The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input data, such as task descriptions, output samples, and rubrics, which are passed as arguments to the autoctx CLI. This creates an attack surface for indirect prompt injection where malicious instructions in the user-provided data could influence the agent's behavior or the results of the improvement loop.
Ingestion points: autoctx solve, autoctx judge, and autoctx improve command arguments in SKILL.md.
Boundary markers: None present in the instructions.
Capability inventory: Executes system commands via the autoctx CLI across multiple instructions in SKILL.md.
Sanitization: None specified in the skill documentation.
[REMOTE_CODE_EXECUTION]: The skill's functionality is dependent on the external autoctx command-line utility. It instructs the agent to execute this binary on the host system to perform knowledge creation and scoring tasks.

autocontext-creator