The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requests a SESSION_ID as an input parameter. While this is a placeholder, passing sensitive session tokens through instruction arguments is a practice that can lead to credential exposure in logs or session histories.
[DYNAMIC_EXECUTION]: The skill utilizes a "live JIT check" against a "real system." Just-In-Time execution represents a dynamic code execution surface where code is compiled and executed at runtime to evaluate system state.
[PROMPT_INJECTION]: The "Finish the work" section contains directives to "repair the fault where it originates" and "use an alternate route" if a step is blocked. Such instructions could be interpreted by the agent as a mandate to bypass security controls, error messages, or safety guardrails if they are perceived as "faults" preventing the requested work from being completed.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external system checks ("observed value") without specifying boundary markers or sanitization steps. This creates a surface where external data could potentially influence the agent's behavior if the system being checked returns malicious content.
Ingestion points: Ingests output from the "live JIT check" into the agent's context (SKILL.md).
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used for external data.
Capability inventory: Performs dynamic code execution (JIT) and writes data to the filesystem at .gm/receipts/ (SKILL.md).
Sanitization: The instructions do not mention any escaping, filtering, or validation of the observed system values.

gm-eliminate