Artemisa generates configuration files; it does not write them to your project. This guide explains how to take a generated bundle and apply it manually, verify its integrity, and handle custom options.
Every bundle includes these core files:
| File | Purpose |
|---|---|
artemisa.blueprint.json |
Canonical model of all your decisions |
manifest.json |
File inventory with SHA-256 hashes |
docs/INSTALL.md |
Step-by-step install and validation guide |
docs/WHY.md |
Explanation of objective, stack, environment and recommendations |
Depending on your answers, additional artifacts are generated:
| Condition | Artifacts |
|---|---|
| Dev, Kiro or portable target | AGENTS.md |
| Skills enabled | skills/<agent>/SKILL.md |
| Artemisa target | artemisa/steering.json, security-policy.json, governance.json, mcps.json |
| Artemisa + RAG | artemisa/rag.json |
| Artemisa + PR review | artemisa/pr-review.json |
| Kiro target | .kiro/steering/<agent>.md |
| Kiro + hooks | .kiro/hooks/<agent>-quality.json |
| Kiro + skills | .kiro/skills/<agent>/SKILL.md |
From the Creator's completion screen:
- "Descargar todo (.zip)" — downloads a ZIP with all artifacts preserving their relative paths, plus
manifest.jsonandartemisa.blueprint.jsonat the root. - "Descargar bundle (JSON)" — downloads the full bundle as a single JSON object (you'll need to unpack it manually).
- Individual download — each artifact can be downloaded separately from the "Archivos" tab.
The ZIP is the recommended option for applying the full bundle.
Before copying anything to your project:
- Open
docs/WHY.md— understand why each recommendation was made. - Open
artemisa/security-policy.json(if generated) — review the allowlist of allowed commands and paths. - Open
artemisa/steering.json(if generated) — review the roles, system prompts and tool permissions. - Check the warnings shown in the Creator — they flag potential issues (e.g., a local database in production, deploy privileges).
Never copy files without reading them first. These files define what your agent can do. A misconfigured steering or security policy can grant excessive permissions.
Unzip the bundle and copy the files to your project root, preserving the relative paths:
cd /path/to/your-project
unzip /path/to/artemisa-agent.zipThe directory structure should look like:
your-project/
├── AGENTS.md
├── docs/
│ ├── INSTALL.md
│ └── WHY.md
├── artemisa/
│ ├── steering.json
│ ├── security-policy.json
│ ├── governance.json
│ └── mcps.json
├── .kiro/ # only if Kiro target
│ ├── steering/
│ ├── hooks/
│ └── skills/
├── skills/ # only if skills enabled
├── artemisa.blueprint.json
└── manifest.json
If you already have an
AGENTS.mdordocs/directory, merge carefully — don't overwrite existing content without reviewing it.
The manifest.json contains a SHA-256 hash for every generated file. After copying, verify that the files match:
cd /path/to/your-project
# Example: verify steering.json
sha256sum artemisa/steering.json
# Compare the output with the hash in manifest.jsonTo verify all files at once:
# Using jq and sha256sum
jq -r '.files[] | "\(.sha256) \(.path)"' manifest.json | sha256sum -c -If any file doesn't match, it was modified during copy — re-download and copy again.
docs/INSTALL.md contains the specific installation steps for your agent configuration. It includes:
- Dependencies to install
- Environment variables to set
- Validation commands to run
- Production checklist (if production environment was selected)
Follow it step by step.
If you used custom:<slug> for any technology (e.g., custom:my-framework), the bundle will include:
- The custom value in
artemisa.blueprint.json - A warning: "adaptador pendiente" (pending adapter)
- Documentation in
WHY.mdexplaining the custom decision
What to do:
- The generated artifacts won't include specific configuration for your custom technology.
- You'll need to write the configuration manually, following the patterns of similar technologies in the bundle.
- Use the
security-policy.jsonallowlist as a template — add your custom tool's commands to the allowlist.
| Target | Where | Use when |
|---|---|---|
AGENTS.md |
Project root | Universal — works with any AI agent that reads AGENTS.md (Cursor, Claude Code, Windsurf, etc.) |
artemisa/ |
artemisa/ directory |
When applying the generated Artemisa steering format (steering, security policy, governance, MCPs) |
.kiro/ |
.kiro/ directory |
When using Kiro (AWS's agentic IDE) — steering, hooks and skills in Kiro's format |
You can have multiple targets in the same bundle (e.g., both AGENTS.md and .kiro/). They're complementary, not mutually exclusive.
| Mistake | Fix |
|---|---|
| Copying files to the wrong directory | Always copy to the project root, preserving relative paths |
Overwriting an existing AGENTS.md |
Merge manually — keep your existing project-specific rules |
| Not verifying hashes | Run sha256sum -c against manifest.json after copying |
| Ignoring warnings | Review all warnings in the Creator before applying |
Copying security-policy.json without review |
This file defines what your agent is allowed to do — review every entry |