The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user inputs—specifically actor, reason, authority_ref, and reopen_gate_id—and interpolates them into a shell command line. While the instructions mandate 'normalized' inputs and the rejection of 'secret-shaped' values, the use of double-quoted variables in a shell context without explicit escaping requirements creates a surface where malicious user input (e.g., containing shell substitutions like $(...)) could potentially influence the execution environment.
Ingestion points: SKILL.md under the Intake section.
Boundary markers: The skill instructs the agent to require 'normalized inputs' but lacks explicit delimiters or escaping instructions for the shell command interpolation.
Capability inventory: Execution of a local Python script via a shell command in the Execute one repository transaction section of SKILL.md.
Sanitization: The skill mentions rejecting 'secret-shaped inputs' and 'aliases/symlinks' during resolution, but does not provide specific instructions for sanitizing user-provided strings against shell metacharacters before command execution.
[COMMAND_EXECUTION]: The skill explicitly instructs the agent to execute a Python script (ticket-autopilot.py) with multiple arguments. This is the primary mechanism for the skill's intended functionality but involves significant interaction with the host system's shell and file system.

change-status-ticket