:

• OutputPreview/smoke-inputpng-1/round-01.png
• OutputEditable/1_smoke_test_final.vsdx

Routing rule

When working inside DrawForge:

• if the issue is round-specific, keep it in review artifacts
• if it looks structural but still needs validation, write a proposal
• if it is already reusable experience, promote it into a lesson
• if the shared fix is clear, patch the owning layer directly

Where to go next

See:

• README.md
• CONTRIBUTING.md
• docs/architecture/FEEDBACK_PROMOTION_LOOP.md