The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script located at scripts/init.sh to perform various initialization tasks, including directory creation (mkdir), file detection (find), and managing symbolic links (ln -s). It also interacts with other tools like skill-manager to handle the installation of additional skills within the project environment.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection as its primary function is to transform project data into agent instructions.
Ingestion points: The skill reads project-specific metadata files (e.g., package.json, pyproject.toml, Cargo.toml), documentation (README.md), and project directory structures as documented in SKILL.md (Step 4).
Boundary markers: While the skill utilizes structured templates (e.g., templates/agents-template.md) to guide the generation of new instruction files, it does not implement explicit boundary markers or warnings that instruct the agent to ignore potentially malicious instructions embedded within the analyzed project data.
Capability inventory: The skill possesses the ability to read from and write to the file system, create directory structures, and execute shell commands through scripts/init.sh and other referenced tools.
Sanitization: The skill does not perform sanitization or validation of the content extracted from external project files before interpolating it into the final AGENTS.md and CLAUDE.md files, which serve as the agent's core instructions.
[SAFE]: The detection of potentially destructive system commands (e.g., rm -rf /, dd, git reset --hard) in assets/claude-settings-template.json and assets/codex-default.rules is benign. These strings are explicitly defined within 'deny' lists and 'forbidden' rules designed to limit the agent's permissions and prevent accidental or malicious execution of harmful commands. This demonstrates a security-conscious design for the target project's configuration.

project-init