The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands for telemetry and notification using curl to localhost:31337 and for logging using echo to a local file (~/.claude/PAI/MEMORY/SKILLS/execution.jsonl) upon every invocation.
[DYNAMIC_EXECUTION]: The skill instructs the agent to check for and load configuration overrides from the directory ~/.claude/PAI/USER/SKILLCUSTOMIZATIONS/FirstPrinciples/. Loading and applying instructions from a user-controlled path can lead to unauthorized behavior modification.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes user-provided inputs and interpolates them into a shell echo command without explicit sanitization or escaping instructions.
Ingestion points: User-supplied problem descriptions and subject matter deconstructed in the Deconstruct workflow.
Boundary markers: None present in the instructions to protect against malicious payloads in user input.
Capability inventory: The skill performs shell execution (curl, echo) and persistent file writing.
Sanitization: No sanitization or escaping requirements are specified for the 8_WORD_SUMMARY placeholder before it is executed in the shell.

FirstPrinciples