The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local project files to detect callable surfaces and backend patterns. Malicious instructions or code comments in the analyzed project could potentially influence the agent's plan or the generated CLI configuration.
Ingestion points: scripts/inspect_project.py reads project manifests and source files to extract metadata.
Boundary markers: Lacks explicit instructions for the agent to ignore or delimit instructions found within the processed project files.
Capability inventory: The skill has the capability to write files via scripts/scaffold_cli.py and execute local commands via scripts/validate_cli.py.
Sanitization: Employs regex-based validation for project and package names and enforces shell=False in the generated bridge code.
[COMMAND_EXECUTION]: The skill uses subprocess calls to validate the generated CLI and run project-specific tests.
Evidence: scripts/validate_cli.py and scripts/test_cli_creator.py invoke subprocess.run to execute the Python interpreter and the generated CLI commands.
Context: This execution is part of the intended validation workflow and uses argument arrays with shell=False to mitigate injection risks.
[DYNAMIC_EXECUTION]: The skill generates an executable Python harness from templates and then executes it locally during the validation phase.
Evidence: scripts/scaffold_cli.py creates a package including backend.py and cli.py using templates; Step 6 of the workflow in SKILL.md triggers the execution of this generated code via scripts/validate_cli.py.
Context: The generation is template-driven, and identifiers are validated against strict regex patterns to prevent code injection into the harness itself.
[PERSISTENCE]: The skill maintains persistent user preferences and durable records using a profile storage utility.
Evidence: scripts/profile_store.py performs atomic writes to a local JSON profile (e.g., in ~/.config/agent-skills/profile.json).
Mitigation: The script blacklists sensitive field names like 'secret' and 'token' to prevent accidental exposure and requires an explicit --confirm flag from the agent to save changes.

lov-cli-creator