The Agent Skills Directory

[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection due to the way it handles input data.
Ingestion points: User-provided data is ingested via the $ARGUMENTS variable in the INPUT section of SKILL.md.
Boundary markers: The skill does not use delimiters (like triple quotes or XML tags) or specific instructions to ignore embedded commands within the $ARGUMENTS payload.
Capability inventory: The skill performs several high-capability actions, including executing project build and test commands (under SELF-HEALING VALIDATION) and writing telemetry logs to the user's home directory (SELF-EVOLUTION TELEMETRY).
Sanitization: No input validation or escaping mechanisms are described to prevent a malicious input from overriding the agent's instructions or exploiting its command execution capabilities.
[COMMAND_EXECUTION]: The 'SELF-HEALING VALIDATION' phase instructs the agent to execute build and test commands (e.g., npm test or make) found within the project's own configuration. This grants the agent the ability to execute arbitrary scripts defined in the target codebase, which is an intended functionality but presents a risk when used on untrusted repositories.
[COMMAND_EXECUTION]: The 'SELF-EVOLUTION TELEMETRY' section involves automated filesystem operations, specifically searching for and appending data to ~/.claude/projects/skill-telemetry.md. This demonstrates localized persistence and logging capabilities within the platform's configuration directory.

logistics-optimize